Acoustic Emission Signatures of Fatigue Damage in Idealized Bevel ...

metals Article

Acoustic Emission Signatures of Fatigue Damage in Idealized Bevel Gear Spline for Localized Sensing Lu Zhang 1 , Didem Ozevin 1, *, William Hardman 2 and Alan Timmons 2 1 2

*

Civil and Materials Engineering, University of Illinois at Chicago, Chicago, IL 60607, USA; [email protected] Naval Air Systems Command (NAVAIR), Patuxent River, MD 20670, USA; [email protected] (W.H.); [email protected] (A.T.) Correspondence: [email protected]; Tel.: +1-312-413-3051

Received: 30 May 2017; Accepted: 26 June 2017; Published: 30 June 2017

Abstract: In many rotating machinery applications, such as helicopters, the splines of an externally-splined steel shaft that emerges from the gearbox engage with the reverse geometry of an internally splined driven shaft for the delivery of power. The splined section of the shaft is a critical and non-redundant element which is prone to cracking due to complex loading conditions. Thus, early detection of flaws is required to prevent catastrophic failures. The acoustic emission (AE) method is a direct way of detecting such active flaws, but its application to detect flaws in a splined shaft in a gearbox is difficult due to the interference of background noise and uncertainty about the effects of the wave propagation path on the received AE signature. Here, to model how AE may detect fault propagation in a hollow cylindrical splined shaft, the splined section is essentially unrolled into a metal plate of the same thickness as the cylinder wall. Spline ridges are cut into this plate, a through-notch is cut perpendicular to the spline to model fatigue crack initiation, and tensile cyclic loading is applied parallel to the spline to propagate the crack. In this paper, the new piezoelectric sensor array is introduced with the purpose of placing them within the gearbox to minimize the wave propagation path. The fatigue crack growth of a notched and flattened gearbox spline component is monitored using a new piezoelectric sensor array and conventional sensors in a laboratory environment with the purpose of developing source models and testing the new sensor performance. The AE data is continuously collected together with strain gauges strategically positioned on the structure. A significant amount of continuous emission due to the plastic deformation accompanied with the crack growth is observed. The frequency spectra of continuous emissions and burst emissions are compared to understand the differences of plastic deformation and sudden crack jump. The correlation of the cumulative AE events at the notch tip and the strain data is used to predict crack growth. The performance of the new sensor array is compared with the conventional AE sensors in terms of signal to noise ratio and the ability to detect fatigue cracking. Keywords: fatigue testing; acoustic emission; spline; gearbox; sensor array

1. Introduction The function of the gearbox in a helicopter transmission system is to keep the engine torque stable at optimum levels and transmit the power. As the spline section of the gearbox is subjected to externally-cyclic loading (e.g., torque, tensile and bend), it is susceptible to develop fatigue cracks and spalling [1–3]. Being a non-redundant component, it is important to monitor the structural state of the spline section using real-time and non-intrusive nondestructive evaluation (NDE) methods. There are several NDE methods implemented to monitor gearbox components, including vibration, acoustic emission, and debris monitoring. For instance, Hussain and Gabbar [4] developed a real-time Metals 2017, 7, 242; doi:10.3390/met7070242

www.mdpi.com/journal/metals

Metals 2017, 7, 242

2 of 21

coded genetic algorithm to detect gear faults, and optimized the process of vibration response feature extraction by band-pass and wavelet filters. Lin et al. [5] studied the signature of spectrograms of vibration signals caused by the crack evolution in gear in terms of B-spline wavelet method. Liu et al. [6] diagnosed the faults in the gearbox according to the vibration signal by means of an empirical mode decomposition and Hilbert spectrum. Bonnardot et al. [7] tracked the rotation behavior of a gearbox based on resampling the acceleration signals. Leturiondo et al. [8] evaluated the condition of the gearbox based on the physics-based model. In their model, not only the dynamics of the rolling element bearings have been considered, but also the influence of the electrical process is included. The change in the gear-mesh stiffness obtained from the vibration signals is typically correlated with the presence of cracking [9–11]. Though the vibration method is a well-established method for monitoring rotating machinery, the sensitivity is insufficient for detecting the initiation of a fatigue crack. Additionally, the operational conditions (e.g., background noises and rotational speed) may influence the baseline vibration properties [12]. Acoustic emission (AE) is a passive NDE method for detecting the active flaws in the gearbox components. The AE method is based on detecting elastic waves released by sudden localized failures (e.g., crack growth, corrosion) and applied in damage detection in various metallic components. Typically, the AE sensors are mounted on the structural surface, which requires elastic waves propagating through several structural layers before reaching the sensor from the crack location in complex structures [13]. While the AE characteristics of the prescribed minor and major defects on the tooth are identified by several researchers [14,15], the applicability of their test results in a realistic condition is questionable as the AE waveform depends on the source function, the structural geometry, and the sensors used. It is important to place AE sensors in proximity to the source in order to reduce the influence of the structural geometry on AE waveforms [16]. The AE and vibration methods have been combined in the literature in order to increase the damage detection ability (e.g., [17–22]). Although there are many laboratory-scale studies aiming to the spline section of gearbox, the application of the method in a realistic field test is challenged by the complexity of AE signal, and the influence of background noise on the dataset. However, the representative AE features obtained in a properly designed laboratory scale experiment can help us to understand the characteristics of AE signal generated by the fatigue testing for localized sensing [23,24]. The majority of the studies studied simple plate geometry [25]; however, the complexity in structural geometry would affect the AE signature. The major goals of this study are (1) to identify the AE signatures due to fatigue crack growth released in the complex geometry of a spline section; and (2) to test a new piezoelectric sensor array with the ability to place the sensor array inside the gearbox to reduce the wave propagation path. To achieve the research goals, the geometrical details of spline section are built in a flattened geometry. The sample is loaded under constant cycle fatigue until the crack length reaches about 7 mm. The AE signatures related to two major source mechanisms as plastic deformation and sudden crack jump are identified. The performance of the piezoelectric sensor array is evaluated in terms of detecting the crack growth in comparison with conventional AE sensors. The outline of this paper is as follows: A brief introduction to the AE method relevant to this study is presented in Section 2. The details of fatigue testing and AE setting are described in Section 3. The numerical model to predict the crack growth behavior is described in Section 4. Section 5 includes the experimental results, and the conclusions of this study are presented in Section 6. 2. Acoustic Emission Method Acoustic emission (AE) is the phenomenon of radiation of transient elastic waves in a material when the sudden redistribution of the stress occurs in the material [26]. As the AE sensors (typically piezoelectric types) are triggered by the transient elastic waves released from sudden localized failure, the method is called a ‘passive’ NDE method. Figure 1 illustrates the measurement chain of an AE signal. In essence, the quantitative AE method requires understanding and modeling all components in Figure 1. The AE signal is the convolution of AE source mechanism, medium transfer function,

Metals 2017, 7, 242 Metals 2017, 7, 242

3 of 21 3 of 21

function, and sensorfunction. transfer function. data acquisition systemalso should also be considered if it and sensor transfer The dataThe acquisition system should be considered if it applies applies analog/digital analog/digital filters orfilters gain.or gain.

Figure 1. The measurement chain of AE signals. Figure 1. The measurement chain of AE signals.

The AE source mechanism is controlled by materials (e.g., metals, composites), loading types The AE source mechanism is controlled by materials (e.g., metals, composites), loading types (e.g., monotonic or fatigue), loading rates (e.g., fast or slow), and the previous load history of the (e.g., monotonic or fatigue), loading rates (e.g., fast or slow), and the previous load history of the material. Currently, a number of studies have attempted to identify the source mechanisms for material. Currently, a number of studies have attempted to identify the source mechanisms for different different materials and loading conditions. Theoretical studies are inspired by the application of materials and loading conditions. Theoretical studies are inspired by the application of seismology. The seismology. The body force equivalents representing dislocations were derived theoretically [27], body force equivalents representing dislocations were derived theoretically [27], and the crack kinetics and the crack kinetics and kinematics were studied by means of the moment tensor analysis [28–30]. and kinematics were studied by means of the moment tensor analysis [28–30]. However, the derived However, the derived source is the representative acoustic emission; the difference between AE source is the representative acoustic emission; the difference between AE waves due to generalized waves due to generalized theory and those of cracking has not been clearly realized. The source theory and those of cracking has not been clearly realized. The source mechanism is significantly mechanism is significantly influenced by the material composition [31–36]. For instance, sources influenced by the material composition [31–36]. For instance, sources associated with composites are associated with composites are matrix cracking, fiber-matrix debonding, fiber pull-out, fiber matrix cracking, fiber-matrix debonding, fiber pull-out, fiber bridging, inter-ply failure, delamination, bridging, inter-ply failure, delamination, and fiber breakage. For brittle materials, such as concrete and fiber breakage. For brittle materials, such as concrete or rock, microcrack and macrocrack initiation or rock, microcrack and macrocrack initiation and growth are two major AE sources. For metallic and growth are two major AE sources. For metallic materials, the source mechanisms are usually materials, the source mechanisms are usually related to the microstructural changes such as related to the microstructural changes such as movement of dislocation, slip, twinning, fracture, and movement of dislocation, slip, twinning, fracture, and debonding of precipitates, in addition to debonding of precipitates, in addition to macrostuctural changes, such as crack jump. macrostuctural changes, such as crack jump. The common sources that generate AE signals in the spline section of gearbox include plastic The common sources that generate AE signals in the spline section of gearbox include plastic deformation, microfracture, wear, bubbles, friction, and impact [37]. Different source mechanisms deformation, microfracture, wear, bubbles, friction, and impact [37]. Different source mechanisms release different energy levels, which affect their detectability by the AE method. If the signal level release different energy levels, which affect their detectability by the AE method. If the signal level is is below electronic or background noise level, the AE method cannot detect the flaw. Poddar and below electronic or background noise level, the AE method cannot detect the flaw. Poddar and Giurgiutiu [38] applied the numerical method to understand the detectable length of fatigue cracking Giurgiutiu [38] applied the numerical method to understand the detectable length of fatigue by means of AE. Therefore, it is important to identify detectable source mechanisms and adapt them in cracking by means of AE. Therefore, it is important to identify detectable source mechanisms and a realistic test environment. In this study, the structural material is made of steel, and the target source adapt them in a realistic test environment. In this study, the structural material is made of steel, and mechanisms are plastic deformation and sudden crack jump. the target source mechanisms are plastic deformation and sudden crack jump. Once the AE signals are detected, there are different signal processing techniques to extract Once the AE signals are detected, there are different signal processing techniques to extract features describing source mechanisms [20,39–41]. For instance, Siracusano et al. [42] conducted the features describing source mechanisms [20,39–41]. For instance, Siracusano et al. [42] conducted the Hilbert-Huang transform on the AE signals to evaluate the damage evolution in the cementitious Hilbert-Huang transform on the AE signals to evaluate the damage evolution in the cementitious material, and the method has been applied successfully in plain and reinforced concrete, and material, and the method has been applied successfully in plain and reinforced concrete, and the AE the AE behavior of nucleation of micro-cracks and the formation of macro-cracks are identified. behavior of nucleation of micro-cracks and the formation of macro-cracks are identified. Hamzah et Hamzah et al. [43] found the correlation between the lubricating condition of the gearbox and RMS al. [43] found the correlation between the lubricating condition of the gearbox and RMS (root mean (root mean square) level. While the relationships between damage states of the gearbox and AE square) level. While the relationships between damage states of the gearbox and AE features have features have been established by several researchers [44–47], the results are based on laboratory-scale been established by several researchers [44–47], the results are based on laboratory-scale experiments with artificially introduced damage. Additionally, almost all of the AE features are related experiments with artificially introduced damage. Additionally, almost all of the AE features are to the burst types of signals (as shown in Figure 1), and the continuous signals are typically considered related to the burst types of signals (as shown in Figure 1), and the continuous signals are typically as noise sources in detecting fatigue crack growth. considered as noise sources in detecting fatigue crack growth.

Metals 2017, 7, 242 Metals 2017, 7, 242

4 of 21 4 of 21

3. Overview of the Idealized Gearbox Sample

spline section sectionof ofthe thehelicopter helicoptergearbox gearboxhas hascomplex complexgeometry geometry with variable cross-sections, The spline with variable cross-sections, as as shown in Figure structural complexity affects the wave and propagation the shown in Figure 2a. 2a. TheThe structural complexity affects the wave field field and propagation whenwhen the crack crack occurs at a tooth. spline Therefore, tooth. Therefore, it is important test a geometry similar geometry underloading fatigue occurs at a spline it is important to test atosimilar under fatigue loading for understanding the characteristics of AE signals due to fatigue However, cracking. it However, it to is for understanding the characteristics of AE signals due to fatigue cracking. is difficult difficult to apply fatigue on a cylinder-shaped spline in the laboratory environment. In this apply fatigue loading on loading a cylinder-shaped spline in the laboratory environment. In this study, the study, the geometry cylindricalis flattened geometrywhile is flattened while the details cross-sectional details and splines cylindrical the cross-sectional and splines are preserved (seeare in preserved Figure 2b). The idealization it possible conduct a fatigue test and Figure 2b). (see The in idealization makes it possible tomakes conduct a fatigue to test and initiate and grow the initiate fatigue and grow fatigue crack. The cross-sectional ofsample the idealized gearbox sample is shown in crack. The the cross-sectional profile of the idealizedprofile gearbox is shown in Figure 2c. The window Figure 2c. The window sectiontoisplace specifically to place thearray, new which piezoelectric section is specifically designed the newdesigned piezoelectric sensor would sensor have a array, direct which wouldthe have a direct path between the crack and sensing element, inside similarthetogearbox. its planned path between crack and sensing element, similar to its planned installment The installment inside the gearbox. The idealized specimen is manufactured using a structural steel idealized specimen is manufactured using a structural steel plate. A notch with a length of 7.6 cm was plate. Aatnotch with section a length 7.6 cm was the formed the spline section of inthe order to grow the crack formed the spline inof order to grow crackatwithin the capacity loading machine. The within theis capacity of by thethe loading machine. The AE signaltoisthe influenced by theThe crack growth AE signal influenced crack growth direction relative sensor position. laboratory direction relative to the sensor The laboratory condition that crack growth is perpendicular condition that crack growth is position. perpendicular to the piezoelectric sensor array simulates the actual to thecondition. piezoelectric sensor array simulates the actual field condition. field

Figure 2. The flattened spline section and dimensions (unit: mm).

In order to show the influence of geometric details (e.g., cross-section and spline), two finite In order to show the influence of geometric details (e.g., cross-section and spline), two finite elements models using COMSOL Multiphysics software (Version 4.2a, COMSOL Inc., Stockholm, elements models using COMSOL Multiphysics software (Version 4.2a, COMSOL Inc., Stockholm, Sweden) are built, and the frequency analysis is conducted to illustrate the influence of geometric Sweden) are built, and the frequency analysis is conducted to illustrate the influence of geometric details on the elastic wave propagation. A tetrahedral mesh is implemented with the minimum details on the elastic wave propagation. A tetrahedral mesh is implemented with the minimum mesh size of 0.25 mm to order to achieve 1 MHz resolution. The numerical model is simulated in the mesh size of 0.25 mm to order to achieve 1 MHz resolution. The numerical model is simulated in frequency domain by applying harmonic perturbation at the location of the notch. The natural the frequency domain by applying harmonic perturbation at the location of the notch. The natural boundary condition is defined. The material properties of steel are defined as the Young’s modulus boundary condition is defined. The material properties of steel are defined as the Young’s modulus (200 GPa), density (7850 kg/m3), and Poisson’s ratio (0.33). One model preserves all the spline details (200 GPa), density (7850 kg/m3 ), and Poisson’s ratio (0.33). One model preserves all the spline details in a typical helicopter gearbox, and the other model has a uniform surface without spline in a typical helicopter gearbox, and the other model has a uniform surface without spline components, components, as shown in Figure 3. The constant amplitude dipole force is introduced as an artificial as shown in Figure 3. The constant amplitude dipole force is introduced as an artificial AE source, AE source, and a sweep of the frequency range from 20 kHz to 400 kHz, with a frequency step of 1 and a sweep of the frequency range from 20 kHz to 400 kHz, with a frequency step of 1 kHz, is kHz, is applied to find the structural response in the frequency domain. The comparison of the applied to find the structural response in the frequency domain. The comparison of the displacement displacement responses at the selected point (position of the AE sensor highlighted as the wideband responses at the selected point (position of the AE sensor highlighted as the wideband (WD) sensor (WD) sensor on the figure) is shown in Figure 4. The geometric details cause the different on the figure) is shown in Figure 4. The geometric details cause the different characteristics of the AE characteristics of the AE signal, especially in the high-frequency range (150 kHz to 300 kHz). The signal, especially in the high-frequency range (150 kHz to 300 kHz). The difference is more apparent difference is more apparent for in-plane responses (x and y directions) because the dipole force is for in-plane responses (x and y directions) because the dipole force is applied on the surface, which applied on the surface, which coincides with the in-plane direction. The numerical results indicate coincides with the in-plane direction. The numerical results indicate that the test structure should that the test structure should preserve as many geometrical details as possible to obtain acoustic signatures representing a given damage mode and structural component.

Metals 2017, 7, 242

5 of 21

preserve as many geometrical details as possible to obtain acoustic signatures representing a given Metals 7,7,242 55of Metals2017, 2017, 242 of21 21 damage mode and structural component.

11

11

0.8 0.8

0.6 0.6

0.6 0.6

100 200 300 100 200 300 Frequency(kHz) Frequency(kHz)

(a) (a)

400 400

00 00

Complex Complexmodel model Simple Simplemodel model

0.8 0.8

0.4 0.4

0.2 0.2

0.2 0.2

11

0.6 0.6

0.4 0.4

0.4 0.4

00 00


0.8 0.8

Norm.IP.X-displacement Norm.IP.X-displacement


Norm.IP.Y-displacement Norm.IP.Y-displacement

Norm.OOP.Z-displacement Norm.OOP.Z-displacement

Figure 3. 3.3. The schematics to study the influence influenceof ofgeometric geometricdetails details AE Figure The schematics of numerical models on AE Figure The schematicsof ofnumerical numericalmodels models to to study study the influence of geometric details onon AE signatures; (a)(a) excitation and sensor; (c) complex model(unit: (unit:mm). mm). signatures; excitation and sensor; (b) simple model; signatures; (a) excitation and sensor;(b) (b)simple simplemodel; model;and and(c) (c)complex complexmodel model (unit: mm).

0.2 0.2


(b) (b)

400 400

00 00


400 400

(c) (c)

Figure Figure 4.4. The The comparison comparison of of displacement displacement responses responses in in frequency frequency domain, domain, (a) (a) zz direction; direction; (b) (b) yy Figure 4. The comparison of displacement responses in frequency domain, (a) z direction; (b) y direction; direction; direction;and and(c) (c)xxdirection. direction. and (c) x direction.

4.4.Experimental ExperimentalDesign Design 4. Experimental Design The Theexperimental experimentalsetup setupincludes includesthree threemonitoring monitoringtechnologies: technologies:AE, AE,strain, strain,and andultrasonics. ultrasonics.The The The experimental setup includes three monitoring AE, strain, and ultrasonics. AE data isis recorded using six sensors: three (wideband sensor AE data recorded using six piezoelectric piezoelectric sensors:technologies: three WD WD sensors sensors (wideband sensor The AE data is by recorded using six sensors: threeplaced WD in sensors (wideband sensor manufactured Mistras Princeton NJ, form for manufactured by MistrasGroup GroupInc., Inc.,piezoelectric PrincetonJunction, Junction, NJ,USA) USA) placed inaatriangulation triangulation form for manufactured by Mistras Group Inc., Princeton Junction, NJ, USA) placed in by abytriangulation form source one sensor (nano-30 FP manufactured Mistras Inc., sourcelocalization; localization; oneNano Nano30 30 sensor (nano-30 FP97 97sensor sensor manufactured MistrasGroup Group Inc.,for source localization; one 30 sensor (nano-30 FP 97 sensorand manufactured by Mistras Group as Inc., Princeton Junction, NJ, USA) in windowed section, two wafer Princeton Junction, NJ,Nano USA)placed placed inthe the windowed section, and twopiezoelectric piezoelectric wafersensors sensors as Princeton NJ,array USA)(manufactured placed in the windowed section, and two piezoelectric waferas part new by Inc., Boston, MA, B2 partof ofaaJunction, newsensor sensor array (manufactured byMetis MetisDesign Design Inc., Boston, MA,USA) USA)named named assensors B2and andas part ofsensors. a new sensor array (manufactured by Metis Inc., Boston, MA, named as B2 and A3 WD were at field conducted at the facility A3 sensors. WD sensors sensors were also also utilized utilized at the theDesign field testing testing conducted atUSA) the NAVAIR NAVAIR facility the sensors were attached to housing [48]. A sensor has where theWD sensors were attached to the the gearbox gearbox housing [48]. A Nano Nanoat30 30 sensor has afacility a frequency frequency A3where sensors. sensors were also utilized at the field testing conducted the NAVAIR where response of 100–400 kHz, and the required contact size in order to fit into the windowed section. response of 100–400 kHz, and the required contact size in order to fit into the windowed section. the sensors were attached to the gearbox housing [48]. A Nano 30 sensor has a frequency response sensors have been established the literature [49]. The have Piezoelectric wafer sensors have been well wellin established in thethe literature [49].section. The sensors sensors have of Piezoelectric 100–400 kHz,wafer and the required contact size order to fitin into windowed Piezoelectric disk-like geometry with the dimensions diameter and 0.5 mm aa flat disk-like geometry withwell the established dimensions of of 3the 3 mm mm diameter andThe 0.5sensors mm thickness, thickness, flat frequency frequency wafer sensors have been in literature [49]. have disk-like geometry response up and dd3131mode. All are using response upto toMHz MHz frequencies, andoperate operate inmm mode. Allof of the sensors arebonded bonded using super with the dimensions offrequencies, 3 mm diameter and 0.5in thickness, athe flatsensors frequency response up super to MHz glue and connected to 40 dB gain pre-amplifiers. The sensor couplant condition is preserved until glue and connected to 40 dB gain pre-amplifiers. The sensor couplant condition is preserved until frequencies, and operate in d31 mode. All of the sensors are bonded using super glue and connected completion of The AE isis collected using data board thedB completion of the the experiments. experiments. The AE data datacondition collected using PCI-8 PCI-8 data acquisition board to the 40 gain pre-amplifiers. The sensor couplant is preserved until theacquisition completion of the manufactured manufacturedby byMistras MistrasGroup GroupInc. Inc.The Themajor majordata dataacquisition acquisitionvariables variablesare arethe thedigital digitalfilter, filter,as as20– 20– experiments. The AE data is collected using PCI-8 data acquisition board manufactured by Mistras 400 400 kHz, kHz, hit hit definition definition time, time, as as 800 800 μs, μs, hit hit lockout lockout time, time, as as 1200 1200 μs, μs, and and 50–70 50–70 dB dB threshold threshold level level Group Inc. The major data acquisition variables are the digital filter, as 20–400 kHz, hit definition depending depending on on the the hit hit rate rate (discussed (discussed in in detail detail below). below). AE AE waveforms waveforms are are recorded recorded with with aa 33 MHz MHz time, as 800 µs, hit lockout time, as 1200 µs, and 50–70 dB threshold level depending on the hit rate sampling samplingrate ratewith withaaduration durationof of11ms. ms.Two Twostrain straingauges gaugesmanufactured manufacturedby byMicro MicroMeasurements Measurementsare are (discussed in detail below). AE waveforms are recorded with a 3 MHz sampling rate with a duration positioned positionedon onthe thenotch notchtip tipand andthe thecorner cornerof ofthe thewindowed windowedsection section(see (seeFigure Figure5b). 5b).The Thestrain straindata dataisis of collected 1collected ms. Two strain gauges manufactured by Micro Measurements arethe positioned on the notch tip using LabView using aa National National Instruments Instruments data data acquisition acquisition system system and and the LabView program. program. The The and the corner of the section (see Figure 5b). test. The strain data to istocollected usingsensors, a National strain calibration isiswindowed conducted the of AE aa strain calibration conducted at at the beginning beginning of each each test. In In addition addition AE and and strain strain sensors, new newpiezoelectric piezoelectricsensor sensorarray arraydesigned designedby byMetis MetisDesign DesignInc. Inc.isisplaced placedon onthe thewindowed windowedsection, section,as as shown shown in in Figure Figure 5c, 5c, for for ultrasonic ultrasonic and and AE AE modes. modes. The The array array has has two two actuators actuators and and six six receivers receivers in in

Metals 2017, 7, 242 Metals 2017, 7, 242

6 of 21

6 of 21

order to map the damage state of spline section using guided wave ultrasonics. Receivers are used as Instruments data acquisition system and the LabView program. The strain calibration is conducted AE sensors during fatigue testing, and UT data is collected when the specimen is under no load. In at the beginning of each test. In addition to AE and strain sensors, a new piezoelectric sensor array this paper, only the results of AE and strain data are presented. designed by Metis Design Inc. is placed on the windowed section, as shown in Figure 5c, for ultrasonic The experimental setup is shown in Figure 5a. The fatigue test was performed using an Instron and AE modes. The array has two actuators and six receivers in order to map the damage state of loading machine. The fatigue load setting was a 4 Hz frequency with max/min load ratio of 0.1 and a spline sectionload using ultrasonics. arethe used as AE sensors during fatigue testing, maximum ofguided 12 kN. wave One specimen wasReceivers tested until crack length reached 7 mm at 590,211 andcycles. UT data collected when specimen is under nofive load. In this paper, theofresults of AE and Theisexperiments werethe conducted about four to hours of each dayonly (total eight days and strain data are presented. 43 h).

(b)

(a)

(c) Figure 5. The experimental setup, (a) data acquisition systems; (b) the sensor positions; and (c) the Figure 5. The experimental setup, (a) data acquisition systems; (b) the sensor positions; and (c) the sensors on the window section. sensors on the window section.

The flattened bevel sample is modeled using Franc3D software (Version 7.0.8, Fracture Analysis Consultants Inc, Newsetup York,is NY, USA) to identify the fatigue crack growth characteristics. The Franc3D The experimental shown in Figure 5a. The test was performed using an Instron modelmachine. of the sample shownload in Figure 6a including all frequency of the details, themax/min prescribedload notch, andofits0.1 loading The isfatigue setting was a 4 Hz with ratio mesh are shown in Figure 6b. Figure 6b shows the image of the predicted crack growth direction, and a maximum load of 12 kN. One specimen was tested until the crack length reached 7 mm at whichcycles. agreesThe well with the experimental result. The four stresstointensity factors due to (total 1000 lb 590,211 experiments were conducted about five hours of each day of (4.5 eightkN) days loading along different crack surfaces are shown in Figure 6c–e. Based on the fracture model and and 43 h). observation in the test, sample the predicted crack growth rate for 12 kN loading is 12.77.0.8, × 10−6Fracture mm/cycle. The The flattened bevel is modeled using Franc3D software (Version Analysis crack length after about 550,000 cycles is calculated as 7 mm. The direction of crack growth and the Consultants Inc., New York, NY, USA) to identify the crack growth characteristics. The Franc3D model crack growth rate obtained through Franc3D shows good agreement with experimental observation of the sample is shown in Figure 6a including all of the details, the prescribed notch, and its mesh are as discussed in Section 5. shown in Figure 6b. Figure 6b shows the image of the predicted crack growth direction, which agrees well with the experimental result. The stress intensity factors due to 1000 lb (4.5 kN) loading along different crack surfaces are shown in Figure 6c–e. Based on the fracture model and observation in the test, the predicted crack growth rate for 12 kN loading is 12.7 × 10−6 mm/cycle. The crack length after about 550,000 cycles is calculated as 7 mm. The direction of crack growth and the crack growth

Metals 2017, 7, 242

7 of 21

rate obtained through Franc3D shows good agreement with experimental observation as discussed in Section 5. Metals 2017, 7, 242

(a)

7 of 21

(b)

(c)

(d)

(e)

Figure 6. The Franc3D model predicted crack growth rate(a) (a)sample samplemodel; model;(b) (b) mesh mesh around Figure 6. The Franc3D model andand thethe predicted crack growth rate around the notch tip, crack direction in the model and experiment; (c–e) crack lines along different crack surfaces the notch tip, crack direction in the model and experiment; (c–e) crack lines along different crack and corresponding SIFs. surfaces and corresponding SIFs.

5. Results 5. Results The AE data are analyzed using 2D localization, individual waveforms, and time/frequency The AEfeatures. data areThe analyzed using 2D localization, and time/frequency domain characteristics of AE signals areindividual evaluated inwaveforms, order to determine the precursor domain features. The characteristics of AE signals are evaluated in order to determine thethe precursor of of crack growth. The strain history is combined with the AE data in order to validate initiation cracktime growth. strain history combined with the piezoelectric AE data in order to validate the initiation time of theThe crack growth. Theisperformance of new sensors is evaluated by comparing of thewith crack The of performance of newThe piezoelectric sensors is evaluated by comparing withare: the thegrowth. waveforms a Nano 30 sensor. major questions sought in the experimental results waveforms of a Nano 30 sensor. The major questions sought in the experimental results are: • What are the typical AE waveforms representing fatigue crack growth developed in the complex geometry of the spline section? • What are the precursor AE signatures indicating the initiation of fatigue crack?

Metals 2017, 7, 242

• • •

8 of 21

What are the typical AE waveforms representing fatigue crack growth developed in the complex geometry of the spline section? What are the precursor AE signatures indicating the initiation of fatigue crack? Is the sensitivity of miniaturized wafer type piezoelectric sensor sufficient to operate as an AE sensor?

5.1. AE Waveforms Representing Fatigue Crack As the AE sensors are highly sensitive to secondary emissions, such as friction, it is important to reliably identify the AE waveforms representing the fatigue crack growth by source localization and the comparison of the AE data with other monitoring data. As 2D source localization requires minimum three sensors in order to locate the source, an AE event (defined as AE signals resulted in a meaningful location by the localization algorithm) is more reliably linked with source mechanisms than individual AE hits (defined as any AE signal above the threshold level). In this study, three WD sensors are placed in a triangular form to localize the AE source as shown in Figure 7. The inputs to the source localization algorithm are wave velocity, sensor positions, and arrival time differences. The prescribed wave speed is determined as 3000 m/s based on the pencil lead break (PLB) simulations. The cluster analysis is performed within 25 mm × 25 mm window in the vicinity of the notch tip, and the AE events occurred in this region are considered as the AE related to the fatigue crack growth. The selection of a fixed wave velocity, fixed threshold to determine the arrival time, and a fixed cluster window limits the target AE events as dominated by surface waves within the energy release similar to PLB. If AE signal is strong such that the first threshold crossing is a longitudinal wave, the located event will be outside the cluster window, which is then considered only in total AE events. The size of cluster is selected to allow ±12.5 mm error in source localization, which corresponds to approximately one wavelength. Figure 7 shows the AE events detected within the different ranges of the fatigue cycles. At the beginning of the fatigue test, almost every event occurred in the top pin area, and those events were identified as the frictions of near pin location. After the pin was replaced, no event was detected. With the increase of the fatigue cycles, the AE event accumulation was observed near the cluster window. It is also important to verify that the accumulations of AE events at the notch tip are related to fatigue crack growth using the strain history. Figure 8 shows the minimum and maximum envelops of strain levels throughout the fatigue testing for strain gauge 1 located near the crack tip and strain gauge 2 located near the window edge (labeled in Figure 5). The amplitude of strain gauge 1 increases abruptly at the end of 380,068 fatigue cycles, while strain gauge 2 has stable response throughout the testing. The peak value of strain gauge 1 occurs between the fatigue cycles of 380,068–484,293. The sudden increase of the strain level near the notch tip is considered as a result of plastic strain accumulation before the crack jump. Once the crack tip passes the strain gauge location, the surface that strain gauge is attached to is exposed to the compression zone, as observed in the strain envelope. Figure 9a shows the total AE events and the AE events within the cluster window shown in Figure 7 versus the cumulative fatigue cycles. The total AE events throughout test are 128 (43 in the cluster window). There is a significant increase in the cumulative AE events after about 380,068 cycles, which coincides with the increase of strain near the crack tip as shown in Figure 8. The major crack growth behavior is observed between 380,068–590,211 cycles. A microscopic image of crack growth observed after 590,211 cycles are shown in Figure 9b,c. The crack direction agrees with the numerical result as discussed in Section 4. In order to get the more accurate crack length, we used the Sonoscan Gen 6 C-SAM acoustic microscope to scan the sample. The sensor with 100 MHz is used, and the resolution can be up to 60 µm. The region of crack is shown in Figure 9d, and the length of the crack is 7.020 mm.

only in total AE events. The size of cluster is selected to allow ±12.5 mm error in source localization, which corresponds to approximately one wavelength. Figure 7 shows the AE events detected within the different ranges of the fatigue cycles. At the beginning of the fatigue test, almost every event occurred in the top pin area, and those events were identified as the frictions of near pin location. Metals 2017,the 7, 242 21 After pin was replaced, no event was detected. With the increase of the fatigue cycles, the9 of AE event accumulation was observed near the cluster window. #cycle:76,752-180,264

150

100

100

100

50

Y(mm)

150

50

50

0

0

0

-50

-50

-50

-50

0 X(mm)

50

-50

(a)

0 X(mm)

-50

50

(b)

#cycle:236,904-290,234

100

Y(mm)

100

Y(mm)

100

Y(mm)

150

50

50

0

0

0

-50

-50

-50

Metals 2017, 7, 242

0 X(mm)

50

-50

(d)

100

100

Y(mm)

150

50

0 X(mm)

(f)

50

9 of 21

0

-50

-50

50

Black dot denotes the position of WD sensor. “+” represents the AE event. The square region with a dotted line is defined as crack active region.

50

0

(g)

-50

#cycle:524,095-590,211

150

0 X(mm)

50

(e)

#cycle:484,293-524,095

-50

0 X(mm)

50

#cycle:380,068-484,293

150

50

0 X(mm)

(c)

#cycle:290,234-380,068

150

-50

Y(mm)

#cycle:180,264-236,904

150

Y(mm)

Y(mm)

#cycle:0-76,752

-50

0 X(mm)

50

(h)

Figure 7. The AE event locations detected during the fatigue tests from day 1 to day 8: (a) day 1; (b) Figure 7. The AE event locations detected during the fatigue tests from day 1 to day 8: (a) day 1; day 2; (c) day 3; (d) day 4; (e) day 5; (f) day 6; (g) day 7; and (h) day 8. (b) day 2; (c) day 3; (d) day 4; (e) day 5; (f) day 6; (g) day 7; and (h) day 8.

It is also important to verify that the accumulations of AE events at the notch tip are related to fatigue crack growth using the strain history. Figure 8 shows the minimum and maximum envelops of strain levels throughout the fatigue testing for strain gauge 1 located near the crack tip and strain gauge 2 located near the window edge (labeled in Figure 5). The amplitude of strain gauge 1 increases abruptly at the end of 380,068 fatigue cycles, while strain gauge 2 has stable response throughout the testing. The peak value of strain gauge 1 occurs between the fatigue cycles of 380,068–484,293. The sudden increase of the strain level near the notch tip is considered as a result of plastic strain accumulation before the crack jump. Once the crack tip passes the strain gauge location, the surface that strain gauge is attached to is exposed to the compression zone, as observed in the strain envelope.

increases abruptly at the end of 380,068 fatigue cycles, while strain gauge 2 has stable response throughout the testing. The peak value of strain gauge 1 occurs between the fatigue cycles of 380,068–484,293. The sudden increase of the strain level near the notch tip is considered as a result of plastic strain accumulation before the crack jump. Once the crack tip passes the strain gauge location, the surface that strain gauge is attached to is exposed to the compression zone, as observed 10 of 21 Metals 2017, 7, 242 in the strain envelope.

Metals 2017, 7, 242

10 of 21

Metals 2017, 7, 242

10 of 21

(a)

(b) Figure 8. The strain history versus cumulative AE events; (a) strain gauge 1; and (b) strain gauge 2.

Figure 9a shows the total AE events and the AE events within the cluster window shown in Figure 7 versus the cumulative fatigue cycles. The total AE events throughout test are 128 (43 in the cluster window). There is a significant increase in the cumulative AE events after about 380,068 cycles, which coincides with the increase of strain near the crack tip as shown in Figure 8. The major crack growth behavior is observed between 380,068–590,211 cycles. A microscopic image of crack growth observed after 590,211 cycles are shown in Figure 9b,c. The crack direction agrees with the numerical result as discussed in section 4. In order to get the more accurate crack length, we used the (b) Sonoscan Gen 6 C-SAM acoustic microscope to scan the sample. The sensor with 100 MHz is used, Figure 8. The strain cumulative AE events; (a)isstrain gauge 1; and 9d, (b) strain gauge 2. of and the resolution canhistory be upversus to 60 μm. The region of crack shown in Figure and the length Figure 8. The strain history versus cumulative AE events; (a) strain gauge 1; and (b) strain gauge 2. the crack is 7.020 mm.

Cumulative AE events

Figure 9a shows the total AE events and the AE events within the cluster window shown in Figure 7140 versus the cumulative fatigue cycles. The total AE events throughout test are 128 (43 in the Clustered cluster window). There AE is event a significant increase in the cumulative AE events after about 380,068 120 Total AE event cycles, which coincides with the increase of strain near the crack tip as shown in Figure 8. The major 100 crack growth behavior is observed between 380,068–590,211 cycles. A microscopic image of crack 80 growth observed after 590,211 cycles are shown in Figure 9b,c. The crack direction agrees with the numerical60result as discussed in section 4. In order to get the more accurate crack length, we used the Sonoscan40Gen 6 C-SAM acoustic microscope to scan the sample. The sensor with 100 MHz is used, and the resolution can be up to 60 μm. The region of crack is shown in Figure 9d, and the length of 20 the crack is 7.020 mm. 0 0

Cumulative AE events

140 120

2 4 Cumulative fatigue cycles

6 x 10

5

(a)

(b)

Clustered AE event Total AE event

100 80 60 40 20 0 0

2 4 Cumulative fatigue cycles

(a) (c)

6 x 10

5

(b) (d)

Figure 9. (a) Cumulative AE events and fatigue cycles, microscopic image of crack growth; (b) back

Figure 9.surface; (a) Cumulative AE events and fatigue cycles,image. microscopic image of crack growth; (b) back (c) front surface; and (d) Sonoscan microscopic surface; (c) front surface; and (d) Sonoscan microscopic image.

Metals 2017, 7, 242

11 of 21

Metals 2017, 7, 242 Metals 2017, 7, 242

11 of 21 11 of 21

The crack crack lengthdata data is onlyrecorded recorded at the end of the experiments. In order to find the The The cracklength length dataisisonly only recordedatatthe theend endofofthe theexperiments. experiments.InInorder ordertotofind findthe thecrack crack crack growth behavior, a numerical model is built to find the strain values at the locations of strain growth growthbehavior, behavior,a anumerical numericalmodel modelisisbuilt builttotofind findthe thestrain strainvalues valuesatatthe thelocations locationsofofstrain straingauges gauges gaugesthe when the crack length increases. The steel sample with different crack lengths of 15 mm, mm, when crack length increases. The steel sample with different crack lengths of 1 mm, 2 mm, when the crack length increases. The steel sample with different crack lengths of 1 mm, 2 mm, 5 mm, 2 mm, 5 mm, and 7.02 mm is modeled, as shown in Figure 10, and loaded with the maximum load and and7.02 7.02mm mmisismodeled, modeled,asasshown shownininFigure Figure10, 10,and andloaded loadedwith withthe themaximum maximumload loadapplied applied applied experimentally. The numerical and experimental strain values are compared in Figure 11. experimentally. The numerical and experimental strain values are compared in Figure 11. experimentally. The numerical and experimental strain values are compared in Figure 11.Although Although Although the numerical straingauge of strain gauge 1 is smaller the than that the experimental results, the the thenumerical numericalstrain strainofofstrain strain gauge1 1isissmaller smallerthan thanthat that theexperimental experimentalresults, results,the theresults resultsofof results of strain gauge 2 shows good agreement. The details of crack model would influence the strain gauge 2 shows good agreement. The details of crack model would influence the absolute strain gauge 2 shows good agreement. The details of crack model would influence the absolute absolute strain distribution near the cracknumerical tip. The numericalindicate results indicate that the strain of strain strain straindistribution distributionnear nearthe thecrack cracktip. tip.The The numericalresults results indicatethat thatthe thestrain strainofofstrain straingauge gauge1 1 gauge 1 drops once the crack length exceeds the position of strain gauge 1, which agrees well with the drops dropsonce oncethe thecrack cracklength lengthexceeds exceedsthe theposition positionofofstrain straingauge gauge1,1,which whichagrees agreeswell wellwith withthe the experimental results. experimental experimentalresults. results.

10. Static element with different lengths the growth direction. FigureFigure 10. Static finite finite element modelmodel with different crack crack lengths in thein growth direction. Figure 10. Static finite element model with different crack lengths in the growth direction. -4

-4

Strain Strain

4 2

4

4

7.02mm crack 7.02mm crack ↓ ↓

2

0 00 0

Strain Strain

x 10 -4 x 10 Strain gauge 1 8 Strain gauge 1 8 Strain gauge 2 Strain gauge 2 6 6

x 10 -4 x 10 Strain gauge 1 8 Strain gauge 1 8 Strain gauge 2 Strain gauge 2 6 6

↑

1

1

7.02mm↑crack 7.02mm crack 2 3 4 5 6 7 2 3 4 5 6 5 7 Number of cycles x 10 5 Number of cycles x 10

(a) (a)

2

7.02mm crack 7.02mm crack ↓

4

↓

2

0 00 0

↑

2

2

↑ crack 7.02mm 7.02mm crack 4 6 8 4 6 8 Length of crack(mm) Length of crack(mm)

10 10

(b) (b)

Figure 11. The comparison of experimental and numerical strain history; (a) experimental Figure 11. The comparison of experimental and numerical strain history; (a) experimental Figure 11. The experimental and numerical strain history; (a) experimental measurement; and comparison (b) numericalofmeasurement. measurement; and (b) numerical measurement. measurement; and (b) numerical measurement.

Once Oncethe theAE AEaccumulation accumulationnear nearthe thenotch notchtip tipisisverified verifiedasasrepresenting representingthe thecrack crackgrowth, growth,anan AE event located near the notch tip and occurred near 400,000 cycles is selected for further Once the AE accumulation near the notch tip is verified as representing the crack growth, an AE event located near the notch tip and occurred near 400,000 cycles is selected for furtheranalysis. analysis. Figure 12 shows the time domain histories and their frequency spectra of three WD sensors AE event located near the notch tip and occurred near 400,000 cycles is selected for further analysis. Figure 12 shows the time domain histories and their frequency spectra of three WD sensors detected same event. While three sensors transfer functions, their Figure 12 by shows time domain histories and theirhave frequency spectra of three WD sensors detected by detected bythe thethe same event. While three sensors havesimilar similar transfer functions, theirwaveforms waveforms show differences. The amplitude of sensor 2 is higher, and the response of sensor 3 shows higher the same event. While three sensors have similar transfer functions, their waveforms show differences. show differences. The amplitude of sensor 2 is higher, and the response of sensor 3 shows higher dispersive behavior. While the isis isotropic, sensor-source direction influences the The amplitude of sensor 2 is higher, and the response of sensor 3 shows higher dispersive behavior. dispersive behavior. While the material material isotropic, sensor-source direction influences the waveforms detected bybythe time signals not While the material is isotropic, sensor-source direction influences the waveforms detected by the the AE waveforms detected theAE AEsensors. sensors.While Whilethe the timedomain domain signalsare are notsimilar, similar, the frequency domain responses show similar The centroids ofofsensor sensors. While the time domain signals are broadband not similar, behavior. the frequency domain responses show similar frequency domain responses show similar broadband behavior. Thefrequency frequency centroids sensor1 1 to kHz, 264.6 245.1 kHz, respectively. other hand, broadband The frequency centroids sensor 1 to sensor 3 are On 224.5 kHz, 264.6 kHz,peak and tosensor sensor3behavior. 3are are224.5 224.5 kHz, 264.6kHz, kHz,and andof 245.1 kHz, respectively. Onthe the other hand, peak frequencies also depend on the sensor-source direction. For instance, the peak frequencies of sensor 245.1 kHz, respectively. On the other hand, peak frequencies also depend on the sensor-source direction. frequencies also depend on the sensor-source direction. For instance, the peak frequencies of sensor 1 1totosensor sensor3 3are are105.8 105.8kHz, kHz,258.2 258.2kHz, kHz,and and144.7 144.7kHz, kHz,respectively. respectively.Similar Similarbehavior behaviorisisobserved observedfor for the other events detected in the cluster window. the other events detected in the cluster window.

Metals 2017, 7, 242

12 of 21

For instance, the peak frequencies of sensor 1 to sensor 3 are 105.8 kHz, 258.2 kHz, and 144.7 kHz, respectively. Similar behavior is observed for the other events detected in the cluster window. 12 of 21 Metals 2017, 7, 242

200

200

0.06

(a)

0 0

1000

800

1000

100 100

1

-1 0

0

-2 -1 0 -2 0

WD sensor 1

200 300 Frequency(kHz)

(d)

WD sensor 2

0 1

200

400 600 Time(μs)


400

500

400

500

800

0 0

-1 -2 0

200

400 600 Time(μs)

(c)

200

400 600 Time(μs)

0.06


(e)

100

WD sensor 3

0

-2 0

1000

WD sensor 2

100

-1

1000

WD sensor 2

(b)

0.05 0.06 0.04 0.05 0.03 0.04 0.02 0.03 0.01 0.02 0 0.01 0

800

400(b) 600 Time(μs)

200

0.06

WD sensor 1

|Voltage| |Voltage|

0.05 0.06 0.04 0.05 0.03 0.04 0.02 0.03 0.01 0.02 0 0 0.01

(a)

400 600 Time(μs)

800

|Voltage||Voltage|

-2 0

400 600 Time(μs)

2

Voltage(V) Voltage(V)

-1 0

1

1 2


400 400

500

1000

1000

WD sensor 3

WD sensor 3

100

0 0

500

800

800

(c)

0.05 0.06 0.04 0.05 0.03 0.04 0.02 0.03 0.01 0.02 0 0.01 0

|Voltage| |Voltage|

WD sensor 1

-2 -1 0

12 of 21

WD sensor 3

WD sensor 2

1 2 0 1

2

2 WD sensor 1 Voltage(V) Voltage(V)


2 2017, 7, 242 Metals


(f)

100


400

400

500 500

(e) spectra of the AE event detected(f)by three WD Figure 12. (d) Time domain histories and frequency Figure 12. Time domain histories and frequency spectra of the AE event detected by three WD sensors, sensors, and located near the notch tip at about 400,000 cycles, where (a,d) are from sensor 1; (b,e) are Figure 12. near Timethe domain and 400,000 frequency spectra of the AEare event by(b,e) threeare WD and located notchhistories tip at about cycles, where (a,d) fromdetected sensor 1; from from sensor 2; and (c,f) are from sensor 3. sensors, and(c,f) located notch3.tip at about 400,000 cycles, where (a,d) are from sensor 1; (b,e) are sensor 2; and are near fromthe sensor from sensor 2; and (c,f) are from sensor 3.

5.2. Precursor AE Signature 5.2. Precursor AE Signature 5.2. Precursor AE Signature The fatigue test was continued for about five to six hours of testing each day. A pattern of The fatigue test was continued for about five to six hours of testing each day. A pattern of increasing continuous emission was for observed aftertoabout 237,000 cycles. each It is considered thatofthe The fatigue test was continued about five six hours of testing day. A pattern increasing continuous emission was observed after about 237,000 cycles. It is considered that the increase of continuous the notchafter tip isabout a precursor sudden crack growth increasing continuous emission emission from was observed 237,000indication cycles. It of is considered that the increase of continuous emission from the notch tip is a precursor indication of sudden crack growth jumps. 13 shows five fatigue cycles oftip different test days and the AE hits increaseFigure of continuous emission from the notch is a precursor indication of amplitudes sudden crackofgrowth jumps. Figure 13 selected shows five fatigue cycles of different test days and the amplitudes of AEthe hits recorded recorded at the fatigue cycles. Three continuous emissions that occurred near peak load jumps. Figure 13 shows five fatigue cycles of different test days and the amplitudes of AE hits atare theselected selectedfrom fatigue cycles. Three continuous emissions that near theAE peak load arethe selected each dataset, labeled the figure. Theoccurred majority of the hits near peak recorded at the selected fatigue and cycles. Threeon continuous emissions that occurred near the peak load from each dataset, and labeled on the figure. The majority of the AE hits near the peak load load are detected sensor 3, which is located across from themajority notch tip. are selected from by each dataset, and labeled on the figure. The of the AE hits near the peakare detected sensor by 3, which from thefrom notch tip. load areby detected sensor is 3, located which isacross located across the notch tip. 100

b b

a a

80 80

c c

Amplitude(dB) Amplitude(dB)

Load(kN) Load(kN)

12 10 10 8 8 6 6 4 4 2 2

60 60 40 40 20 20

0 01.2951 1.2951

1.2951 1.2951 Time(s) Time(s)

(a) (a)

1.2952 1.2952

0 0

4 104

x x 10

14 14 12 12 10 10 8 8 6 6 4 4 2 2

0 0

100 100

80 80

a

b

a

b

Amplitude(dB) Amplitude(dB)

100

14 12

Load(kN) Load(kN)

14

c

60

c

60

40

40 20

4217 4217

4217.5 Time(s) 4217.5 Time(s)

20

0 4218 0 4218

(b) (b)

Figure 13. The selected continuous emission recorded near peak load; (a) day 5, and (b) day 6. Figure 13. The selected continuous emission recorded near peak load; (a) day 5, and (b) day 6. Figure 13. The selected continuous emission recorded near peak load; (a) day 5, and (b) day 6.

Figures 14 and 15 show time histories and frequency spectra of six examples of continuous Figures 14 and 15 show time histories and frequency spectra of six examples of continuous emission signals detected by sensor 3 near the peak load. There is a distinct frequency near 200 kHz. emission signals detected by sensor 3 near the peak load. There is a distinct frequency near 200 kHz. Typical mechanical friction sources generate AE signals below 100 kHz [50–52]. The increase in the Typical mechanical friction sources generate AE signals below 100 kHz [50–52]. The increase in the frequency frequency, frequency frequencycentroid) centroid)can canbebeconsidered consideredasasthe the initiation frequency features features (e.g., (e.g., peak peak frequency, initiation of of plastic sensors are are positioned positionedin inproximity proximitytotothe thelocation locationofofcrack crackgrowth. growth. plastic deformation deformation if if the the sensors It It is is

Metals 2017, 7, 242

13 of 21

Figures 14 and 15 show time histories and frequency spectra of six examples of continuous emission signals detected by sensor 3 near the peak load. There is a distinct frequency near 200 kHz. Typical mechanical friction sources generate AE signals below 100 kHz [50–52]. The increase in the frequency features (e.g., peak frequency, frequency centroid) can be considered as the initiation Metals 2017, 7, 242 13 of 21 of Metals plastic deformation if the sensors are positioned in proximity to the location of crack growth. 2017, 7, 242 13 of 21 It is important emphasizethat thathigh-frequency high-frequencysignals signalsare aremore moreattenuative attenuativethan thanlow-frequency low-frequency important totoemphasize important to increase emphasize thatdistance high-frequency signals areand more attenuative than low-frequency signals with the of the between the sensor the source; therefore, the attenuation signals with the increase of the distance between the sensor and the source; therefore, the signals with the increase of the distance between the sensor and the source; therefore, the characteristics should be considered if the sensor isifplaced further sourcefrom implemented attenuation characteristics should be considered the sensor is from placedthe further the sourcein attenuation characteristics should be considered if the sensor is placed further from the source this study. implemented in this study. implemented in this study.

-0.2 -0.2

0.02 0.02

500 Time( 500μs) Time(μs)

0.005 0.005 0 0 0 0

100 200 300 400 100 Frequency(kHz) 200 300 400 Frequency(kHz)

-0.2 -0.2

0.02 0.02

WD sensor 3 WD sensor 3

0.01 0.01

0 0

-0.4 0 -0.4 0

1000 1000

(a) (a)

0.015 0.015

0.2 0.2


-0.4 0 -0.4 0

0.02 0.02


0.005 0.005 0 0 0 0

-0.2 -0.2

(b) (b)

0.01 0.01

500 500

0 0

1000 1000

0.015 0.015



1000 1000

(c) (c)


0.015 0.015 0.01 0.01 0.005 0.005 0 0 0 0

500 500


500 500

(f) (f)

(e) (e)

(d) (d)


0.2 0.2


0 0

0.4 0.4


|Voltage| |Voltage|


0.2 0.2

-0.4 0 -0.4 0

|Voltage| |Voltage|

0.4 0.4


|Voltage| |Voltage|


0.4 0.4

Figure 14. Examples of continuous AE signals and frequency spectra detected by sensor 3 for day 5 Figure 14.14. Examples spectra detected detectedby bysensor sensor3 3for forday day5 5 Figure Examplesofofcontinuous continuousAE AEsignals signalsand and frequency frequency spectra (290,234–380,068 cycles), where (a,d), (b,e), and (c,f) are from three different signals, respectively. (290,234–380,068 cycles), where respectively. (290,234–380,068 cycles), where(a,d), (a,d),(b,e), (b,e),and and(c,f) (c,f)are arefrom fromthree three different different signals, signals, respectively.

-0.1 -0.1

0.014 0.014 0.012 0.012 0.01 0.01 0.008 0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 0 0


(a) (a)


(d) (d)

500 500

0 0 -0.1 -0.1 -0.2 -0.2 0 0

1000 1000


0.1 0.1


0 0

0.014 0.014 0.012 0.012 0.01 0.01 0.008 0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 0 0

0.2 0.2




(e) (e)

500 500

0 0 -0.1 -0.1 500 Time( 500μs) Time(μs)

1000 1000

(c) (c)

(b) (b)



0.1 0.1

-0.2 0 -0.2 0

1000 1000

|Voltage| |Voltage|

|Voltage| |Voltage|


0.1 0.1

-0.2 0 -0.2 0

0.2 0.2


|Voltage| |Voltage|


0.2 0.2

0.014 0.014 0.012 0.012 0.01 0.01 0.008


0.008 0.006 0.006 0.004 0.004 0.002 0.002 0 0 0 0


500 500

(f) (f)

Figure 15. Examples of continuous AE signals and frequency spectra detected by sensor 3 for day 6 Figure 15. Examples of continuous AE signals and frequency spectra detected by sensor 3 for day 6 (380,068–484,293 cycles), where (a,d), andand (c,f)frequency are from three different signals, respectively. Figure 15. Examples of continuous AE(b,e), signals spectra detected by sensor 3 for day 6 (380,068–484,293 cycles), where (a,d), (b,e), and (c,f) are from three different signals, respectively. (380,068–484,293 cycles), where (a,d), (b,e), and (c,f) are from three different signals, respectively.

5.3. The Comparison of Burst and Continuous AE Signals 5.3. The Comparison of Burst and Continuous AE Signals A distinct difference between the frequency spectra of burst and continuous AE signals is A distinct difference between the frequency spectra of burst and continuous AE signals is observed as shown in Figure 16. The spectrograms are obtained using the complex Morlet Wavelet observed as shown in Figure 16. The spectrograms are obtained using the complex Morlet Wavelet transform, which gives the best resolution in time and frequency [53,54]. While the main AE energy transform, which gives the best resolution in time and frequency [53,54]. While the main AE energy is in the frequency bandwidth of 100–150 kHz for the burst signal, it shifts to 200–250 kHz for is in the frequency bandwidth of 100–150 kHz for the burst signal, it shifts to 200–250 kHz for continuous emission. Examples of burst and continuous AE signals are selected, and their AE continuous emission. Examples of burst and continuous AE signals are selected, and their AE characteristics (including peak frequency and frequency centroid) are extracted, as shown in Figure

Metals 2017, 7, 242

14 of 21

5.3. The Comparison of Burst and Continuous AE Signals A distinct difference between the frequency spectra of burst and continuous AE signals is observed as shown in Figure 16. The spectrograms are obtained using the complex Morlet Wavelet transform, which gives the best resolution in time and frequency [53,54]. While the main AE energy is in the frequency bandwidth of 100–150 kHz for the burst signal, it shifts to 200–250 kHz for continuous emission. Examples of burst and continuous AE signals are selected, and their AE characteristics (including peak frequency and frequency centroid) are extracted, as shown in Figure 17. In this figure, Metals 2017, 7, 242 14 of 21 Metals 2017,arithmetic 7, 242 14 of 21 the weighted average of the burst and continuous signals recorded at different fatigue cycles 17. In this figure, the weighted arithmetic average ofhits the recorded burst and by continuous recorded is presented. For continuous signals, about 100,000 AE channel 3signals (located acrossatnotch 17. In thisfatigue figure, cycles the weighted arithmetic average of the burst about and continuous signals recorded by at different is presented. Forare continuous 100,000 AE hits recorded tip), which occur near maximum loading selected;signals, the AE events located by three WD sensors different fatigue cycles is presented. For continuous signals, about 100,000 AE hits recorded by channel 3 (located across notch tip), which occur near maximum loading are selected; the AE events during each 3day are selected as the statistical samples for burst loading signals. In selected; general, theAE burst signals channel (located across notch tip), which occur maximum events located by three WD sensors during each day arenear selected as the statisticalare samples forthe burst signals. havelocated lower by peak frequency and frequency centroid thanasthe signals (see insignals. Figure 17), three WD signals sensors during eachpeak day are selected thecontinuous statistical samples forthe burst In general, the burst have lower frequency and frequency centroid than continuous which is consistent with the results in literature [55]. The AE amplitude does not indicate any distinct In general, the burst signals have lower peak frequency and frequency centroid than the continuous signals (see in Figure 17), which is consistent with the results in literature [55]. The AE amplitude difference to identify two signal types. The other time domain features, such as duration and rise signals (see in Figure 17), which is consistent with the results in literature [55]. The AE amplitude does not indicate any distinct difference to identify two signal types. The other time domaintime, does indicate any distinct difference identify two signal types. The other timeare domain cannot be not applied to type Thus, features in to thecontinuous frequency domain needed features, such as continuous duration and risesignals. time, tocannot be applied type signals. Thus, to features, such as duration and rise time, cannot be applied to continuous type signals. Thus, separate the AE source mechanisms. features in the frequency domain are needed to separate the AE source mechanisms. Burst type-WD sensor 3 Continuous type-WD sensor 3 Burst type-WD sensor 3 Continuous type-WD sensor 3

Norm.Voltage Norm.Voltage

0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 0

100 100

200 300 Frequency(kHz) 200 300 Frequency(kHz)

400

500

400

500

600

600

600

600

0.2

400

0.2 0.15

1.5

400

1.5 1

400

1 0.5

200 200 0 0 0 0

0.5 500 Time( 500 μs)

1000 1000

Frequeycy(kHz) Frequeycy(kHz)

1 1 0.8

Frequeycy(kHz) Frequeycy(kHz)

features in the frequency domain are needed to separate the AE source mechanisms.

400

0.15 0.1

200

0.1 0.05

200

0.05

0 0 0 0

500 Time(μs) 500 Time((c) μs)

1000 1000

Time( μs) (b) (a) (c) (b) (a) Figure 16. (a) The comparison of frequency spectra between burst type and continuous type signals; Figure 16. (a) The comparison of frequency spectra between burst type and continuous type signals; Figure 16. (a) Theofcomparison of frequency between burst type and type signals; (b) spectrogram burst type signal and (c) spectra spectrogram of continuous typecontinuous signal. (b) spectrogram of burst type signal ofcontinuous continuoustype type signal. (b) spectrogram of burst type signaland and(c) (c)spectrogram spectrogram of signal. 350

Burst signal Continuous signal Burst signal Continuous signal

300 250 250 200 200 150 150 100 100 50 50

290234-380068 290234-380068

380068-484293 484293-524095 524095-590211 Cycles 380068-484293 484293-524095 524095-590211 Cycles

(a) (a)

Frequency centroid(kHz) Frequency centroid(kHz)

PeakPeak frequency(kHz) frequency(kHz)

350 350 300

Burst signal Continuous signal Burst signal Continuous signal

350 300 300 250 250 200 200 150 150 100 100 50 50

290234-380068 290234-380068

380068-484293 484293-524095 524095-590211 Cycles 380068-484293 484293-524095 524095-590211 Cycles

(b) (b)

Figure 17. The histogram of burst and continuous signals; (a) peak frequency, and (b) frequency Figure 17. The histogram of burst and continuous signals; (a) peak frequency, and (b) frequency centroid. Figure 17. The histogram of burst and continuous signals; (a) peak frequency, and centroid.

(b) frequency centroid. 5.4. Performance of Lead Zirconate Titanate (PZT) Wafer Sensor in the AE Mode 5.4. Performance of Lead Zirconate Titanate (PZT) Wafer Sensor in the AE Mode In orderoftoLead evaluate the performance of PZT wafer sensors as AE sensors, 5.4. Performance Zirconate Titanate (PZT) Wafer Sensor in the Mode their waveforms due In order to evaluate the performance of PZT wafer sensors AE sensors, their waveforms due to PLB simulation and fatigue crack, and the ability to detect theasfatigue crack growth are compared In order to evaluate the performance of PZT wafer sensors as AE sensors, their waveforms to PLB simulation and fatigue crack, and the ability to detect the fatigue crack growth are compared with the performance of a Nano 30 sensor. The sources of AE signals compared in this section are due with the by performance of a Nano 30and sensor. The sources of AEFigure signals compared in this section are to PLB simulation fatigue crack, thesensors, ability to shown detect the fatigue crack are compared verified the and triangulation array of WD as 18. The red growth dots represent the by located the triangulation array of WD sensors, as shown Figure 18. The dotsinrepresent the are with verified theevents performance ofthree a Nano 30 sensor. The sources of AE signals compared this AE by WD sensors when PLB simulation (Figure 18a) is red performed and section fatigue AE events by three WD sensors PLB (Figure 18a) performed and fatigue crack (Figure 18b)array occurs. wave velocity is selected similar tois the section as AE verified bygrowth the located triangulation of The WD when sensors, assimulation shown Figure 18. The redprevious dots represent the crack growth (Figure 18b) occurs. The wave velocity is selected similar to the previous section as 3000 m/s. AE signals detected by the Nano 30 sensor and PZT wafer sensors at the same times of the events located by three WD sensors when PLB simulation (Figure 18a) is performed and fatigue crack 3000 m/s. AE signals detected Nanoanalyzed. 30 sensor and PZT wafer sensors at the same times of the AE(Figure events shown in Figure 18by arethe further growth 18b) occurs. The wave velocity is selected similar to the previous section as 3000 m/s. AE events shown in Figure 18 are further analyzed.

AE signals detected by the Nano 30 sensor and PZT wafer sensors at the same times of the AE events shown in Figure 18 are further analyzed.

Metals 2017, 7, 242

15 of 21

Metals 2017, 7, 242

15 of 21

150

150

100

100

Y(mm)

15 of 21

Y(mm)

Metals 2017, 7, 242

150

150

50

50

100

100

Y(mm)

Y(mm)

0

50

-50 0

0

50

-50 0

-50

-50

0 X(mm)

-50

50 -50

(a)

0 X(mm)

50

(b)

-50by all the 0 50 0 50 notch tip detected Figure 18. The selected AE-50events near the sensors; (a) an artificial event

Figure 18. The selected AE events X(mm) near the notch tip detected by X(mm) all the sensors; (a) an artificial event generated by PLB simulation; and (b) a real event generated by fatigue crack growth. (b) generated by PLB simulation; and (a) (b) a real event generated by fatigue crack growth.

Signal is applied tonear improve the of(a)PZT wafer event sensors. The Figuredenoising 18. The selected AE events the notch tip signal detectedto bynoise all the ratio sensors; an artificial generated by PLB simulation; (b) a realwavelet event by noise fatigue de-noising function a one-dimensional function to findcrack the growth. lowest wavelet Signal denoising is isapplied toand improve the generated signal to ratio of PZT wafercoefficients, sensors. The whichfunction typically is represent noise. Symletswavelet 5 is selected as thetoorthogonal wavelet wavelet and five coefficients, levels of de-noising a one-dimensional function find the lowest Signal denoising is applied to improve the signal to noise ratio aofpre-defined PZT wafer sensors. The decomposition are implemented. After wavelet coefficients below fixed threshold which typically represent noise. Symlets 5 is selected as the orthogonal wavelet and five levels of de-noising function is a one-dimensional wavelet function to find the lowest wavelet coefficients, (value of 2.5125 selected in this study) are removed from the dataset, signals are reconstructed decomposition are implemented. After wavelet below a pre-defined fixed which typically represent noise. Symlets 5 is coefficients selected as the orthogonal wavelet and fivethreshold levels of (value using an inverse wavelet transform. It is important to identify that the signal denoising process arestudy) implemented. After wavelet below aare pre-defined fixed threshold of 2.5125decomposition selected in this are removed from thecoefficients dataset, signals reconstructed using an inverse does not influence the major signal frequencies. Figure 19 shows the frequency spectra of (value of 2.5125 in this study) arethat removed from the dataset, process signals are reconstructed wavelet transform. isselected important to identify the signal denoising does not influence pre-trigger andItburst signal regions of piezoelectric wafer sensors under the PLB simulations. The the using an inverse wavelet transform. It is important to identify that the signal denoising process major electronic signal frequencies. Figure 19 shows the frequency spectra of pre-trigger and burst signal regions in thethe pre-trigger region has peak frequency near 100 The burst signal does notnoise influence major signal frequencies. Figure 19 shows thekHz. frequency spectra of has of piezoelectric wafer sensors under the PLB simulations. The electronic noise in the pre-trigger region frequencies 200 kHz–400 kHz.ofFigure 20 compares the denoised signals. pre-triggernear and burst signal regions piezoelectric wafer sensors under theand PLBoriginal simulations. The It is has peak frequency near 100denoising kHz. Theprocess burst has signal frequencies near 200 kHz–400 kHz.has Figure clearly shown that increased the signal to 100 noise ratio while the 20 electronic noise inthe the pre-trigger region peakhas frequency near kHz. The burst preserving signal major frequency ofkHz. the burst signal. The RMS-type features calculations in time frequencies nearcomponents 200 FigureIt 20 compares the denoised and original signals. It increased is and compares the denoised andkHz–400 original signals. is clearly shown that the denoising process frequency domains a short window could be used avoidtothe influence ofofreflections the AEThe clearly shown thatin the denoising process thetosignal noise ratio while preserving the the signal to noise ratio while preserving theincreased major frequency components the bursttosignal. major [56]. frequency components of the burst signal. The RMS-type features calculations in time and features RMS-type features calculations in time and frequency domains in a short window could be used to frequency domains in a short window could be used to avoid the influence of reflections to the AE avoid thefeatures influence of reflections to the AE features [56]. [56].

(a) Pre-trigger Response Pre-trigger

0.025

Response

0.02

0.01

0.005

0.005

(b) Pre-trigger

0.02

|Voltage|

0.01

0.025

|Voltage|

|Voltage|

0.015

|Voltage|

(b)

(a)

0.015

Response Pre-trigger Response

0.015

0.015

0.01

0.01

0.005

0.005

0 00

0

500 500 Frequency(kHz) Frequency(kHz)

(c) (c)

1000

1000

0

0 0 0

500

500 Frequency(kHz) Frequency(kHz)

1000

1000

(d)(d)

Figure 19. Time and frequency domain response of pre-trigger and burst signal regions of PZT wafer sensors, where (a,c) are from Metis A3; and (b,d) are from Metis B2.

Metals 2017, 7, 242

16 of 21

Metals 2017, 7,Figure 242 19. Time and frequency domain response of pre-trigger and burst signal regions of PZT wafer sensors, where (a,c) are from Metis A3; and (b,d) are from Metis B2. 0.3

0.3

Denoised

0.1 0 -0.1 -0.2

0.1 0 -0.1 -0.2

0

200

400 600 Time(μs)

800

1000

0

0.3

0 -0.1 -0.2 200

400 600 Time(μs)

0.1 0 -0.1

800

1000

0

200

400 600 Time(μs)

800

1000

(b)

-3

4 Denoised Original

2 1

x 10

-3

Denoised Original

3 |Voltage|

3 |Voltage|

1000

Original

(a)

0 0

800

-0.2

0

4

400 600 Time(μs)

0.2 Voltage(V)

0.1

200

0.3

Original

0.2 Voltage(V)

Denoised

0.2 Voltage(V)

Voltage(V)

0.2

x 10

16 of 21

2 1

200


800

1000

0 0

200


(c)

800

1000

(d)

Figure 20. The time and frequency domain responses of PZT wafer sensors generated by PLB

Figure 20. The time and frequency domain responses of PZT wafer sensors generated by PLB simulation, where (a,c) are from Metis A3, and (b,d) are from Metis B2. simulation, where (a,c) are from Metis A3, and (b,d) are from Metis B2. The denoising process has been applied to all the waveforms obtained by PLB simulations and

Voltage(V)

Voltage(V)

Voltage(V)

fatigue testing. process Figure 21has shows time histories andwaveforms frequency spectra of signals from Nano 30 and The denoising beenthe applied to all the obtained by PLB simulations and two PZT wafers obtained by PLB simulations. A windowed section of complete time histories fatigue testing. Figure 21 shows the time histories and frequency spectra of signals from Nano 30 is shown in Figure 22. While the AE source is the same and the sensors’ locations are similar, the and two PZT wafers obtained by PLB simulations. A windowed section of complete time histories responses of Nano 30 and PZT wafer sensors are different due to different transfer functions. The is shown Figure 22. have While the amplitude AE source is to theitssame andsmall the sensors’ locationsthey are exhibit similar, the PZTin wafer sensors lower due relatively size. Additionally, responses of Nano and (i.e., PZThigher waferdamping) sensors are different due totime different transfer The PZT a lower quality30factor with lower ringing than the Nano 30functions. sensor. wafer sensors have lower amplitude due to its relatively small size. Additionally, they exhibit a lower 0.3 0.3 quality factor (i.e., higher damping) with lower ringing time Metis A3than the Nano 30 sensor. Metis B2 Nano 30 0.2 0.2 1 Figure 23 compares the waveforms 0.1during the fatigue experiment, 0.1 of two sensor types detected and the selected AE event occurred near0400,000 cycles. Similar to PLB0 simulations, the amplitudes of 0 -0.1 -0.1 PZT wafer sensors are lower than the amplitude of Nano 30. It is interesting to note that the waveforms -1 -0.2 -0.2 and frequency distributions of AE signals due to fatigue crack growth are similar to PLB simulations, 0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 800 1000 Time(μtimes s) Time( μs) as shown in Figure 22,μ600 fatigue crack signal is about ten smaller than Time( s)while the amplitude of the (b) (a) signal. This indicates the difference the amplitude of PLB of energy levels of two(c)source mechanisms with similar frequency spectrum.

fatigue testing. Figure 21 shows the time histories and frequency spectra of signals from Nano 30 and two PZT wafers obtained by PLB simulations. A windowed section of complete time histories is shown in Figure 22. While the AE source is the same and the sensors’ locations are similar, the responses of Nano 30 and PZT wafer sensors are different due to different transfer functions. The PZT wafer Metals 2017, 7, 242sensors have lower amplitude due to its relatively small size. Additionally, they exhibit 17 of 21 a lower quality factor (i.e., higher damping) with lower ringing time than the Nano 30 sensor. 0.3

0.3

Nano 30

0 -1

0.1 0 -0.1

200

400 600 Time(μs)

800

0

1000

200

Metals 2017, 7, 242 (a) Metals 2017, 7, 242

0.03

800

0

1000

200

400 600 Time(μs)

0.03 0.02 0.02 0.01 0.01 200 300 Frequency(kHz) 200 300 Frequency(kHz)

400

500

400

500

-3

x 10

4

-3

Metis A3

4

Metis A3

3 2 2 1

x 10

(d) (d)

500

400

500

Metis B2 Metis B2

2 2 1 1

400

-3

3

0 0 0 0


17 of 21 17 of 21

3

1

100

1000

-3

x 10

0 0 0 0

100

800

(c)

3 |Voltage| |Voltage|

|Voltage| |Voltage|

4

Nano 30

100

400 600 Time(μs)

|Voltage| |Voltage|

0.04

x 10

4

Nano 30

100

0 -0.1

(b)

0.04

0 0 0 0

0.1

-0.2

-0.2

0

Metis B2

0.2 Voltage(V)

Voltage(V)

1 Voltage(V)

Metis A3

0.2

100 100


400

500

400

500

(f) (f)

(e) (e)

Figure 21. Complete time history waveforms and their frequency spectra generated by PLB Figure 21. Complete time historyhistory waveforms and their frequency spectra generated by PLB simulations, Figure 21. Complete waveforms and their frequency spectra generated by PLB simulations, where (a,d)time are from Nano 30; (b,e) are from Metis A3; and (c,f) are from Metis B2. where (a,d) are from Nano 30; (b,e) are from Metis A3; and (c,f) are from Metis B2. simulations, where (a,d) are from Nano 30; (b,e) are from Metis A3; and (c,f) are from Metis B2. 0.15 0.1

Original

0 -0.5 -0.5 -1 0 -1 0

20 20

Time(μs)

0.1 0.05

-0.050 -0.05 -0.1

60

-0.1 0

40

60

0

Time(μs)

1

1 0.8

Denoised Original Denoised

0.8 0.6

Original

0.6 0.4 0.4 0.2 0.2 0 0 0 0

500 Frequency(kHz) 500 Frequency(kHz) (d)

(d)

1000 1000



1

Original

0.05 0

40

(a) (a)

Denoised Original Denoised Voltage(V) Voltage(V)

0.5 0

0.15



0.2

Original

0 0 -0.2 -0.2

60

0

20

Time(μs) 20 40 Time(μs)

60

0

20

1


0.8 0.6

Original

0.6 0.4 0.4 0.2 500 Frequency(kHz) 500 Frequency(kHz) (e)

(e)

1000 1000

Time(μs)

40

60

40

60

Time(μs)

(b) (b)

1 0.8

0.2 0 0 0 0

0.2

40

20



1 0.5


1

(c) (c)

1 0.8


0.8 0.6

Original

0.6 0.4 0.4 0.2 0.2 0 0 0 0

500 Frequency(kHz) 500 Frequency(kHz)

1000 1000

(f) (f)

Figure 22. Windowed time history waveforms and their frequency spectra generated by PLB Figure 22. Windowed history frequency spectra generated by PLB simulations, where (a,d) time are from Nanowaveforms 30; (b,e) areand fromtheir Metis A3; and (c,f) are from Metis B2. Figure 22. Windowed time history waveforms and their frequency spectra generated by PLB simulations, where (a,d) are from Nano 30; (b,e) are from Metis A3; and (c,f) are from Metis B2. simulations, where (a,d) are from Nano 30; (b,e) are from Metis A3; and (c,f) are from Metis B2.

Figure 23 compares the waveforms of two sensor types detected during the fatigue experiment, Figure 23 compares waveforms two sensor detected during the fatigue experiment, and the selected AE eventthe occurred nearof 400,000 cycles.types Similar to PLB simulations, the amplitudes and the selected AE event occurred near 400,000 cycles. Similar to PLB simulations, the amplitudes Based on the simulations and fatigue experiments, it is concluded that new piezoelectric sensor of PZT wafer sensors are lower than the amplitude of Nano 30. It is interesting to note that the of can PZTbe wafer areAE lower than While the amplitude of Nano 30. is interesting tosimilar note that the array utilized as an sensor. their sensitivities areItcrack lower than bulky commercial waveforms andsensors frequency distributions of AE signals due to fatigue growth are to PLB waveforms and frequency distributions of AE due tooffatigue crack growth are similar toofPLB AE simulations, sensors, they be in placed in22, proximity toamplitude crack locations inside the spline section the as can shown Figure while thesignals the fatigue crack signal is about ten simulations, as shown in Figure 22, while the amplitude of the fatigue crack signal is about ten gearbox, the transmission loss ofThis signal passingthe through all the layers of the gearbox times which smallereliminates than the amplitude of PLB signal. indicates difference of energy levels of two times smaller than the amplitude of PLB signal. This the difference of energy levels of two source mechanisms with similar spectrum. components. The cumulative AE frequency signals detected byindicates piezoelectric wafer sensors near maximum source mechanisms with similar frequency spectrum. on the simulations and fatigue experiments, it is noise concluded piezoelectric loading Based with amplitude greater than 52 dB (above electronic level)that arenew shown in Figure sensor 24. The Based on the simulations and fatigue experiments, it is concluded that new piezoelectric sensor array can be utilized as an AE sensor. While their sensitivities are lower than bulky commercial AE9a. piezoelectric wafer sensors detect crack growth behavior similar to WD sensors, as shown in Figure array canthey be utilized as an AE sensor. While their locations sensitivities are the lower thansection bulky commercial AE sensors, can be placed in proximity to crack inside spline of the gearbox, sensors, they can be placed in proximity to crack locations inside the spline section of the gearbox, which eliminates the transmission loss of signal passing through all the layers of the gearbox which eliminates the transmission loss of signal passing through wafer all thesensors layers near of the gearbox components. The cumulative AE signals detected by piezoelectric maximum components. The cumulative AE signals detected by piezoelectric wafer sensors near maximum loading with amplitude greater than 52 dB (above electronic noise level) are shown in Figure 24. loading with amplitude greater detect than 52 dB (above aresensors, shown in The piezoelectric wafer sensors crack growthelectronic behavior noise similarlevel) to WD as Figure shown 24. in The piezoelectric wafer sensors detect crack growth behavior similar to WD sensors, as shown in Figure 9a. Figure 9a.

Metals 2017, 7, 242

18 of 21

Metals 2017, 7, 242 Metals 2017, 7, 242

18 of 21 18 of 21

0.4 0.4

0

-0.2 -0.2

20 40 20 Time(μs) 40 Time(μs)

0

-0.05 -0.05 0 0

60 60

1 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 0

1000 1000

1 1 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 0

Denoised Denoised Original Original


1000 1000

20

40 Time(μs) Time(μs)

40

60

60



1000 1000

(f) (f)

(e) (e)

(d) (d)

20

(c) (c)

(b) (b)



0


0.4 0.4 0.2 0.2 0 0 0 0

60 60

-0.02 -0.02 0 0

(a) (a)


1 1 0.8 0.8 0.6 0.6

20 Time( s) 40 20 Time( s)μ40 μ

0


-0.4 -0.4 0 0

0




0

0.05 0.05

0.02 0.02

0.2 0.2

Figure 23. Windowed time history waveforms and their frequency spectra generated by fatigue Figure history waveforms and their frequency spectra generated by fatigue crack Figure23. 23.Windowed Windowedtime time history waveforms and their frequency spectra generated by fatigue crack growth, where (a,d) are from Nano 30, (b,e) are from Metis A3, and (c,f) are from Metis B2. growth, where (a,d) from (b,e) from A3, and MetisMetis B2. B2. crack growth, whereare (a,d) areNano from 30, Nano 30,are (b,e) are Metis from Metis A3,(c,f) and are (c,f)from are from

Cumulative AE Signals Cumulative AE Signals

70 70 60 60 50 50 40 40 30 30 20 20 10 10 0 0 0 0

Detected by Metis B2(CH5) Detected by Metis B2(CH5) Detected by Metis A3(CH6) Detected by Metis A3(CH6)

1

1

2 3 4 2 3 fatigue cycles 4 Cumulative Cumulative fatigue cycles

5

5

6 65 x 10 5 x 10

Figure 24. Cumulative AE signals from Metis A3 and Metis B2 and fatigue cycles. Figure24. 24.Cumulative Cumulative AE signals signals from Figure from Metis MetisA3 A3and andMetis MetisB2 B2and andfatigue fatiguecycles. cycles.

6. Conclusions Conclusions 6.6.Conclusions The detection of a fatigue crack in the idealized spline component of a gearbox using the AE Thedetection detectionof of afatigue fatigue crack crack in in the the idealized spline ofofa agearbox using thethe AE The idealized splinecomponent component method is presenteda in this paper. The crack length reached is about 7 mm in gearbox an angledusing directionAE at method is presented in this paper. The crack length reached is about 7 mm in an angled direction at method is cycles, presented in agrees this paper. Thenumerical crack length reached is about 7 mm in an angled direction 590,211 which with the model. A detailed waveform analysis indicates that 590,211 cycles, which agrees with the numerical model. A detailed waveform analysis indicates that at 590,211 cycles, which agrees the numerical model. A detailed indicates the plastic deformation and with the crack jump have two distinct AEwaveform signaturesanalysis in the forms of the plastic deformation and the crack jump have two distinct AE signatures in the forms of thatcontinuous the plasticemission deformation and the crack jump have two distinct AE signatures in the forms of and burst emission, respectively. Frequency domain features play significant continuous emission and burst emission, respectively. Frequency domain features play significant continuous emission damage and burst emission, respectively. FrequencyAdomain features play significant roles in separating modes in complex metallic structures. new piezoelectric sensor array is roles in separating damage modes in complex metallic structures. A new piezoelectric sensor array is roles in separating modes complex metallic structures. A new crack piezoelectric sensor introduced with damage the intention of in placing them in proximity to potential locations. Whilearray the introduced with the intention of placing them in proximity to potential crack locations. While the sensors have low compared to conventional AE sensors dueWhile to their is introduced with thesensitivity intention of placing them in proximitypiezoelectric to potential crack locations. the sensors have low sensitivity compared to conventional piezoelectric AE sensors due to their relatively size, theycompared can detecttothe major crack growth events. As the due sensor arrayrelatively can be sensors have small low sensitivity conventional piezoelectric AE sensors to their relatively small size, they can detect the major crack growth events. As the sensor array can be placed inside thedetect gearbox, the propagation path andAs thethesignal loss willcan not affect the signal small size, they can the major crack growth events. sensor array be placed inside placed inside the gearbox, the propagation path and the signal loss will not affect the signalthe signature and energy as compared to conventional AE sensors placed on the gearbox housing. The gearbox, the propagation path and the signal loss will not affect the signal and energy signature and energy as compared to conventional AE sensors placed on thesignature gearbox housing. The as research will continue by adapting the AE patterns of two damage modes to the AE data recorded compared to conventional AE sensors on theof gearbox housing. Theto research research will continue by adapting theplaced AE patterns two damage modes the AE will datacontinue recordedby from the realistic test stand at damage the NAVAIR facility. adapting the AE patterns of two modes to the AE data recorded from the realistic test stand at from the realistic test stand at the NAVAIR facility. the NAVAIR facility.

Acknowledgments: This material is based upon work supported by the U.S. Naval Air Systems Command Acknowledgments: This material is based upon work supported by the U.S. Naval Air Systems Command (NAVAIR) underThis contract No. isN68335-13-C-0417 entitled “Hybrid State-Detection System for Gearbox Acknowledgments: material based upon work supported by the U.S. Naval Air Systems (NAVAIR) under contract No. N68335-13-C-0417 entitled “Hybrid State-Detection System for Command Gearbox Components” awarded to the Metis Design Corporation. The authors would like to thank Seth Kessler and (NAVAIR) under contract No.Metis N68335-13-C-0417 entitled “Hybridwould State-Detection System for Gearbox Components” awarded to the Design Corporation. The authors like to thank Seth Kessler and Gregory James for their inputs onDesign the design of the testThe specimen. acknowledge Daniel Bailey and for detailed Components” awarded to the Metis Corporation. authorsWe would like to thank SethP.Kessler Gregory Gregory James for their inputs on the design of the test specimen. We acknowledge Daniel P. Bailey for detailed James for their inputs on theand design of the test We acknowledge Daniel P. Baileyand for detailed editing editing of the abstract discussion of specimen. the manuscript. Any opinions, findings, conclusions or editing of the abstract and discussion of the manuscript. Any opinions, findings, and conclusions or

Metals 2017, 7, 242

19 of 21

of the abstract and discussion of the manuscript. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of NAVAIR. Author Contributions: Lu Zhang and Didem Ozevin conducted the experiments and analyzed the data. Lu Zhang also conducted the literature survey and wrote the background literature. William Hardman and Alan Timmons performed the Franc3D simulations and provided feedback on fatigue experiments and data processing. Conflicts of Interest: The authors declare no conflict of interest.

References 1. 2. 3. 4. 5. 6. 7.

8. 9. 10.

11. 12. 13. 14. 15. 16. 17.

18.

19.

Sinha, P.; Bhattacharyya, S. Failure investigation of spline-shaft of an under slung crane. J. Fail. Anal. Prev. 2013, 13, 601–606. [CrossRef] Yu, Z.; Xu, X. Failure analysis of a diesel engine crankshaft. Eng. Fail. Anal. 2005, 12, 487–495. [CrossRef] Reuben, L.C.K.; Mba, D. Bearing time-to-failure estimation using spectral analysis features. Struct. Health Monit. 2014, 13, 219–230. [CrossRef] Hussain, S.; Gabbar, H.A. Gearbox fault detection using real coded genetic algorithm and novel shock response spectrum features extraction. J. Nondestruct. Eval. 2014, 33, 111–123. [CrossRef] Lin, S.T.; McFadden, P.D. Gear vibration analysis by B-spline wavelet-based linear wavelet transform. Mech. Syst. Signal Process. 1997, 11, 603–609. [CrossRef] Liu, B.; Riemenschneider, S.; Xu, Y. Gearbox Fault Diagnosis Using Empirical Mode Decomposition and Hilbert Spectrum. Mech. Syst. Signal Process. 2006, 20, 718–734. [CrossRef] Bonnardot, F.; El Badaoui, M.; Randall, R.B.; Danière, J.; Guillet, F. Use of the acceleration signal of a gearbox in order to perform angular resampling (with limited speed fluctuation). Mech. Syst. Signal Process. 2005, 19, 766–785. [CrossRef] Leturiondo, U.; Salgado, O.; Galar, D. Validation of a physics-based model of a rotating machine for synthetic data generation in hybrid diagnosis. Struct. Health Monit. 2016. [CrossRef] Pandya, Y.; Parey, A. Failure path based modified gear mesh stiffness for spur gear pair with tooth root crack. Eng. Fail. Anal. 2013, 27, 286–296. [CrossRef] Li, Z.; Yan, X.; Tian, Z.; Yuan, C.; Peng, Z.; Li, L. Blind vibration component separation and nonlinear feature extraction applied to the nonstationary vibration signals for the gearbox multi-fault diagnosis. Meas. J. Int. Meas. Confed. 2013, 46, 259–271. [CrossRef] Obuchowski, J.; Wylomanska, A.; Zimroz, R. Vibration Engineering and Technology of Machinery. Mech. Mach. Sci. 2015, 23, 401–410. Yunusa-Kaltungo, A.; Sinha, J.K.; Nembhard, A.D. A novel fault diagnosis technique for enhancing maintenance and reliability of rotating machines. Struct. Health Monit. 2015, 14, 1–18. [CrossRef] Elasha, F.; Greaves, M.; Mba, D.; Addali, A. Application of Acoustic Emission in Diagnostic of Bearing Faults within a Helicopter Gearbox. Procedia CIRP 2015, 38, 30–36. [CrossRef] Eftekharnejad, B.; Mba, D. Seeded fault detection on helical gears with acoustic emission. Appl. Acoust. 2009, 70, 547–555. [CrossRef] Gu, D.; Kim, J.; An, Y.; Choi, B. Detection of faults in gearboxes using acoustic emission signal. J. Mech. Sci. Technol. 2011, 25, 1279–1286. [CrossRef] Aggelis, D.; Shiotani, T.; Papacharalampopoulos, A.; Polyzos, D. The influence of propagation path on elastic waves as measured by acoustic emission parameters. Struct. Health Monit. 2012, 11, 359–366. [CrossRef] Khazaee, M.; Ahmadi, H.; Omid, M.; Moosavian, A.; Khazaee, M. Classifier fusion of vibration and acoustic signals for fault diagnosis and classification of planetary gears based on Dempster-Shafer evidence theory. Proc. Inst. Mech. Eng. Part E 2014, 228, 21–32. [CrossRef] Li, C.; Sanchez, R.-V.; Zurita, G.; Cerrada, M.; Cabrera, D.; Vásquez, R.E. Gearbox fault diagnosis based on deep random forest fusion of acoustic and vibratory signals. Mech. Syst. Signal Process. 2016, 76–77, 283–293. [CrossRef] Loutas, T.H.; Roulias, D.; Pauly, E.; Kostopoulos, V. The combined use of vibration, acoustic emission and oil debris on-line monitoring towards a more effective condition monitoring of rotating machinery. Mech. Syst. Signal Process. 2011, 25, 1339–1352. [CrossRef]

Metals 2017, 7, 242

20.

21. 22.

23.

24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39.

40. 41.

42.

43.

20 of 21

Qu, Y.; He, D.; Yoon, J.; Van Hecke, B.; Bechhoefer, E.; Zhu, J. Gearbox tooth cut fault diagnostics using acoustic emission and vibration sensors—A comparative study. Sensors 2014, 14, 1372–1393. [CrossRef] [PubMed] Rezaei, A.; Dadouche, A. Development of a turbojet engine gearbox test rig for prognostics and health management. Mech. Syst. Signal Process. 2012, 33, 299–311. [CrossRef] Al-Ghamd, A.M.; Mba, D. A comparative experimental study on the use of acoustic emission and vibration analysis for bearing defect identification and estimation of defect size. Mech. Syst. Signal Process. 2006, 20, 1537–1571. [CrossRef] Bigus, G.A.; Travkin, A.A. An evaluation of the flaw-detection characteristics for the detection of fatigue cracks by the acoustic-emission method in samples made of steel 20 that have a cast structure. Russ. J. Nondestruct. Test. 2015, 51, 32–38. [CrossRef] Yu, J.G.; Ziehl, P. Stable and unstable fatigue prediction for A572 structural steel using acoustic emission. J. Constr. Steel Res. 2012, 77, 173–179. [CrossRef] Rastegaev, I.; Danyuk, A.; Afanas’yev, M.; Merson, D.; Berto, F.; Vinogradov, A. Acoustic emission assessment of impending fracture in a cyclically loading structural steel. Metals 2016, 6, 266. [CrossRef] Huang, M.; Jiang, L.; Liaw, P.K.; Brooks, C.R.; Seeley, R.; Klarstrom, L.D. Using Acoustic Emission in Fatigue and Fracture Materials Research. Mater. Res. 1998, 50, 1–13. Burridge, R.; Knopoff, L. Body Force Equivalents for Seismic Dislocations. Seismol. Res. Lett. 2003, 74, 154–162. [CrossRef] Ohtsu, M. Acoustic emission theory for moment tensor analysis. J. Res. Nondestruct. Eval. 1995, 169–184. [CrossRef] Ohtsu, M. Source kinematics of acoustic emission based on a moment tensor. NDT Int. 1989, 22, 14–20. Ohtsu, M.; Ono, K. The generalized theory and source representations of acoustic emission. J. Acoust. Emiss. 1986, 5, 124–133. Healy, D.; Jones, R.R.; Holdsworth, R.E. Three-dimensional brittle shear fracturing by tensile crack interaction. Nature 2006, 439, 64–67. [CrossRef] [PubMed] Kao, C.-S.; Carvalho, F.C.S.; Labuz, J.F. Micromechanisms of fracture from acoustic emission. Int. J. Rock Mech. Min. Sci. 2011, 48, 666–673. [CrossRef] Ohira, T.; Pao, Y. Quantitative characterization of microcracking in A533B steel by acoustic emission. Metall. Trans. A 1989, 20, 1105–1114. [CrossRef] Ohno, K.; Uji, K.; Ueno, A.; Ohtsu, M. Fracture process zone in notched concrete beam under three-point bending by acoustic emission. Constr. Build. Mater. 2014, 67, 139–145. [CrossRef] Sause, M.G.R.; Horn, S. Simulation of Acoustic Emission in Planar Carbon Fiber Reinforced Plastic Specimens. J. Nondestruct. Eval. 2010, 29, 123–142. [CrossRef] Yu, J.; Ziehl, P.; Matta, F.; Pollock, A. Acoustic emission detection of fatigue damage in cruciform welded joints. J. Constr. Steel Res. 2013, 86, 85–91. [CrossRef] Li, R.; Seçkiner, S.U.; He, D.; Bechhoefer, E.; Menon, P. Gear fault location detection for split torque gearbox using AE sensors. IEEE Trans. Syst. Man Cybern. Part C 2012, 42, 1308–1317. [CrossRef] Poddar, B.; Giurgiutiu, V. Detectability of Crack Lengths from Acoustic Emissions Using Physics of Wave Propagation in Plate Structures. J. Nondestruct. Eval. 2017, 36, 41. [CrossRef] Caesarendra, W.; Kosasih, B.; Tieu, A.K.; Zhu, H.; Moodie, C.A.S.; Zhu, Q. Acoustic emission-based condition monitoring methods: Review and application for low speed slew bearing. Mech. Syst. Signal Process. 2015, 72, 134–159. [CrossRef] Li, R.; He, D. Rotational machine health monitoring and fault detection using EMD-based acoustic emission feature quantification. IEEE Trans. Instrum. Meas. 2012, 61, 990–1001. [CrossRef] Soua, S.; Van Lieshout, P.; Perera, A.; Gan, T.H.; Bridge, B. Determination of the combined vibrational and acoustic emission signature of a wind turbine gearbox and generator shaft in service as a pre-requisite for effective condition monitoring. Renew. Energy 2013, 51, 175–181. [CrossRef] Siracusano, G.; Lamonaca, F.; Tomasello, R.; Garescì, F.; Corte, A.L.; Carnì, D.L.; Carpentieri, M.; Grimaldi, D.; Finocchio, G. A framework for the damage evaluation of acoustic emission signals through Hilbert-Huang transform. Mech. Syst. Signal Process. 2016, 75, 109–122. [CrossRef] Raja Hamzah, R.I.; Mba, D. The influence of operating condition on acoustic emission (AE) generation during meshing of helical and spur gear. Tribol. Int. 2009, 42, 3–14. [CrossRef]

Metals 2017, 7, 242

44.

45. 46. 47. 48.

49. 50. 51. 52. 53. 54.

55. 56.

21 of 21

Loutas, T.H.; Sotiriades, G.; Kalaitzoglou, I.; Kostopoulos, V. Condition monitoring of a single-stage gearbox with artificially induced gear cracks utilizing on-line vibration and acoustic emission measurements. Appl. Acoust. 2009, 70, 1148–1159. [CrossRef] Tan, C.K.; Mba, D. Identification of the acoustic emission source during a comparative study on diagnosis of a spur gearbox. Tribol. Int. 2005, 38, 469–480. [CrossRef] Tan, C.K.; Mba, D. Limitation of acoustic emission for identifying seeded defects in gearboxes. J. Nondestruct. Eval. 2005, 24, 11–28. [CrossRef] Toutountzakis, T.; Mba, D. Observations of acoustic emission activity during gear defect diagnosis. NDT E Int. 2003, 36, 471–477. [CrossRef] Ozevin, D.; Cox, J.; Hardman, W.; Kessler, S.; Timmons, A. Fatigue Crack Detection at Gearbox Spline Component using Acoustic Emission Method. In Proceedings of the Annual Conference of the Prognostics and Health Management Society, Forth Worth, TX, USA, 29 September–2 October 2014. Giurgiutiu, V. Structural Health Monitoring: With Piezoelectric Wafer Active Sensors; Academic Press: Cambridge, MA, USA, 2007. Boyun, G.; Song, S.; Ghalambor, A.; Lin, T.R. Offshore Pipelines Design, Installation, and Maintenance; Gulf Professional Publishing: Houston, TX, USA, 2014; p. 383. Hase, A.; Mishina, H.; Wada, M. Correlation between features of acoustic emission signals and mechanical wear mechanisms. Wear 2012, 292–293, 144–150. [CrossRef] Rogers, L.M. The application of vibration signature analysis and acoustic emission source location to on-line condition monitoring of anti-friction bearings. Tribol. Int. 1979, 12, 51–58. [CrossRef] Lin, J.; Qu, L. Feature Extraction Based on Morlet Wavelet and Its Application for Mechanical Fault Diagnosis. J. Sound Vib. 2000, 234, 135–148. [CrossRef] Mostavi, A.; Kamali, N.; Tehrani, N.; Chi, S.W.; Ozevin, D.; Indacochea, J.E. Wavelet based harmonics decomposition of ultrasonic signal in assessment of plastic strain in aluminum. Meas. J. Int. Meas. Confed. 2017, 106, 66–78. [CrossRef] Pomponi, E.; Vinogradov, A. A real-time approach to acoustic emission clustering. Mech. Syst. Signal Process. 2013, 40, 791–804. [CrossRef] Martínez-Jequier, J.; Gallego, A.; Suárez, E.; Juanes, F.J.; Valea, Á. Real-time damage mechanisms assessment in CFRP samples via acoustic emission Lamb wave modal analysis. Composites Part B 2015, 68, 317–326. [CrossRef] © 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).