A non-contact method for part-based process performance monitoring ...

Int J Adv Manuf Technol DOI 10.1007/s00170-015-7523-2

ORIGINAL ARTICLE

A non-contact method for part-based process performance monitoring in end milling operations Hadi Fekrmandi1 · Muhammet Unal2 · Amin Baghalian1 · Shervin Tashakori1 · Kathleen Oyola1 · Abdullah Alsenawi1 · Ibrahim Nur Tansel1

Received: 11 August 2014 / Accepted: 1 July 2015 © Springer-Verlag London 2015

Abstract Surface response to excitation (SuRE) method was originally developed for structural health monitoring (SHM) applications. SuRE was used to evaluate the performance of completed milling operations. The method generates surface waves on the plate and studies the spectrum changes at selected points to detect defects and change of compressive forces. In this study, the length, depth, and width of a slot were changed step by step. The surface of the aluminum plate was excited in the 20–400 kHz range with a piezoelectric element. A laser scanning vibrometer was used to monitor the vibrations at the predetermined grid points after the dimensions of the slot were changed methodically. The frequency spectrums of measured vibrations were calculated by using the Fast Fourier Transformation (FFT). The sums of the squares of the differences (SSD) of the spectrums were calculated to evaluate the change of the spectrums. The SuRE method was able to determine if the dimensions were changed in each case at all the selected points. The scanning laser vibrometer is not feasible to be used at the shop floor. However, the study demonstrated that a piezoelectric element attached to any of the grid points would be able to evaluate the completed machining process.

 Hadi Fekrmandi

[email protected] 1

Mechatronics Research Laboratory, Mechanical and Materials Engineering Department, Florida International University, 10555 W Flagler Street EC3420, Miami, FL 33174, USA

2

Department of Electronics & Computer Education, Marmara University, Goztepe Campus, Kadikoy, Istanbul, Turkey

Keywords Structural health monitoring method · Manufacturing process monitoring · Milling · Surface response to excitation method

1 Introduction Application of CNC milling machines plays an important role in manufacturing automation of manufacturing in automotive and aerospace industries. Still the quality control aspect of manufacturing is not fully automated and usually requires expert human work force. This makes the manufacturing process expensive and slow and leaves the products vulnerable to human errors. In order to stay competitive in a market with low-payment workforce, development of a full scope of automated manufacturing is necessary. There are two fundamental approaches in the machining condition monitoring (MCM) filed; direct and indirect methods [1]. Direct methods such as vision-based or optical methods directly measure the dimensions of tool or workpiece to identify the dimensional accuracy. But their application requires the tool and work-piece to be cleaned from chips and fluids. There are still significant challenges for industrial application of current MCM systems. A lot of studies have been devoted to develop effective MCM systems. During the metal cutting operations, metal chips and cooling fluids could easily block the vision of camera. Due to these reasons, although being very accurate, practical application of direct methods have been limited to laboratory. On the other side, the indirect methods have the advantage of not interrupting the operation. During the manufacturing operations, several parameters influenced and could be employed for monitoring state of tool or metal removal process [2]. Among those parameters measurement of cutting forces [3], acoustic emission signal [4], and ultrasonic

Int J Adv Manuf Technol

signal [5] have been used more frequently. Based on the measurement of angular cutting force characteristic and IAS analysis, Lamraoui et al. [6] investigated the efficiency of Short Angular Fourier Transform (SAFT) methods in the angular frequency domain and developed chatter indicators. However, their method may not be used to monitor the chatter phenomenon in an industrial process due to the use of an expensive dynamometer. Machining process monitoring could be addressed from different aspects. There is a significant amount of literature related to the tool condition monitoring (TCM) filed. Hsieh et al. [7] applied a neural network method for spindle vibration-based tool wear monitoring in micro-milling. However, their methodology was not able to distinguish if the change in energy level of the other frequency domain signals were due to the worn tool or they were affected by changes in parameters such as material and noise. Cus et al. [8] used of the Adaptive NeuroFuzzy Interference System (ANFIS) to predict the flank wear of the tool in end milling process. Neural network was used as a decision-making system to predict the condition of the tool, and the cutting forces were used as an indicator of the tool flank wears variation. However, due to the high computational power required for training the neural network, their method required parallel processing for the monitoring of the cutting process with high reliability. They suggested different decision making tools, such as fuzzy logic to be applied to obtain a smaller error of detection. Once the signal was collected from the sensor, various signal processing methods could be used to process the data. Wavelet transform [9], Hilbert-Huang Transform [10] and singular value decomposition method [11] were used as processing techniques for the above mentioned signal readings. Also, in multi sensor methods, a combination of the sensor data has been used and their sensitivity compared to each other [12]. The last step in developing a process monitoring technique is to find a decision making approach. During developing any machining process monitoring method, the amount of complexity of entire system should be considered carefully. Otherwise, the excessive complexity could compromise any future industrial application. Unlike the classical tool condition monitoring that focuses on the wear of cutting tool, the new emerging manufacturing process monitoring considers the quality of product as the purpose of monitoring the performance of manufacturing operation. The manufacturing originated tolerances in the components can cause unreliability in the performance of the systems [13]. Brecher et al. [14] used NC kernel data for surface roughness monitoring in milling operations. They used experimentation in order to obtain the data to be modeled with artificial neural networks for surface roughness average parameter predictions. The major drawback of their method was that their method required performing several

prior cutting tests in the corresponding machine tool, cutting tool, tool holder, and material combination. Huang [15] developed an intelligent neural-fuzzy model for an in-process surface roughness monitoring system in end milling operations. When the number of fuzzy sets were increased, there becomes a need for training data to be increased to fulfill all the possible IF-THEN rules. Marinescun et al. [16] developed an on-line automated monitoring method based on acoustic emission measurement for surface anomalies during milling of aerospace alloys. Their method reduced surface anomalies by a process monitoring solution that detects work piece anomalies associated with the cutting tooth of a milling cutter through signal analysis, which causes improvement of fault detection and the avoidance of surface anomalies or tool malfunctions. Although their method was efficient in detecting in controlling the tool to minimize process malfunctions in milling operations, they reported that in higher feeds and speeds the system was limited by processing speed of the operating system and amount of data processing needed to detect the malfunctions. Quintana et al. [17] developed a process solution that controls the surface process based on artificial neural network models by capturing the vibrations that occur during metal removal operation with the help of piezoelectric accelerometers. The technique calculates the current in-process roughness average and looks at the cutting parameters and applies the neural network developed using dynamic parameters measurement. Like similar methods extensive number of experiments required to be carried out to increase the software performance by training of the network on the basis of a large amount of experience. Bisu et al. [18] proposed a method based on the Vibration analysis that refers to advance analysis of vibrations for spectral envelope analysis based on Hilbert transform, and identifies mechanical defects to obtain a better response on the milling process quality. Surface response to excitation (SuRE) was developed for detecting structural health problems [19] and proved to be efficient using different implementation techniques [20]. Recently, SuRE was used for detecting the irregularities in dimension of work-piece [21]. The purpose of this study is to develop and efficient non-contact milling process monitoring based on the SuRE method. In this study, an indirect method for milling monitoring has developed based on measuring high-frequency surface guided waves. The purpose was to be able to develop a monitoring process that is capable of monitoring not only the surface roughness but also dimensional accuracy of the work-piece. To this end, laser-scanning vibrometer was used to monitor the behavior of high-frequency surface waves on an aluminum plate in the presence of milling operation. Due to the non-contact nature of this study, no interference occurs with the milling process. Results showed that the

Int J Adv Manuf Technol Fig. 1 Experimentation set-up for SuRE system

laser measurements followed by processing data were sensitive to the dimensional parameters of the milling operation.

The sum of the squared differences (SSD) for each scan point is calculated from:  Dm×n (2) S1×n = m

2 Method In this study, a structural health monitoring method was used for milling process monitoring. SuRE was used for this purpose and proper modifications in sensing, signal processing and data analysis was made to adapt this method feasible for the machining process monitoring. The SuRE method monitors the condition of structure actively by exciting the high-frequency surface waves. The surface of structure is excited using a piezoelectric element, and the response is monitored in another position on the plate. Usually the type of excitation is a sweep sine wave over a certain frequency range between 20–400 kHz. The Fast Fourier Transform (FFT) is then used to calculate the frequency spectrum of the transfer function. Studies have shown that this spectrum is consistent for any point on the structure as long as no change is occurred on the structure. As soon as condition changes such as compressive force or fatigue crack is introduced, the frequency spectrum changes to some extent. To quantify the change, the squared difference (SD) of frequency spectrum with respect to a reference one is used: Dm×n = ||Am×n − Rm×n ||2

(1)

Here, R and A are the reference and altered data matrices. The dimension of each data matrices is m rows by n columns. Each column includes the frequency spectrums of a certain scan point distributed over the frequency range.

In order to find the Normalized SSD (NSSD), the normalized matrix of differences, D matrix, is calculated from this formulation: D m×n =

1 Dm×n  d

(3)

d is the average value of matrix of differences that is calculated from: 1  d = Dm×n (4) mn m n The normalized differences matrix could be used to find the NSSD and in a similar way that the SSD matrix was found by:  S= D m×n (5) m

S is a matrix with the size of scanning grid that contains a normalized value for each scanning point. This normalized value quantitatively represents the amount of change in the spectrum for each scanning point. Depending on the configuration of scan, S could be a one-dimensional or two-dimensional array. SSD is an index that represents the change in the system spectrum. Any machining operation that removes the matter from work-piece, changes its properties such as the mass, damping and stiffness. These changes vary the pattern of wave propagation on the work-piece, and therefore cause the frequency spectrum to change. In other words, SSD values are directly affected by any changes in properties of the work-piece such as the mass, damping, and stiffness.

Int J Adv Manuf Technol Table 1 Four sizes of HSS flat head milling heads used for the experiments of this study

Milling head Size (in)

Fig. 2 Work-piece used in experimentations with scan points and milling areas specified; R1-R4 was for milling length study and R5 was used for milling depth and milling width study

In this study, the SSD and NSSD values were used for monitoring the process of three machining milling operations. Based on the SuRE method first, a reference scan was captured from a set of scan points. Each operation was performed through certain number of steps and at every step a laser scan captured the spectrums of all scan points. Based on the method described in this section, the SSD values and NSSD values were calculated. The behavior of SSD values and NSSD values were examined for milling length, depth and width changes and reliability of SuRE algorithm was evaluated for the purpose of machining process monitoring operations.

3 Experimental set up An aluminum plate was used in this experimentation with (4 in)(11 in)(0.4 in) dimensions. In order to excite surface waves on the beam an APC piezoelectric model D-.750”2 MHz-850 WFB was attached to the middle of the plate. To remove oil and any other possible contaminants, the surface of aluminum plate was cleaned with acetylene, ethanol,

Fig. 3 Real-time FFT of sine wave excitation (blue), spectrum created by capturing the peaks of FFT using peak hold (red)

First

Second

Third

Forth

1st 1/8 × 3/8

2nd 3/16 × 3/8

3rd 5/16 × 3/8

4th 3/8 × 3/8

and water. The bonding agent was LOCTITE Hysol Product E-30CL epoxy adhesive with a curing time of 24 hours. An applicator gun simultaneously mixed and dispensed the bonding agents from a dual-cylinder cartridge by passing the ingredients through a mixing nozzle. A clamp was used to position the plate in front of the laser head. Figure 1 shows the experimental set-up for SuRE system. The scan points were marked on the clamp to make sure that at every step of experimentation the same points are scanned. The piezoelectric element was used as an exciter to create sweep sine waves with a frequency range of 20– 400 kHz on the surface of the beam. A RIGOL DG1022 function/arbitrary waveform generator with peak-to-peak amplitude of 2.5 V generated the waves. In order to have higher signal to noise ratio in measurement points, the waveform amplified by five times through passing a TEGAM power amplifier model 2348. In this study, laser-scanning vibrometer measured the surface waves. The laser scanning Doppler vibrometer (LSDV) model Polytech PSV-400 remotely measured the surface vibrations from a grid of scan points on the aluminum plate. The scan grid included 50 scan points that were arranged in five rows and ten columns with the piezo element in the middle. Dimensions of scan area was (3.5 in) × (9.5 in). Scan grid in the both sides of piezoelectric included five columns of scan points. In total nineteen

Fig. 4 The coolant and cutting chips are covering the scan points in area close to milling operation

Int J Adv Manuf Technol

4 Results

Table 2 Milling parameters for the experiments of this study Milling step

First

Second

Third

Forth

Length (in) Depth (in) Width (in)

0.375 0.03 0.125

1.475 0.06 0.188

4.51 0.09 0.313

– – 0.375

milling operations were performed in the areas between the scan points. In Fig. 2, each operation region, located between two scan columns, is demonstrated. The laser vibrometer is composed of the scanning head, vibrometer controller, and junction box. Due to the limitation of sampling frequency of analogue to digital converter of laser junction box, an external data acquisition system was used. Data Translation simultaneous A/D convertor model DT9832-A was employed to capture the peak holds of the transfer function for the frequency input sweep sine wave. The maximum sampling rate of the device was set to 1 million samples per second. This allowed a maximum frequency of 400 kHz to be sampled. Since the SuRE algorithm requires the frequency domain data, the FFT of the input time data was used. The DT9832-A has a built in FFT package. The FFT size was set to 16384 and Hanning window function was used as smoothing window. Since the spectrum was captured by peak hold of FFT response while sweeping the excitation and the frequency band was spanning all the way from 20 to 400 kHz; in order to get the highest possible resolution, the maximum applicable FFT size available by software was chosen that was equal to 16384. Capturing the peak of Fast Fourier Transform of measured signal during a complete sweeping cycle created the frequency spectrum (Fig. 3). Fig. 5 a 1st Reference spectrum vs. 2nd reference spectrum, b 1st reference spectrum vs. 1st milling spectrum

4.1 Design of experiments The purpose of this study was to evaluate the potential of SuRE method for remote process monitoring of milling operation. Four different sizes of HSS milling heads shown in Table 1 were used. Experiments were divided into three categories for analysis of effect of length, depth, and width of the milling on laser scanner measurements. For each study, the milling operations were performed in a stepwise approach and at each step the specimen was scanned. For example, in the case of milling length analysis, the milling operation was performed at three steps using an ACRA milling machine. In this study, the experiments were not conducted in real time and at every step the work piece was removed from milling machine and placed in front of laser after completion of the operation. Figure 4 shows that the operation area is covered with the metal chip and coolant. The data for every scan point was analyzed in the frequency domain. Before any milling operation, the reference scan was performed on the intact plate. The reference scan was repeated in order to evaluate the consistency of the procedure. Then the milling operation was performed for study of effect of length, depth, and width of milling on the measurements. The milling for each case was performed in certain number of steps according to the Table 2. After every step of operation, a laser scan was captured from the same scan points on the plate. Figure 5a compares the measured frequency spectrums from the first and second reference scan. It is obvious that majority of peaks and valleys of both reference spectrum overlap. Figure 5b compares the frequency spectrum of reference scan with a frequency spectrum that

Int J Adv Manuf Technol

Fig. 6 Contour map of SSD values on the scan area for the (a) 1st milling, (b) 2nd milling, (c) 3rd milling. The color-map is created from scan points of R1-R4

was captured after the first step of milling operation was performed. The change in the frequency spectrum could be observed in three ways. Some peaks dropped, some peaks raised, and some drifted with respect to the original position in the reference scan. According to the SuRE, the SSDs of each spectrum from reference spectrum was calculated for all scan point at each step of milling operation. Since the SSD values were calculated for all scan points, contour map plot of those values at every step could give a visual understanding of the procedure. 4.2 Study of milling length The study of the length of milling was performed on the left side of the plate within areas R1 to R4 (Fig. 2) for each of the four milling sizes. Contour map of the SSD values over the surface of the plate is shown in Fig. 6. Considering that the area R1 was located on the left side of the plate, and the operation was performed on this area, the red spot of the contour map successfully identifies of the operation on the plate. At every step, the maximum values of SSDs occurred close to the location of operation. A secondary weaker red spot is also is appeared away from the first one. Appearance of the secondary weak spot is due to the fact that the plate size in this study has small dimensions and because of the reflections of waves between the machined area and the exciting piezo, minor secondary spot can occur, which in comparison to the original read spot is negligible. Our study revealed that the SSD values in this area show similar pattern to those ones close to the location of operation and therefore could be used for the purpose of monitoring the operation. The importance of this phenomenon is that in machining operations that involve heavy formation of chip and usage of lots of coolant, laser could capture measurements from this area.

Fig. 7 Maximum SSD values on the scan area for the study of milling length (a) 1st milling step, (b) 2nd milling step, (c) 3rd milling step

Another interesting phenomenon that was observed during this study was the correlation between the SSD values and the length of the milling on the aluminum beam. This correlation is demonstrated in the Fig. 7 where the SSD values increased by increasing the length of milling. This behavior could be used for monitoring the state of the progress of milling operation. 4.3 Study of milling depth The operation for study of the depth of the operation was performed in the area R5, which is located on the right side of the plate within 3rd and 4th columns of scan grid. In Fig. 8, the three-dimensional contour map of the SSD values correctly identifies the location of operation on the plate. This property of SuRE method that is useful in damage localization in SHM applications could be used to monitor the correct location of each operation in automated machining monitoring applications. Dimensional accuracy is another important property in machining. Figure 9 includes SSD values for the three consequent steps of the millings where the depth of milling has increased in each

Fig. 8 Three-dimensional demonstration of SSD values on the scan area

Int J Adv Manuf Technol

5 Conclusions

Fig. 9 SSD values versus depth of milling increasing in 3 steps

step according to Table 2. As the milling depth increases, the maximum value of the SSDs has also increases. 4.4 Study of milling width The study of milling width was performed in four steps for all milling head sizes of this study in the R5 region between 3rd and 4th scan columns (Fig. 2). Although the change in the value of SSD in the study of milling width had similar value to those of depth and length, due to the larger SSD values in this level of study, the relative value of changes in the SSD seems to be small in Fig. 10. Therefore, among the three basic parameters that were studied, SSD values showed the least sensitivity to the milling depth. Especially, in scan points close to edges of the plate, the behavior of SSDs was not consistent. But if the position of the measurement point is chosen carefully in interior area of the plate, a more reasonable behavior could be observed.

The purpose of this study was to evaluate how effectively SuRE could be used to evaluate the quality of the completed machining process in milling operations. A piezoelectric element excited surface waves on the surface of the aluminum workpiece. Laser scanning vibrometer was used as a non-contact sensor to measure the vibrations at a grid of scan points. The spectrums of the vibration data were calculated by using the FFT. The variation of the spectrums was evaluated by calculating the sum of squares of differences between the reference and spectrums after each machining operation. The SSD values quickly increased when the size of the slot changed. Our study confirmed that the dimensions of a completed machining operation may be inspected by using the SuRE method. The SSD values showed the sensitivity to three basic parameters of the slot; namely length, depth, and width. The sensitivity was very good against length and depth. It was minimum to width change. The study indicates that instead of monitoring the machining operation with sensors during the machining operations, the performance of the completed operation may be evaluated with piezoelectric sensors attached to the workpiece in milling operations. This would reduce the cost of inspection and initial investment minimum. In addition, the same sensors may be used to evaluate the performance of multiple operations. Acknowledgments The authors greatly acknowledge the Florida International University Graduate School for providing support for this research in the form of Dissertation Year Fellowship (DYF). Also, authors gratefully acknowledge Army Research Office for funding the shared facilities used in this research at Florida International University (Grant Number 58940-RT-REP). The test pieces were prepared with Richard-Todd Zicarelli’s help at the Engineering Manufacturing Center (EMC) of the Florida International University. His help is sincerely appreciated.

References

Fig. 10 SSD values versus width of milling increasing in 4 steps

1. Kalvoda T, Hwang YR (2010) A cutter tool monitoring in machining process using Hilbert–Huang transform. Int J Mach Tools Manuf 50(5):495–501 2. Lu MC, Wan BS (2013) Study of high-frequency sound signals for tool wear monitoring in micromilling. Int J Adv Manuf Technol 66(9–12):1785–1792 3. Nikranjbar A, Atai AA (2013) Online model-based milling process condition monitoring. Int J Mechatronics Manuf Syst 6(2):195–212 4. Hase A, Wada M, Koga T, Mishina H (2014) The relationship between acoustic emission signals and cutting phenomena in turning process. Int J Adv Manuf Technol 70(5–8):947–955 5. Dinakaran D, Sampathkumar S, Sivashanmugam N (2009) An experimental investigation on monitoring of crater wear in turning using ultrasonic technique. Int J Mach Tools Manuf 49(15):1234– 1237

Int J Adv Manuf Technol 6. Lamraoui M, Thomas M, El Badaoui M, Girardin F (2014) Indicators for monitoring chatter in milling based on instantaneous angular speeds. Mech Syst Signal Process 44(1):72–85 7. Hsieh WH, Lu MC, Chiou SJ (2012) Application of backpropagation neural network for spindle vibration-based tool wear monitoring in micro-milling. Int J Adv Manuf Technol 61(1– 4):53–61 ˇ s F, Zuperl ˇ 8. Cuˇ U (2011) Real-time cutting tool condition monitoring in milling. Strojniˇskivestnik-J Mech Eng 57(2):142–150 9. Tangjitsitcharoen S, Saksri T, Ratanakuakangwan S (2013) Advance in chatter detection in ball end milling process by utilizing wavelet transform. J Intell Manuf 26(3):485–499 10. Cao H, Lei Y, He Z (2013) Chatter identification in end milling process using wavelet packets and Hilbert–Huang transform. Int J Mach Tools Manuf 69:11–19 11. Kalvoda T, Hwang YR (2010) Analysis of signals for monitoring of nonlinear and non-stationary machining processes. Sens Actuators A: Phys 161(1):39–45 12. El Ouafi A, Barka N (2014) An ANN based multi-sensor integration approach for in-process monitoring of product quality in turning operations. J Autom Control Eng 2(3):289–293 13. Samadani M, Behbahani S, Nataraj C (2013) A reliability-based manufacturing process planning method for the components of a complex mechatronic system. Appl Math Model 37(24):9829– 9845

14. Brecher C, Quintana G, Rudolf T, Ciurana J (2011) Use of NC kernel data for surface roughness monitoring in milling operations. Int J Adv Manuf Technol 53(9–12):953–962 15. Huang PB (2014) An intelligent neural-fuzzy model for an inprocess surface roughness monitoring system in end milling operations. J Intell Manuf 1–12. doi:10.1007/s10845-014-0907-6 16. Marinescu I, Axinte DA (2011) An automated monitoring solution for avoiding an increased number of surface anomalies during milling of aerospace alloys. Int J Mach Tools Manuf 51(4):349– 357 17. Quintana G, Garcia-Romeu ML, Ciurana J (2011) Surface roughness monitoring application based on artificial neural networks for ball-end milling operations. J Intell Manuf 22(4):607–617 18. Bisu CF, Zapciu M, Cahuc O, G´erard A, Anica M (2012) Envelope dynamic analysis: a new approach for milling process monitoring. Int J Adv Manuf Technol 62(5–8):471–486 19. Fekrmandi H, Rojas J, Campbell J, Tansel IN, Kaya B, Taskin S (2014) Inspection of the integrity of a multi-bolt robotic arm using a scanning laser vibrometer and implementing the Surface Response to Excitation Method (SuRE). Int J Progn Health Manag 5.1:1–10 20. Fekrmandi H, Rojas J, Tansel IN, Yapici A, Uragun B (2015) Investigation of the computational efficiency and validity of the surface response to excitation method. Measurement 62:33–34 21. Tansel IN, Yapici A (2013) Part based process performance monitoring (PbPPM). J Manuf Process 15(3):329–337