FPGA based Signal Prefiltering System for Vibration ... - CyberLeninka

Available online at www.sciencedirect.com

Procedia Technology 4 (2012) 442 – 448

C3IT-2012

FPGA based Signal Prefiltering System for Vibration Analysis of Induction Motor Failure Detection Saikat Kumar Shomea, Uma Dattab, SRK Vadalic b

a School of Mechatronics, CSIR-CMERI, Durgapur, Pin: 713209, India Electronics & Instrumentation Division, CSIR-CMERI, Durgapur, Pin: 713209, India c Robotics & Automation Division, CSIR-CMERI, Durgapur, Pin: 713209, India

Abstract Timely diagnosis of faults in industrial equipments is a concern to guarantee the overall production process efficiency. For instance, vibration analysis has been one of the few important approaches for motor failure detection. However, vibration sensory signals are often corrupted with random noise, processing of which leads to inaccurate results. Thus, there is a necessity of low cost instrumentation with proper filtering capabilities for online vibration measurement which can be permanently fixed to the machine under test (MUT) for continuous monitoring and reliable diagnosis. The present work compares three signal averaging based filtering techniques for the purpose of analysis. The filters have been implemented in Field Programmable Gate Arrays (FPGA) which are characterized by reduced power consumption and high operational speed for real time applications. To test the functionality of the proposed algorithms, case study of an accelerometer data attached to an Induction motor has been taken up and the results have been analyzed.

© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of C3IT Keywords: FPGA; Vibration Analysis; MUT; EWMA

1. Introduction Machine condition monitoring, especially detection of failures at an early stage is one of the major concerns of industry. Several techniques have been developed for detection of failures in induction motors, such as, those based on electrical variables [1], thermography, ultrasound technologies and oil analysis, which are mainly carried out offline, requiring periodic halts in the production flow. However, analysis through vibration has been proved to be very effective because each fault within a machine produces a vibration with distinct characteristics [2]. Vibration signals can be analysed in either time domain or frequency domain. Although time domain approach provide meaningful insight about the physical nature of vibrations, the presence of multi-tone signals render such analysis practically infeasible. On the contrary, analysis in the frequency domain with reference to both amplitude and phase spectra, is more useful because it associates change in dynamics with almost 100% failures in rotating machines [3]. However, the presence of noise signals corrupt the sensory output, which in turn modify the original frequencies. 2212-0173 © 2012 Published by Elsevier Ltd. doi:10.1016/j.protcy.2012.05.070

Saikat Kumar Shome et al. / Procedia Technology 4 (2012) 442 – 448

Spectral analysis of such corrupt signals may reveal the presence of misleading frequencies and hence lead to faulty diagnosis [4,5]. Frequently used instruments for vibration analysis like the Dynamic Signal Analyser (DSA) are general purpose, costly, time consuming, power inefficient and cannot be permanently mounted to the machine for real time analysis. Quite a few researches have been undertaken to introduce new characteristics and advanced features to the previous designs.[6] propose a vibration analyzer with multichannel data acquisition system, nevertheless without filtering and requiring a PC for spectral analysis. Bilks et al [7] proposed a spectral analyzer in Labview environment, while [8] emphasizes in the reducing processing time and operational cost through high speed hardware devices. A Fast Fourier Transform (FFT) based analyzer with adjustable parameter using a PC is described in [9]. These researches focus on various failure detection techniques simulating the spectra of each failure using a signal generator. In [10], an entirely different paradigm considering FPGA based implementation for online spectral computation without requiring any PC is seen. As may be observed from the cited references, the acquired sensor data have been ported to the vibration analyzers, mostly offline. The captured sensory data being unavoidably corrupt with noise results in an unreliable output with reduced information content and usability. Researches also suggest a reconfigurable, low cost embedded SoC based standalone instrument for effective vibration based condition monitoring. In the present work, we propose a cut-off frequency independent, computationally inexpensive, averaging based digital filter with very little hardware resource utilization, in FPGA. FPGA platform has been chosen for its high operating speed suitable for real time analysis with highly reconfigurable property that allows further processing to be implemented in the same VLSI chip. The present paper is organized as follows: Section 2 provides a Theoretical Background and System Model for signal pre-processing. Section 3 provides results from FPGA implementation of the proposed module while Section 4 concludes the paper. 2. Theoretical Background Many techniques have been developed to detect motor failures as mentioned earlier, with Motor Current Signature Analysis (MCSA) and vibration analysis being the most preferred ones. MCSA procedure analyses the motor stator current to relate the measured quantity with possible faults based on either start up transient analysis or steady state [11,12]. Although MCSA technique has a good performance, it demands dedicated instrumentation coupled with specific algorithms for automatic detection of failures. On the contrary, vibration based analysis can be used as a generalized method for fault detection and identification because almost 100% of motor failures are reflected as vibration frequency changes. Research studies aimed at relating particular motor failures with their vibration characteristics are present in literature [13]. In most applications as in [14], FFT has been employed as the mathematical tool for spectrum computations. Faulty models can effectively be identified based on vibration. During normal operation, the vibration spectra is characterized by peak present at shaft rotation frequency, Fw, followed by a number of sub-harmonics. Each failure in the motor present either at the rotor or stator, of mechanical or electrical origin, causes a specific alteration of the spectrum as compared to the healthy one, as seen in Table 1. The presence of a particular tone at a given frequency indicates the existence of a specific fault characterized by that frequency while the magnitude of the peak reflects the severity of the damage. The System Model is considered with a Broadband Gaussian noise ሺሻwith zero mean and variance ɐଶ so that ሾ ሺሻሿ ൌ Ͳ (1) ሾ ଶ ሺሻሿ ൌ ɐଶ ሺʹሻ

443

444


Assuming the signal ሺሻ and noise ሺሻ are uncorrelated, the recovered signal ሺሻ can be expressed as ሺሻ ൌ ሺሻ ൅ ሺሻ(3) In case of simple moving average filter, equal weightage is given to the present input sample and the previous averaged output sample and is governed by the following equation: ‫ݕ‬ሺ݅ሻ ൌ ͲǤͷ‫ݔ‬ሺ݅ሻ ൅ ͲǤͷ‫ݕ‬ሺ݅ െ ͳሻ(4) where, ‫ݕ‬ሺ݅ሻ is the filtered output at ith sample. This technique places equal emphasis on all data points [15]. Thus a value in the past will have the same influence as the current measurement, when calculating the filtered signal. This problem can however be reduced by generating the filtered value in a slightly different manner. Table 1. Frequency of the vibration and possible related failures

Frequency

Related Failure

1 x RPM

Unbalance, misalignment of gear or pulley, resonance, electrical problems

2 x RPM

Mechanical offset, misalignment, Alternative forces, resonance

3 x RPM

Misalignment, axial mechanical breath

< 1 x RPM

Slip, oil whirl

Supply Freq

Electrical problems

RPM Harmonics Damaged gears, aerodynamic forces, hydraulic forces, mechanical offset, alternative forces

Figure-1: Exponential Weighing Factors in EWMA

In dynamic systems, however, the most current values tend to reflect better the state of the process. A filter that places more emphasis on the most recent data would therefore be more useful. The Exponentially Weighted Moving Average (EWMA) filter [16] places more importance to more recent data by discounting older data in an exponential manner, as in Fig.1. Thus, in calculating the filtered value, more emphasis is given to more recent measurements as compared to Simple Moving Average. Hence, the filter output can be written as: ‫ݕ‬ሺ݅ሻ ൌ ሺͳ െ ߙሻ‫ݔ‬ሺ݅ሻ ൅ ߙ‫ݕ‬ሺ݅ െ ͳሻ(5) Equation (5) represents the Exponentially Weighted Moving Average Filter, where ‫ݕ‬ሺ݅ሻ is the current average output sample, ‫ݕ‬ሺ݅ െ ͳሻ is the previous averaged output sample, ‫ݔ‬ሺ݅ሻ is the current sample. ߙ is the weighting factor constant, known as degree of filtering, which is expressed by ௣ where, p is the number of samples taken for consideration. ߙ ൌ ௣ାଵ

3. Experimental Setup & Results of FPGA Implementation The proposed work is constituted by a dual axis piezoelectric accelerometer of 0.1g, one Data Acquisition System (DAS) and FPGA based hardware signal processing unit as generalised in Fig. 2. Experiments were conducted on unbalanced 3-ĭ, 1 HP Induction Motor rotating at 2700 rpm using Simple Moving Averaging and Exponential Moving Averaging Technique (Fig. 3-4) of varying weighing factors with tools listed in Table 2. The EWMA filter which gives more weightage to recent observations has a better performance as compared to SMA.


Figure-2: Position P of propo osed module in entire e architecturre

Figure-3: Sysstem Floor Plan of Exponentially Weighted W Movinng

Average Filterr Table 2 : Design Tools for Syynthesis

Platfoorm HDL Languag FPGA A Family Packaage Targeet Device Speedd Grade

Xilinxx ISE Design Suuite 13.1 Veriloog HDL Spartaan 3E VQ1000 XC3S S250E -5

Figure -4: RTL Schematic View of o the Top Level Block B

Figure-5: Test Bench (Timing Diagram) Resullts of Simple Mooving Averagingg Filter

Figure-6: Test Bench (Timing Diagram) Resullts of Exponential Moving Averrager with p = 100

445

446


Figure-7: Acceleerometer data attacched to an Inductioon Motor

Figure-8: Simpple Moving Averag ge Output

Figure-9: Expoonential Moving Average A Output : p= =4

Figure-10: Expponential Moving Average A Output : p=10 p


mmary of Spartan 3E 3 Figure-11: Device Utilization Sum

Figure-12: HDL H Synthesis Report R

ISim timing diagrams and Matlab M simulaation results aree shown in Figg. 5-10. The Deevice Utilizatioon Summary aloong with partiaal HDL Synthesis report aree reported in Fig. F 11-12. Thee minimum tim me period is 2.1002 ns with a maximum m frequeency of 475.7337 MHz using a total memoryy of 229 kB. 4. Conclusioon Condition monitoring using g vibration anaalysis is a techhnique of grow wing importancce. It is felt thaat available speectral vibration n analysis schhemes are not highly reliable in a noisy scenario. s In thhe present workk, a case study of acceleromeeter data attachhed to an Inducction motor has been taken up u for analysis. It was obserrved that speccial purpose appplication specific pre-proceessing modulees would assist in better diagn nosis, especiallly in FFT basedd vibration anaalysis. Three fiiltering modulees have been suuggested and teested with raw sensor data. The T filters weree implemented in a single, low w cost FPGA chip c with recon nfigurable opeen architecture and are obserrved to be of low complexityy, while introduucing minimal latency. An advantage off the proposedd module is thhe flexibility to t selectively chhoose any of the t three sugggested filtering schemes whicch effectively requires only a few additionnal indexing operations. o Foor this reasonn, it may be concluded thaat such simplle preprocessingg modules maay be used as a prefiltering module m for predictive mainttenance tools in i industrial appplications baseed on SoC apprroach. References 1. 2. 3. 4. 5. 6.

M E H Benbouzid, M Vieira, C Theys, “Induction “ Motorss Fault Detection and a Localization Using U Stator Currennt Advaanced Signal Proceessing Techiques””, IEEE Trans.on Power P Electronics,, vol 14, No 5, pp 14-22, 1999. J.S. Mitchell, M Introduction to machinery analysis and monitoring, 2nd ed., Peennwell Publishingg Company, 1993. S.L. Marple, Digital Sp pectral analysis, with w applications. Englewood E Cliff, NJ: N Prentice Hall, 1987. S. Janngi and Y. Jain, “E Embedding spectraal analysis in equippment,” IEEE Speectrum, pp. 4043, 1991. 1 G. K Sing, S A K S Ah hmed, “Vibration signal analysis usiing wavelet transfoorm for isolation and a identification of o electrrical faults in indu uction machine”, Elsevier E Electric Poower System Research, pp. 119-136, 2003. S. A.. Ansari and R. Baaig, “A PC-based vibration v analyzerr for condition monnitoring of processs machinery,” IEE EE Trans. Instrum. Meas., vol. 47, no. 2, pp.. 378–383, Apr. 19998.

447

448

Saikat Kumar Shome et al. / Procedia Technology 4 (2012) 442 – 448 7. 8. 9. 10. 11. 12. 13. 14.

15. 16.

P. Bilski and W. Winiecki, “Virtual spectrum analyzer based on data acquisition card,” IEEE Trans. Instrum. Meas., vol. 51, no. 1, pp. 82–87,Feb. 2002. P. Bilski and W. Winiecki, “A low-cost real-time virtual spectrum analyzer,” IEEE Trans. Instrum. Meas., vol. 56, no. 6, pp. 2169–2174,Dec. 2007. G. Betta, M. D’Apuzzo, C. Ligouri, and A. Pietrosanto, “An intelligent FFT-analyzer,” IEEE Trans. Instrum. Meas., vol. 47, no. 5, pp. 1173–1179, Oct. 1998. J. A. Vite-Frias, R. J. Romero-Troncoso, and A. Ordaz-Moreno, “VHDL core for 1024-point radix-4 FFT computation,” in Proc. Int. Conf. ReCon-Fig, Puebla, Mexico, 2005, p. 24. P. J. Tavner and Penman, Condition Monitoring of Electrical Machines.Chichester, U.K.: Wiley, 1987. M. El Hachemi Benbouzid, “A review of induction motors signature analysis as medium for faults detection,” IEEE Trans. Ind. Electron., vol. 47, no. 5, pp. 984–992, Oct. 2000. R. A. Collacot, Vibration Monitoring and Diagnosis. London, U.K.:George Godwin Ltd., 1979. G. Betta, C. Ligouri, A. Paolillo, and A. Pietrosanto, “A DSP-based FFT analyzer for the fault diagnosis of rotating machine based on vibration analysis,” IEEE Trans. Instrum. Meas., vol. 51, no. 6, pp. 1316–1322, Dec. 2002. Bonnie Baker, “The power of moving-average digital filters”, Magazine, Dec 16, 2005, pp 30. S H Hwang, H Chen and C Chang, “An exponentially weighted moving average method for identification and monitoring of stochastic systems Industrial & Engineering Chemistry Research, vol. 47(21), 2008, pp. 82398249.