2011 International Conference on Electronic & Mechanical Engineering and Information Technology Design and Development of Rolling Bearing Vibration Signal Analysis System Wensheng Su Wuxi Branch, Jiangsu Province Special Equipment Safety Supervision Inspection Institute Wuxi, China e-mail: swsing@163.com Fengtao Wang Research Institute of Vibration Dalian University of Technology Dalian, China e-mail: wangft@dlut.edu.cn Abstract—Taking the respective advantages of LabView and Matlab, an on-line vibration signal analysis system of rolling bearing was designed and developed. First, system scheme design was introduced. Then each module of the system was discussed. Finally, system application examples were given, which verified the convenience and effectiveness of this system. Keywords-rolling bearing; system development; data acquisition; Lab View;Matlab I. INTRODUCTION Open system V Acquire signal V Input signal T De-noising Demodulation Feature parameters extraction Exit system Figure 1. Overall structure of the system 1. De-noising 1.1 EMD de-noising 1.2DWT-HMT 1.3DTCWT-HMT1 1.4DTCWT-HMT2 1.5 DTCWT soft-threshold 1.6 DTCWT hard-threshold 1.7 Traditional soft-threshold 1.8 Traditional hard-threshold The vibration signal processing methods of rolling bearing generally include de-noising, demodulation and feature extraction [1, 2]. Typically computer program is written with Matlab which requires repeated modifying in application, so specific implementation is inconvenient. For this reason, the authors developed an on-line vibration signal analysis system of rolling bearing based on LabView and Matlab hybrid programming. LabView and Matlab are widely used in engineering field. They have respective advantages at different aspect. LabView is regard as the most outstanding representative of graphical programming language. Matlab is known as the most powerful computing language. In developing this system, full using of their advantages can achieve high efficiency programming. Signal acquisition and graphical user interface were developed with LabView, while numerical calculation and signal processing were developed with Matlab, this hybrid method was adopted. LabView 8.6 and Matlab 7.1 version, respectively, was used to develop this system. II. SYSTEM SCHEME DESIGN T 0 Original vibration signals 0.1 System signal acquisition 0.2 Off-line signal <-V 2 Demodulation Determination of band-pass filter parameters 2.1 Spectral kurtosis method 2.2 Optimal Morlet wavelet filter Envelope analysis 2.3 Hilbert demodulation 2.4 Square envelope demodulation Enhancement method 2.5 Autocorrelation enhancement algorithm The overall structure of the system is shown in Fig. 1. First, open the system; and then acquire signal. It can acquire test-bed signal or directly into an existing signal which was obtained from other way. Third, signal analysis, including de-noising, demodulation and feature parameter extraction. Finally, exit the system. Detailed analysis process of the system is shown in Fig.2; the arrows represent the flow of data. Although some demodulation, de-noising and feature parameters extraction methods are listed in this figure, it should be noted that the system is extendible. We all know these methods are so much that any system can't contain all methods, the same to this system. The system uses a modular programming style. In future research work, some analysis methods can be added to enrich the contents of the system. 3 Feature parameters extraction 3.1 RMS value 3.2 Kurtosis value 3.3 approximate entropy 3.4 sample entropy 3.5 wavelet packet sample entropy I <-V 4 Frequency analysis (FFT) Figure 2. Analysis process of the system The analysis process of the system includes four parts: 978-l-61284-088-8/ll/$26.00 ©2011 IEEE 2679 12-14 August, 2011 The one is original vibration signal acquisition. It comes from two aspects: one is data acquisition system developed on the rotating machinery vibration analysis and fault diagnosis system which is purchased by our research institute; the second is rolling bearing vibration signal data obtained from other sources including the CWRU signal and a petrochemical enterprise signal [3]. The second is de-noising, including EMD de-noising [4], wavelet packet de-noising [5], wavelet domain Hidden Markov Tree Model de-noising [6], dual-tree complex wavelet domain Hidden Markov Tree Model de-noising [7] and wavelet threshold de-noising [8]. For the de-noised signal, it can be processed with spectrum analysis. The third is demodulation. In this system, there are two ways of band-pass filter parameter selection for resonance demodulation. One is spectral kurtosis [4]; the second is optimal Morlet filter method [3]. For the band-pass filtered signal, you can use Hilbert demodulation or square envelope demodulation. The demodulated signal can be processed by spectrum analysis and get the envelope spectrum. The fourth is feature parameters extraction, including RMS value, kurtosis value, approximate entropy [9], sample entropy [10], wavelet packet sample entropy [11] and so on. It should be noted that rolling bearing vibration signal usually is processed through one or a few parts of the following three processes, namely, de-noising, demodulation and feature extraction, then you can achieve fault diagnosis and determine fault type. III. SYSTEM IM PLEMENTATION not have database access function. There are several ways to solve this problem [12]. Figure 3. Test platform A. Data acquisition module Figure 4. Compact RIO module NI Compact RIO is a reconfigurable I/O data acquisition and industrial control systems, with the advantages of high performance and customization. It contains a reconfigurable FPGA chips and a real-time controller, which is compatible with reliable independent embedded or distributed applications system. It also includes hot-swappable industrial I/O module, built-in signal condition circuit, can be directly connected with sensors, achieve a variety of signal acquisition. Since Compact RIO has an open framework, users can easily access to the underlying hardware devices. Moreover, Compact RIO embedded system can use the LabView for rapid development. Rotating machinery vibration analysis and fault diagnosis test platform is shown in Fig.3. Based on the test platform, a rolling bearing data acquisition system based on NI Compact RIO is developed, which consists of an integrated controller NI cRIO-9073 and I/O module NI cRIO-9205. Vibration signal sampled by the cRIO-9205 is processed with sample computation by 2M gate FPGA and then transmitted to controller cRIO-9073 for real-time storage through FIFO channel. PC can read the signal stored in real-time controller cRIO-9073. B. Data management module It inevitably need database access when developing application software with LabView. But LabView itself does 1) Call dynamic link library (DLL) for database access. First, program DLL using other tools such as Visual C++. Then utilize LabView function nodes CLFN (Calling Library Function Node) to call this DLL to access the database. This method requires complex underlying programming and it is not realistic for non-professional programmer. 2) Utilize ActiveX function of LabView to call Microsoft ADO object and realize database access with SQL language. This method requires the user deep understanding Microsoft ADO and SQL language. 3) Visit database through a third party developed free toolkit LabSQL. LabSQL uses Microsoft ADO and SQL language to complete database access. It packages complicated underlying ADO and SQL operation into a series of LabSQL Vis, so it is easy to use. 4) Use NTs database interface package LabView SQL Toolkit for database access. The toolkit integrates a series of advanced modules which encapsulate the majority of database operations and a number of advanced database access function. It is easy to understand and simple to operate. The user need not learn SQL syntax. For the collected signal from test-bed, LabView SQL Toolkit is used as the database access method. Using the toolkit, users can create automated test systems that manage complex tasks, store test data and summarize test results. 2680 For external signals, there are many different types of data formats, including text format (.Txt), binary format (.Dat), Matlab data format (.Mat) and so on. Therefore, Lab View need to adopt different reading procedures for different formats data. File reading module adopts selection structure. Reading a Matlab data file uses the Matlab Script of Lab View; Reading a text file or binary file uses Read Lvm nodes of Lab View. Storing analysis results data use Write Lvm node of Lab View. C. Signal analysis module Signal processing includes three parts: de-noising, demodulation and feature parameter extraction. They are respectively made as a subroutine which is called by the main program for signal analysis. The subroutine module can be extended to enrich the system analysis. Lab View is used for program interface development. For complex computer programs, Matlab code need be called for analysis. Lab View and Matlab have their own advantages. Combining them can improve development efficiency. Currently, the hybrid programming methods with Lab View and Matlab mainly include the followings. 1) Based on Matlab Script node. Lab View and Matlab hybrid programming is based on their support of ActiveX automation technology. Matlab Script node actually communicates with the Matlab Server through ActiveX Control. Calling Matlab function is actually that Matlab acts as an ActiveX server, create ActiveX channel, send the function to Matlab through ActiveX channel and implement it in Matlab background. Matlab Script can be correctly debugged in Matlab environment and then imported into Matlab Script node. 2) Based on ActiveX automation ActiveX automation is the ability that a program controls another program by using its methods and properties. Lab View connects to Matlab automation server through identification number (Refhum) and invokes the methods and properties provided by Matlab ActiveX interface through automation node function. Using ActiveX automation method has greater control ability to Matlab. ActiveX automation is suitable for larger application program development. 3) Based on COM technology The core of COM technology is the binary interface specification which is independent of programming languages and operating systems. The Matlab Builder tool can convert M file into independent COM component object, and automatically register in the operating system. The generated COM component object can be used in any application program that supports COM object, such as Visual C++, Visual Basic, Lab View and so on. Lab View control to COM component is also achieved by the automation node function of ActiveX sub-module. This system adopted the COM technology to realize the development of complex program modules. IV. SYSTEM APPLICATION Click on 'signal import', select a signal, then the result is shown in Fig.6 which includes time domain and frequency domain waveforms. Figure 5. Main interface of analysis system Figure 6. Interface of signal input Figure 7. Menu demo interface Figure 8. Denoise method based on EMD filter The main interface of the system is shown in Fig.5, which provides the background information for the system. System takes a menu form. Fig.7 is used as an example for signal analysis. Click on 'De-noising' -*• 'EMD de-noising' -*'Threshold-based processing', implement the 2681 de-noising method based on EMD filter, the result is shown in Fig.8. According to the cross-correlation coefficient and kurtosis, we select the imf (Intrinsic Mode Function) component 1 and 2, click on 'de-noise' and get the denoised signal, click on 'save', then the denoised signal will be saved. Similarly, you can get the results based on Dual-tree complex wavelet transform (DTCWT) threshold de-noising (Fig.9) method and optimal Morlet wavelet band-pass filter (Fig. 10). Fig.8 to 10 are the denoised or filtered results with different methods. It can be found that the impact time point can be easier determined based on DTCWT method and the rolling bearing fault can be diagnosed in time-domain waveform. Vibration signal parameters calculation is done in Fig. 10. By selecting calculating contents, the corresponding results can be obtained. All of the above verify the system is convenient and effective. Figure 11. Vibration signal feature extraction REFERENCES Figure 9. Denoise method based on DTCWT Figure 10. Bandpass filtering method based on optimal Morlet wavelet ACKNOWLEDGMENT The work is supported by National Natural Science Foundation of China (No.50805014) and the Key Project of Chinese Ministry of Education (No.109047). [I] QIU H, LEE J, LIN J, YU G. Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics [J]. Journal of Sound and Vibration, 2006, 289: 1066-1090. [2] Chen Jin. Application of signal processing in machinery fault diagnosis [J]. Journal of Vibration and Shock, 1999,18(3): 91-93. [3] Wensheng Su, Fengtao Wang, Hong Zhu, Zhixin Zhang, Zhenggang Guo. Rolling Element Bearing Faults Diagnosis Based on Optimal Morlet Wavelet Filter and Autocorrelation Enhancement [J]. Mechanical Systems and Signal Processing, 2010, 24(5) : 1458-1472. [4] Su Wensheng, Wang Fengtao, Zhang Zhixin,Guo Zhenggang, Li Hongkun. Application of EMD De-noising and Spectral Kurtosis in Fault Early Diagnosis of Rolling Element Bearing [J]. Journal of Vibration and Shock, 2010,29(3): 18-21. [5] Wang Fengtao, Ma Xiaojiang, Zou Yankun, Zhang Zhixin. Local power feature extraction method of frequency bands based on wavelet packet decomposition [J]. Transaction of the Chinese society for agriculture machinery, 2004,35(5): 177-180. [6] Crouse M S, Nowak R D, Baraniuk R G. Wavelet-based statistical signal processing using hidden Markov models [J]. IEEE Transactions on Signal Processing, 1998, 46(4):886-902. [7] Su Wen-sheng, Wang Feng-tao, Zhu Hong, Zhang Zhixin, Li Hongkun, Guo Zhenggang. Denoising method based on hidden Markov tree model in dual tree complex wavelet domain and its application in mechanical fault diagnosis [J]. Journal of Vibration and Shock (Accepted). [8] Donoho D L. De-noising by soft-thresholding [J] . IEEE Transaction on Information Theory, 1995, 41(3):613—627. [9] Pincus S M. Approximate entropy as a measurement of system complexity [J]. Proc Natl Acad Sci USA, 1991, 88 (6):2297-2301. [10] Richman J S, Moorman R J. Physiological time-series analysis using approximate entropy and sample entropy [J]. Am J Physiol Heart Circ Physiol, 2000, 278: 2039-2049. [II] Su Wensheng, Wang Fengtao, Zhu Hong, Guo Zhengang, Zhang Zhixin, Zhang Hongyin. Study on the feature extraction of rolling element bearing fault based on wavelet packet sample entropy [J]. Journal of Vibration,Measurement & Diagnosis (Accepted). [12] Bi Hu, Lu Fangcheng, Li Yanqing, Li Heming. Several methods of accessing database in Lab View. Microcomputer Information [J], 2006,22(1):131-134. [13] Yang Leping, Li Haitao, Zhao Yong, et al. Lab VIEW advanced programming [M]. Beijing: Tsinghua University Press, 2003. 2682
