Fault Feature Extraction Technique Based On Patch Near-Field Acoustical Holography

At some situations, the fault diagnosis technique based on vibration signals has its restrictions. As the result of vibration emission in air, machine sound signal carries affluent information about the working condition of machine. It processes the advantage of non-contacting measurement. It can partly take place of vibration signal based fault diagnosis and to be used for mechanical fault diagnosis. The fault diagnosis technique based on sound signal is named by acoustical diagnosis technique. To implement this technique, it should take into account of the characters of machines'noise and make deep research into the technique of acoustical feature extraction. So, the extracted feature can describe machineries'condition well. Traditional acoustical feature extraction method can only describe the law for fault features changing on time, frequency, but they can not reveal the changing information of fault features according to the locations of sound sources.For more efficiently perform fault diagnosis for machineries using sound signal, a fault feature extraction technique based on acoustical holography is presented in this work. This technique use an array consisted of a small number of microphones to acquire sound pressure. Traditional wave superposition method can not reconstruct the sound field precisely because of the absence of equivalent source configuration information. A method named Joint Patch Near-field Holography (JPNAH) is proposed based on the modified statistically optimal near-field acoustical holography and the wave superposition method. The basic idea is that the fictitious sources'location essential to the wave superposition method can be decided by the modified statistically optimal near-field acoustical holography, and then complex sound field will reconstructed by the wave superposition method with high efficiency and precision. Besides the reconstruction of the patch sound field, the JPNAH has many other virtues. Firstly, the JPNAH technique is not sensitive to the sound source outside the patch area. Secondly, compared to the NAH, the JPNAH performs better in the low frequency range. Finally, using the JPNAH, it is possible to achieve a real-time mapping of the complex sound field. Once the sound field reconstructed, normal templates and abnormal templates based on acoustical holography can be constructed. Comparing the object operating condition with these templates, the differences between them can be found. Then, fault features can be found. Further more, fault diagnosis can be performed with some operating parameters of machines. The main contents of this dissertation can be summarized as follows.Firstly, the background of fault diagnosis will be introduced. The research history of machinery fault diagnosis technique and acoustical fault diagnosis technique will be summarized. Then, the development of noise source identification and acoustical holography technique will be reviewed. Among them, the near-field acoustical holography and equivalent sources method and modified statistically optimal near field acoustic holography are analyzed in specific. With the aim at extracting fault features for machineries in situation by using acoustical feature methods, a number of noise source identification methods are compared with advantages and disadvantages. A foundation of this work is constructed.After that, the basic theory and reason of sound radiation from vibrating structure of machineries are analyzed. And also the mechanical sound radiation problems are described in numerical formulations. And the near-filed acoustical holography (NAH) algorithm is deduced. Some filters in wave number domain for NAH are also discussed. Then, the NAH are discredited. Through numerical simulation, it is shown that NAH can accurately identify sound sources certain cases. It also shows that it causes"wrap-round errors"and windowing effects in the calculations. To overcome these shortcomings, the holography surface muse as large as two times of source surface. As for large scale objects at high frequency, it will need a large number of microphones. The test and reconstruction calculation will cost lots of time. Also, the testing costs will increase. These disadvantages hinder the broad applications of NAH in practice. Through theoretical analysis, a foundation is constructed for the acoustical holography based fault feature extraction technique.Based on the SONAH and WSM, a new method named JPNAH is presented. This method can reconstruct the partial sound field without the information of the sound source. The simulation show that several influence factor is essential to the reconstruction, such as sound source type, measurement plane, reconstruction plane, measurement error and the number of the k-vector. The conclusions are: 1) JPNAH is better than NAH because the measurement array can be random style; 2) JPNAH can simplify the measurement and reduce the experiment cost; 3) JPNAH is more sensitive to the amplitude mismatch than to the phase mismatch; 4) In JPNAH, the number of the k-vector must larger 200. Besides the reconstruction of the patch sound field, the JPNAH has many other virtues. Firstly, the JPNAH technique is not sensitive to the sound source outside the patch area. Secondly, compared to the NAH, the JPNAH performs better in the low frequency range. Finally, using the JPNAH, it is possible to achieve a real-time mapping of the complex sound field. The regularization method is essential to the reconstruction of the sound field based on JPNAH. The Tikhonov approaches in conjunction with different regularization parameter selection strategies have the ability to restrain the effect caused by the measurement noise. In practice, it is hard to obtain the knowledge of the noise variance, so the Tikhonov in conjunction with regularization parameter selection strategies that does not require the knowledge of the noise variance must be used in JPNAH. Based on the accurate reconstruction of the sound field, the acoustical-based diagnosis of the machinery will be achieved. In this paper, several regularization methods are discussed, and the best method is selected. Tikhonov in conjunction with Engl's criterion is the best to the JPNAH.After that, experiments are done to evaluate feasibility and accuracy of the techniques present by this work: the JPNAH and the acoustical fault feature extraction technique base on it. That will make a basis for its application in situation. The present hardware of the vibration laboratory is introduced. A set of microphone array and data acquisitions system are designed and implemented. The principle of the experiment is described. On basis of that, experiments are performed in a semi-anechoic and anechoic chamber. The sound source models are made up of a sound box, a motor and the fan of the computer. Sound source identification and fault feature extraction are performed on the sound source model. After acquisition of the sound pressure data, the experimental results are analyzed. The efficiency and accuracy of the technique are proved by experimental results.Conclusions are given at last. The innovations are summarized. At the same time, some advices are given for the future research for the fault feature extraction based on acoustical holography. Also, remarks are given on some problems.