Acoustic-based Condition Monitoring Of Machinery Using Blind Signal Processing

Acoustical signals, similar to mechanical vibration signals, indicate a lot about the mechanical system, because the acoustical features will change along with the working condition of equipments. Thus, acoustic signature analysis is regarded as a very powerful technique for detection and diagnosis of faults in rotating machinery. Moreover, acoustical signal measurement has several advantages, such as non-destruction, non-contact and simple using. However, acoustic-based diagnosis is confronted with many difficulties. For example, the sound field is extraordinarily complex in practice and the captured signals from microphones are complicated mixtures. It is too difficult to extract useful information from measured mixtures directly, since the target signal is usually corrupted by other equipments'signals or noise. Consequently, in acoustic-based diagnosis, it is crucial to remove or restrain interference signals or background noise, and accurately extract the target signal from the mixed signals of low signal-noise ratio.Currently, blind signal processing (BSP) technology becomes a powerful tool in the field of separation of mechanical acoustical signals, because of its capacity of restoring or estimating original sources from an array of signals nearly without any prior knowledge. Therefore, the present research is supported by the National Natural Science Foundation of China (No. 50805071) and Scientific Research Foundation of Department of Education of Yunnan province (No.08J0009). The research, based on the theory of BSP, mainly focuses on random non-stationary acoustical signal extraction and separation in actual sound fields, which combines theoretical investigation, experimental investigation and system investigation. The main contents are given as below:(1) From the viewpoint of engineering application, this paper generally illustrates the background and significance of this study, and summarizes the state of the acoustic-based diagnosis, the theory of BSP and the application of BSP in acoustic-based diagnosis.(2) To solve the problem of monitoring acoustical signals in a short range, two improved algorithm are proposed:blind extraction algorithm for convolutive mixtures based on clustering and blind deconvolution algorithm based on genetic algorithm. The two algorithms can solve the uncertain problem of separated results which is created by various delay times in a wide range. Some simulations and experiments are carried out, which can verify the practicability and reliability of the improved algorithms.(3) With the consideration of the fact that acoustical signals are measured in distant range, a theoretical framework of blind deconvolution method for extraction of impulse signal is proposed. Base on the framework, a blind deconvolution algorithm based on block-based model and optimization and a blind deconvolution algorithm using reference signal are proposed. Experiments with simulated signals and real acoustical signals from a test rig with faulty bearing in a room are conducted to demonstrate the effectiveness of the improved algorithms. The results show that the adaptability and ability of the algorithms are powerful.(4) According to the practical application of BSP, the underdetermined problem and the number estimation of sources problem are studied. By means of constructing high dimensional matrix, principle and minor component analysis and multi-resolution sub-band decomposition, a blind deconvolution algorithm using a single signal is proposed. Moreover, an improved blind deconvolution algorithm is proposed based on the self-adaptive parametric algorithm of FCM. This algorithm can not only separate the mixtures but also estimate the number of main acoustical sources when the number of observations is less than the number of source signals. Simulation and experiment examples are given to demonstrate the effectiveness of the both algorithms. The results show that the algorithms can instruct the engineering application of BSP to some extent.(5) An acoustic-based monitoring system (ABD-BSPTool) has been implemented based on theoretical investigation of BSP. This system is developed by using MATLAB. The implemented ABD-BSPTool has been tested in the laboratory, in which a test environment has been built. The results show that the system is effective. The ABD-BSPTool is of value to promote acoustic-based condition monitoring and diagnosis based on BSP to practical application.