The Research Of Identification Method For Rolling Bearing Early Fault Based On Manifold

Rolling bearing is one of critical and failure-prone components for rotating machinery and equipment, and its running status directly affect the performance of the whole equipment system. Therefore, condition identification and fault diagnosis of rolling bearings are of great significance to ensure the safety and efficient operation of machines in engineering applications. Signal noise reduction and fault feature extraction are the core contents of condition identification and fault diagnosis. Nevertheless, rolling bearing running environment is complex, including much interference and noise, and most of the fault signals are non-stationary and nonlinear, thereby the efficiencies of traditional noise reduction and fault feature extraction methods reduce greatly. To this end, in this paper, contraposing the characteristics of early weak fault vibration signals under strong background noise, rolling bearing fault signal noise reduction and feature extraction are intensive studied combing manifold method and other modern vibration signal analysis method. The main contents are as follows:1. The background and significance of the selected topic and the development process of rolling bearing fault diagnosis at home and abroad are discussed; The two key problems noise reduction and fault feature extraction in contemporary rolling bearing fault diagnosis are detailed reviewed; Rolling bearing common failure forms and its causes and consequences are dissected; The rolling bearing vibration mechanism and the specific using methods of time domain, frequency domain and time-frequency domain vibration analysis theory are expounded.2. In view of the problem for rolling bearing early weak fault features easily being overwhelmed by noise and interference components, an improvement EMD early weak fault signal noise reduction method is proposed combining with KPCA manifold, EMD and LTSA manifold. Firstly, to make a first KPCA manifold noise reduction before EMD, then extract the low dimensional manifold components of all IMF coefficients using LTSA manifold algorithm. Lastly, sum the above manifold components to obtain new signal, thereout to achieve signal noise reduction. This method not only make full use of the EMD completely adaptive analysis advantage for non-stationary nonlinear signals, but also can effectively overcome the noise influence for EMD, in addition, this method is very good to solve the fault information leakage problems suffered by ignoring most of the components in conventional EMD applications. Comparative analysis of simulation and the engineering practice signal verify the validity and superiority of the proposed method.3. Direct at how to effectively extract the sensitive characteristics of non-stationary nonlinear fault signal, a time-frequency fault feature extraction method is put forward for rolling bearing based on two dimensional manifold and Hilbert time-frequency spectrum. Firstly, on the basis of Hilbert time-frequency analysis, a two-dimensional manifold method is used to extract signal manifold ingredient to reduce dimensions and extract the sensitive parameters. Secondly, singular values entropy is defined to quantitative measure the differences of manifold ingredient under different fault status. This novel method directly use two-dimensional information as research object and thus avoiding the information loss for one dimensional manifold algorithm in the necessary process transforming two-dimensional information into vector. For another, it can easily find more local data characteristics hidden in high-dimensional data manifold structure compared with PCA method. The effectiveness of the proposed method is verified by engineering signal analysis. Finally, manifold singular value entropy combined with probabilistic neural network are used to confirm the high reliability of this method.4. For engineering practical operation and application problems, a set of rolling bearing fault vibration analysis system is developed. This development adopts progressive type and modularization combining signal processing technology, fault diagnosis technology, database knowledge, virtual instrument technology and human-machine interaction technology. This system is convenient and practical and meets the project site needs very well.