| In this paper, based on the project of the main reducer testing machine for several well-known domestic enterprises, the intelligent diagnosis methods based on identification of vibration spectra were put forward. The research is focusing on the key techniques: signal processing methods, Fusion and decomposition of image features, image compression and feature extraction methods, intelligent multi-class classification algorithm, decision fusion methods. Moreover the research results were applied to performance testing of main reduce, engineering applications show that these kinds of detection methods have a high practicality. Main contents are as follows:First of all, against the edge frequency of the vibration signal modulation characteristics, a non-linear wavelet technology - morphological wavelet was introduced. Based on morphological wavelet theory, the MWP&BT algorithm was advanced, and its principle of demodulation was discussed, moreover it was applied in demodulating vibration signal of rolling bearings and main reduce. For non-stationary and non-Gaussian signals, modern signal processing methods, such as time-frequency analysis, bispectrum, bispectrum cycle, were analyzed and compared, so as to find out the processing methods which are suitable to online signal.Second, In view of contrast sensitivity function (CSF) in human vision system (HVS) model, the image fusion method based on morphological wavelet and CSF was advanced, so the fusion image would be more In line with the characteristics HVS, in addition the image fusion method was used in intra-class and inter-class Fault feature fusion. The decomposition method of vibration feature by processing of vibration spectrum was proposed and its principal was analyzed, and further some operator for image calculation was put forward, its results was compared with the EMD.Third, the morphological wavelet based image compression method was introduced. For image feature extraction of vibration spectrum, the two kinds of gray-level moment algorithm were advanced, and their related factors which impact the result of feature extraction, furthermore they were used in feature extraction of SPWVD spectrum and bispectrum.Fourth, the concept of artificial immune system was introduced and the general framework of intelligent fault diagnosis based on immune network model was established. The stability of dynamic modal, Farmer modal, of idiotypic network was analyzed, combining the advantage of Farmer modal and aiNet, a kind of artificial immune network optimization algorithm (AINOA) was put forward. The feasibility of multi-class classification referring to the modal of antigen-antibody binding energy was proved. In reference of the Optimization capacity of Farmer-aiNET for the sample space and ability of classification of binding energy, a kind of classifier named artificial immune network classification algorithm (AINCA) was advanced, and then it was used in fault diagnosis of rolling bearings.Fifth, the concept of fuzzy measure and fuzzy integral were delivered, and thus the decision-making fusion model based on fuzzy integral was introduced, with a view to improving classification accuracy. Through decision-making fusion of four artificial immune network classifiers (AINC) by fuzzy integral, the classification accuracy of rolling bearings significantly increased.Finally, the dynamic model of the main reducer was established, and their dynamic characteristic was simulated, so as to provide a theoretical support for fault identification. Then, the working principle and main functions of the main reducer test machine were introduced, and three types of measured vibration signal analysis were analyzed. In accordance with the aforementioned image feature extraction methods, the feature vectors were extracted, and then they are identified using the classification algorithm AINCA. In order to improve recognition accuracy, four AINC was built and their results were integrated by fuzzy integral-based decision-making fusion.
|
PDF Full Text Request