Investigation Of The Fault Diagnosis And State Assessment Technique For Bearing Clearance Of Reciprocating Compressor

Reciprocating compressor is an essential power equipment to compress and transport gas and used widely in petroleum and chemical industry. With the increasing emphasis on production safety in present society, it is necessary to implement fault diagnosis technique for reciprocating compressor. Owing to the complex structure and numerous excitation sources, it is necessary to conduct further investigation of the fault diagnosis technique for reciprocating compressor. Taken bearing clearence fault of reciprocating compressor as research object, this paper proposed a fault diagnosis and state assessment technique for bearing clearance fault of reciprocating compressor by further research on the multi-body dynamics simulation, signal adaptive decomposition, nonlinear signal quantitative description and intelligent pattern recognition and assessment method during the data acquisition, feature extraction and pattern recognition and assessment procedures. The main research contents are as follows:Due to the difficulties to perform multiple types fault experiments and insufficient fault data, multi-body dynamics simulation method of bearing fault states for reciprocating compressor with joint clearence was studied. Parameters of joint clearance model are the key factors for dynamic response, and the difference of simulated and experimental data was significant by using the design parameter values. To obtain a simulated acceleration more similar to experiments, the parameters of joint clearance contact force model were taken as the factors and optimized based on Genetic Algorithms method. The relation between parameters of model for different clearance states is the basis for the application of optimized parameters. By discussing the effect of increased clearance on model parameters, the relation between parameters of model for different clearance states was concluded, and the optimized parameters was developed for multi-body dynamics simulation of different clearance states according to the concluded relation.Owing to complex transmission path caused by the moving location of bearing and nonlinear characteristics of vibration signal, an feature extraction method based on multifractal and singular value decomposition for bearing clearance fault location of reciprocating compressor was proposed. Data of single-sensor is difficult to fully observing the state information, so datas of multi-sensor was used to extend the observation range. The generalized fractal dimension can characterize local scale behavior of signal more appropriately, so an initial feature matrix was built by calculating the generalized fractal dimension of multi-sensor signal. The eigenvalues of matrix were extracted by singular value decomposition method, and were taken as eigenvectors. Taken support vector machine as pattern classifier, the validity of this method is proved by the test of simulation and experimental datas.According to the strong nonstationarity, nonlinearity and features coupling characteristics of vibration signal, a feature extraction method based on LMD and MSE for bearing clearance fault extent of reciprocating compressor was proposed. Decomposition results of trandtional LMD method will be distorted due to the poor envelope accuracy for strong non-stationary signal. In order to improve the envelope accuracy of local mean and envelope estimation, Hermite rational interpolation method is used to construct envelope curves, and the envelope curves were optimized based on the extreme symmetry point. Multiscale entropy is an effective method to describe the nonlinearity of vibration signal. The vibration signal was decomposed into a set of PFs, and then multiscale entropy of the first several PFs were calculated as eigenvectors with different scale factors. According to the maximum of average Euclidean distances between different eigenvectors, the eigenvectors which have the best divisibility were selected using Genetic Algorithms.In order to diagnose and assess faults of reciprocating compressor intelligently, pattern recognition and state assessment method based on binary tree SVM was studied. Separability measure is the basis for binary tree SVM hierarchy construction. A new separability measure was constructed by the average distance of samples in one class and the average distance of samples between different classes, and an improved binary tree SVM hierarchy construction method was proposed. An independent parameter SVM algorithm in which parameters of SVM sub-classifiers were optimized respectively was proposed for the parameter optimization process. Different types of binary tree algorithm were compared using UCI standard data sets and reciprocating compressor fault features, and the improved binary tree SVM can improve identification accuracy effectively. The SVM state assessment model was bulit based on the probability statistics characteristics of SVM for the sample set of one class, and this model can assess fault states accurately for the experimental and simluated data.