Gearbox Performance Degradation Evaluation Based On Clustering And Data Reconstruction

With the improvement of modern industry,health condition monitoring of machinery is a crucial task to guarantee reliability in industrial processes.Gearbox is the most critical component in mechanical equipment since it is more susceptible to failure.Bearing and gear are important components of gearbox.Therefore,It is of great practical significance to improve the reliability of equipment condition monitoring.This thesis focuses on the boundary-based model and reconstruction-based model in the performance degradation assessment(PDA)by theory and experiment.Research on the boundary-based PDA model,a performance degradation assessment method based on K-medoids clustering is proposed.In this part,the energy of wavelet packet decomposition is used to select the frequency band,and the Renyi entropy of the useful frequency bands are calculated as the feature vector.Using the K-medoids clustering model,the center points of the normal and failure features are found.The membership function is constructed by calculating the distance between the test data and the two kinds of centers,and the membership value of the test data and the normal center is used as the degradation index.Bearing is an important component of gear box.Results on artificially induced faults and bearing run-to-failure data demonstrate the proposed method is able to track the progress of bearing faults and detect them at incipient stage.Research on the reconstruction-based PDA model,a performance degradation assessment method based on autoregressive m odel and sparse decomposition is proposed.This method mainly uses the AR model to extract the state characteristics of the gear vibration signal,and uses the coefficients to dictionary learning to obtain a normal dictionary for sparse decomposition and reconstruction,and take the reconstruction error as the degradation index.The single factor analysis method is used to analyze the model hyper-parameters of the model and find the optimal parameters.Gear is an important component of gear box.The analysis of the gear run-to-failure data shows that the model can effectively identify the fault degrees of the gear and detect early failures.Research on the multilayer reconstruction-based PDA model,combining with the atuoencoder and sparse decomposition,a double layer data reconstruction type fault performance degradation assessment model is proposed.Time and frequency domain features,wavelet packet entropy,and AR model coefficients are extracted as comprehensive features.The atuoencoder reconstruct the original features,and the sparse decomposition reconstruct the hidden layer of the autocoder.Using autoencoder and dictionary learning to build double layer data reconstruction type fault PDA model,The reconstruction error is used as the degradation index.Aiming at the problem that the model has a large number of hyperparameters that require global optimization,PSO algorithm is used to obtain the hyper-parameters of the model,and the Mean-Square-Error-Ratio(MSER)is proposed to evaluate the effect of reconstruction as a new metric.The bearing data and the gear data are used to verify the proposed method.Bearing and gear experimental results demonstrate the efficiency of the proposed method for describing performance degradation and detecting early failure.This method provides an approach for reconstruction-based PDA model.