The Research And Application Of Fault Diagnosis Methods Based On Signal’s Sparse Representation And Deep Learning

Equipment condition monitoring and fault diagnosis techniques are critical to the efficient and stable operation of machinery,and any unpredictable failures and failures can result in significant loss of life and property.Therefore,researching effective equipment fault diagnosis methods is of great significance for maintaining life safety and ensuring industrial safety.With the development of high-performance acquisition and transmission equipment,the amount of data that can be used for fault diagnosis has also exploded,and the emergence of deep learning methods has been well adapted to the diagnostic tasks under large data volumes.But it also brings some new problems,such as the lack of model interpretability and the high cost of calculation.Fortunately,the theory of compressed sensing born at the beginning of this century clearly reveals the essential characteristics of natural acquisition data with redundancy and sparse compressibility.Based on this feature,this paper studies the data sparseness and model sparsity of deep learning,and proposes the Ensemble Sparse Supervised Model(ESSM).And tested on the bearing fault data.In order to integrate it into the Machine Health Monitoring System(MHMS),the MHMS system based on B/S architecture is also discussed in detail from technical points to development trends.Finally,based on the discussion,the ESSM model is implemented in the MHMS system by separating the two ends.The specific work is as follows:1)An integrated sparse supervised feedback model is proposed for the problem of excessive dependence and excessive separation of input characteristics and model relationships in artificial machinery-based rotating machinery fault diagnosis.The model firstly separates the learning of the feature and the learning of the model under the sparse constraints,and then uses them as the feedback of the other party to correct the learning error of the other model.The process is cycled until a predetermined critical termination condition is reached.The proposed model has been proven to be effective on real bearing data.2)For deep learning,the operator needs to manually select the hyperparameters in the training,which will increase the labor cost and time consumption problem,and introduce the Talos library into the proposed integrated sparse supervised feedback model,only at the beginning of the training.Set various super-parameters in the feature learning phase and the model learning phase.After starting the training,select the optimal set of hyper-parameters according to the generated reports and metrics,which greatly simplifies the operation steps and saves time in training.3)The front-end interface of the MHMS system is designed by using the Restfulstyle api,and the front-end interface of the MHMS system is built using Vue.The new MHMS system has been greatly improved in terms of scalability,development efficiency,maintenance costs and front and back office interaction.