Research On Equipment Fault Recognition Methods Based On Ensemble Learning

With the continuous improvement of the performance requirements of rotating machinery in today ’s society,its internal structure has become more complex.When the equipment fails,it will bring economic losses and casualties.Therefore,it is of great significance to monitor and diagnose the running state of the equipment in real time.Based on the ensemble learning method,combined with data acquisition and signal processing technology,this paper conducts in-depth research on the fault diagnosis task of rotating machinery equipment.The specific research contents are as follows:(1)The performance of the ensemble learning Light GBM model is limited by the traditional feature extraction method and the Le Net-5 network has poor diagnostic performance in the face of complex working conditions of the equipment.This method first converts the one-dimensional vibration signal into a two-dimensional gray image,and performs histogram equalization on the gray image to enhance the contrast.Then,the improved Le Net-5 network is used to extract the features of the gray image,and the extracted features are input into the Light GBM model for recognition and classification.Finally,the effectiveness and superiority of the proposed method are verified by the rolling bearing fault simulation test of Western Reserve University.(2)Aiming at the problem that a single model is prone to poor generalization ability and limited performance improvement potential in fault diagnosis,it is proposed to heterogeneous ensemble multiple single models through the Stacking framework to diagnose mechanical equipment.Firstly,the fault features of the equipment are extracted by wavelet packet transform.Then,the extracted features are put into the primary learner optimized by hyperparameters for prediction,and the prediction results are input into the secondary learner LR for fault classification.Finally,the diagnostic performance of the model is verified by fault simulation experiments.(3)Aiming at the limitations of shallow machine learning models that need to manually extract features and single neural network models in diagnosing equipment faults,a fault diagnosis method based on multi-objective cooperative differential evolution ensemble deep neural network model is proposed.Firstly,the network model with different activation functions is selected according to whether the model can converge.Then,the MOEA / D algorithm is used to iteratively evolve the hyperparameters of the network model,and then the DE algorithm is used to assign weights to the optimized network model.Finally,the weighted network model is ensemble by voting method and applied to the fault simulation experiment of axial piston pump,shaft and rotor system.