Mechanical Fault Diagnosis Based On Deep Learning

Dynamic monitoring of the mechanical equipment operating conditions is very important in the current field of mechanical equipment fault detection,and the purpose is to diagnose the potential faults of mechanical equipment in time by analyzing the information from monitoring.In addition,the problem of electronic circuit fault diagnosis of mechanical equipment has been one of the hottest topics studied by experts and scholars in this field.With the rapid development of artificial intelligence technology,especially deep learning in recent years,and the successful application and good performance of this technology in many vertical fields,it is a direction worthy of attention and research that applying deep learning with feature extraction of electronic circuit fault information and fault diagnoses.The research shows that the characteristic information of electronic circuit faults is often disturbed by environmental noise,circuit element tolerance and original non-linear data.Great demands are being placed on the accuracy and timeliness of feature extraction and fault diagnose by growing scale of electronic circuit in mechanical equipment.Basing on the fact above,we propose a new a deep-learning-based fault diagnosis model for mechanical equipment electronic circuits,which combines the deep learning method and sparse auto-encoders to solve the difficulty in extracting fault information.The main research of this paper including the following three aspects:Firstly,aiming at the feature extraction problem of equipment fault information,this paper proposes a feature extraction method based on sparse marginalized denoising auto-encoder(SmDAE).This method improves the error function of sparse auto-encoder by computing correlation entropy then enhances robustness to non-Gaussian noise.Furthermore,a convolutional neural network is used to reduce edge noise of auto-encoder.Finally,we use the improved SmDAE to extract fault information of electronic circuit by combining the spare auto-encoder with improved error function and marginalized denoising auto-encoder.Secondly,this paper proposes an ISmDAE-SVM based model to solve fault diagnose problem.This model takes the feature of fault information extracted by the ISmDAE as its input,and then connects BT-SVM classifiers.The output is whether the mechanical device may malfunction.Finally,in the electronic circuit fault diagnosis experiment of the automatic welding machinery equipment-three-point welding machine,the ISmDAE-SVM electronic circuit fault diagnosis model proposed in this paper has a good fault diagnosis effect.