Research On Machinery Equipment Fault Diagnosis Based On Deep Contractive Auto-encoders Network

With the continuous development of computers,sensors and communication technologies,modern industrial system is becoming increasingly sophisticated.It is important to ensure the reliable operation of machinery equipment and locate faults accurately in case of industrial equipment failure.Fault diagnosis is a technique to determine the cause of a fault when a device fails.It can help to perform quick troubleshooting and failure recovery in actual production.There are many kinds of fault diagnosis methods.At present,fault diagnosis based on classic machine learning is widely used.This method can directly retrieve the fault characteristics of the equipment from the collected industrial equipment operation data without relying on the prior knowledge of the system.However,classic machine learning algorithms have problems such as dimension disasters,overfitting,etc.Besides,classic machine learning based methods require artificial construction features for different devices so that the generalization ability is poor.This article addresses these issues by introducing deep learning into fault diagnosis of machinery equipment.The main tasks of this article are:1.This paper proposes a fault diagnosis method based on deep auto-encoder network.The deep network is constructed by contractive auto-encoders,which realizes automatic mining of fault features and removes the dependence on artificial structural features so that it improves the applicability and generality of the method.2.This paper proposes a one-class classification fusion algorithm,which can improve the accuracy of the model by distinguishing the existing failure types and unknown failure types.3.The effectiveness of the proposed method is verified by experimental data and compared with the commonly used machine learning-based fault diagnosis methods.The factors affecting the accuracy of the model are analyzed in detail.