Research And Application Of Sensor Fault Diagnosis Algorithm

As an important tool for industrial equipment monitoring,sensors are mainly used for signal and information acquisition to provide decision-making basis for control systems.Sensors usually work in harsh environments,and are prone to failures such as precision drop,deviation,drift and so on,which affect the reliability and accuracy of data,and cause serious consequences such as the collapse of the control system.Therefore,sensor fault diagnosis is very important to ensure the safe operation of the system.The task of sensor fault diagnosis consists of two parts: fault detection and fault diagnosis.Fault detection determines whether the sensor is working within the normal measurement range,and fault diagnosis identifies what kind of fault occurs in the sensor.This paper studies the fault detection and fault diagnosis of sensors based on deep learning.The main work is as follows:(1)Aiming at the problem that the fault signal features are not obvious in the process of sensor detection,and it is difficult to extract deep features,a sensor fault detection algorithm based on convolutional neural network is proposed: MFCC-Res Net.Using the ability of MFCC in speech feature extraction,the dynamic characteristics of the fault data are described by calculating the Mel cepstral coefficient of the fault data,and the extracted fault features are used as the input of Res Net for fault detection.The experimental results show that MFCC-Res Net can extract deeper fault features.Compared with the Res Net,SVM,and MFCC-SVM models,the detection accuracy of the sensor fault detection algorithm based on MFCC-Res Net is respectively improved by 1.84%,2.98% and 1.21%.(2)Aiming at the problems of less fault data and low fault diagnosis accuracy in the process of sensor diagnosis,a sensor fault diagnosis algorithm is proposed: GANRes Net.The algorithm utilizes the characteristics of generative adversarial networks to fit sensor fault data,and generates a large number of virtual sensor fault data to balance and expand the existing dataset.Aiming at the problem that the traditional activation function will compress the fault information in the process of nonlinear feature extraction,an activation function STAC-tanh based on the tanh function is designed.By automatically adjusting the shape of the activation function,the relationship between the nonlinear feature transformation and the input signal is established.relationship to improve the accuracy of fault diagnosis.The experimental results show that in the case of insufficient fault data,the sensor fault diagnosis algorithm of GANRes Net can retain effective fault information and improve the fault diagnosis performance.(3)Based on the fault detection and diagnosis algorithm proposed in this paper,the wind turbine sensor fault diagnosis system is designed and developed,which realizes the fault detection and fault diagnosis of the wind turbine sensor.The system has the functions of data acquisition,data storage,real-time data monitoring,fault detection,fault diagnosis,etc.It is easy to operate and has high diagnostic accuracy.In summary,this paper studies from two aspects: strengthening fault feature extraction and expanding unbalanced data sets,and proposes a sensor fault detection algorithm and fault diagnosis algorithm based on deep learning,and applies the proposed algorithm to the wind turbine sensor fault diagnosis system.design development.The example verification shows that the algorithm has better fault detection performance and diagnosis effect,and the diagnosis system has better stability and ease of use.