Fault Diagnosis Method For Sensor Of Hydraulic Condition Monitoring System Based On Bayesian Network

With the development of advanced manufacturing theory,a new round of manufacturing reform has become the focus of attention around the world.As the foundation and core of a new round of manufacturing reform,the intelligent monitoring and health management of industrial equipment has gradually received attention from everyone.In all kinds of industrial equipment,hydraulic device as the basis and power source of equipment,its importance is self-evident.At the same time,the health monitoring of hydraulic device has been the research focus of the industry.However,because the sensor in the condition monitoring system of the hydraulic device needs to perform long-term uninterrupted high-frequency data collection,and the working environment is very harsh,so it is easy to cause aging and failure and even affect itself.The working status can even directly cause system failures.Therefore,how to accurately and quickly find and judge the types of faults is particularly important and has high research value.First of all,in order to use Bayesian network to identify and judge sensor faults in the hydraulic condition monitoring system better,this paper starts from the causal judgment between two variables,and addresses the problems of low accuracy and poor stability in the causal direction judgment.A causal direction reasoning framework based on parallel integration is proposed,and it is verified by theory and experiment that this framework can achieve higher accuracy and stability on both simulated data sets and real data sets.Secondly,when the traditional Bayesian network makes predictions,its structure often relies on the problems of artificial experts and lack of reliability and accuracy.Based on the consideration of time efficiency and final structure accuracy,this paper proposes two Bayesian structure generation methods,which are reinforcement learning based Bayesian network model generation method based on modern machine learning theory and causal direction based Bayesian network model generation method based on modern mathematical theory.Among them,the Bayesian network model generation method based on reinforcement learning uses reinforcement learning to improve the search performance of the optimal directed acyclic graph in the Bayesian network structure generation,thereby improving the generation accuracy of the Bayesian network structure.The Bayesian network model generation method based on the causal direction brings the causal relationship model to the judgment of the relationship between variables,and improves the traditional K2 score search algorithm by judging the causal strength between the variables to form a new Bayesian Network model generation method.In view of the above two methods,experiments are carried out based on the Bayesian network model and data generated by simulation.The experimental results show that their accuracy and variance are less than those of various traditional methods,and they have higher stability.Finally,this paper proposes a sensor fault diagnosis method of hydraulic condition monitoring system based on Bayesian network.This method calculates the theoretical value of the sensor by using the position of the sensor in the hydraulic condition monitoring system,so as to obtain the residual between the sensor and the observed value of the sensor,and finally judge whether the sensor has a fault and the specific fault type.At the same time,the actual effect of the method is verified through the open data set.The experimental results show that the method can effectively solve the above problems.