Application Of Improved Bayesian Method In Equipment Fault Detection

With the rapid development of modern technology,the country has paid more and more attention to the development of the manufacturing industry,especially after the artificial intelligence technology advanced,which has brought new opportunities for the manufacturing industry development.As the manufacturing industry gradually becomes more intelligent,it is increasingly demanding fault detection and diagnostic techniques.In recent decades,with the large-scale development of industrial equipment,traditional,man-power-based fault detection methods are no longer feasible,and instead they are independent monitoring,analytical models and others.With the development of computer and digital control technology,data-driven methods have great advantages in the field of fault detection.In particular,the Bayesian method,which based on the probability theory,comprehensively uses prior knowledge,and the target object's various information is subject to pattern recognition and detection,providing a simple but powerful fault detection technology.As a new data-driven method,Bayesian method has gradually attracted people's attention,and now it has good performance in many fields.This paper mainly adopts weighted naive Bayes algorithm to detect faults,determines the weights of weighted naive Bayes algorithm by multiple linear regression model,and solves the influence of traditional Bayesian independence condition assumptions.The improved algorithm applied to the Tennessee-Eastman fault detection process.The main tasks of this article are:1.Bayesian classification is a commonly used efficient classification method in data analysis,but its attribute independence assumption influences its classification performance.For this type of problem,this paper proposes a weighted naive Bayes algorithm based on multiple linear regression model.Firstly,analysis correlation between attributes by multiple linear regression model and then the correlation value is used as a weight coefficient.Finally,a weighted naive Bayes classification algorithm is used for classification.The experimental results indicate that the improved algorithm improves accuracy however,as attribute weight estimation process increases the complexity of the algorithm,the algorithm's consumption time increases.The Naive Bayes algorithm overall performance is still improved.2.For industrial processes with time-varying,multi-operating conditions and complex data distribution,this paper proposes an Improve Weighted Naive Bayes fault detection method,use the test data to estimate the weight and select attribute based on the relationship between attributes.Tennessee-Eastman(TE)data is basis in this article,design different test scenarios and compared with similar methods in the literature,verifies the effectiveness and practicality of the improved algorithm in this paper.