Fault Detection And Diagnosis Based On Non-negative Matrix Factorization With Sparseness Constraints

With the development of computer applications and distributed controlsystems, more variables can be monitored and handled quickly in industrialprocess, which makes industrial systems more complicated and intelligentwith fast step. In order to assure the safety and reliability of the equipmentand improve production efficiency and product quality, fault detection anddiagnosis technology is particularly necessary. Data-driven fault detectionand diagnosis method is the most popular one in the field of fault detectionand diagnosis, and its core is how to extract its feature information from largeamounts of data.As a new matrix decomposition method, Non-negative matrixfactorization(NMF) can learn the parts of large amounts of data and get twofactors which is non-negative and sparse. Non-negativity means onlyaddictive computing, and sparseness means local feature of the data. So, theNMF method shows stronger interpretation than traditional matrix methods,that is, parts-based representation. In view of the above advantages, this article will introduce NMF into thefield of fault detection and diagnosis. A fault detection model based NMFwith Sparseness Constraints(NMFSC) is proposed, and a new methoddetermining fault propagation paths is presented through Component SignedDigraph(CSDG) and data-reconstruction. Which is a useful attempt in thefield of fault detection and diagnosis.Specifically, the main work of this article include the following aspects:Firstly, on the base of NMF algorithm, non-negative matrix factorizationwith sparseness constrains (NMFSC) is proposed by applying sparsenessconstrains on coefficient matrix H. The algorithm is able to identify the maininformation about the differences between data individuals in the featurespace.Secondly, On the basis of NMFSC, a fault detection and diagnosismodel based on NMFSC is proposed by presenting two suitable monitoringstatistics: CUSUM and SPE. When a fault is detected, the main variablescausing the fault can be determined by data-reconstruction method.Thirdly, a new method based on NMFSC is proposed to analysis faultpropagation path in a system. To solve the hardship when modelinglarge-scale system with too many components and complicated topology, aGeneric Component Model(GCM) is presented. The CSDG of a system can be established easily through GCM. The most likely fault propagation pathcan be determined by Reconstruction index of compatible path direction inCSDG.Finally, the proposed fault detection and diagnosis method based onNMFSC and data-reconstruction is applied on Antarctic Zhongshan Stationmicro-grid system. Aging batteries can be detected by the micro-grid batteryperformance monitoring, making maintenance more convenient.