Multi-Label Based Simultaneous Fault Diagnosis In Industrial Process

An accurate and reliable fault diagnosis system is of great importance for personal safety, as well as the economy of the plant. In this thesis, the investigations are mainly focus on multi-label based simultaneous fault diagnosis classification and its application in:Tennessee-Eastman Process (TEP), Fuel cell system and its experimental platform. The main contributions of the dissertation are as follow:1. Through building the fault simulation platform for fuel cell system, the fan overheating, battery flooded and other three typical fault simulations are realized. And it solves the problem of insufficient data on the fuel cell experiment platform. Also it provides the basis for the study of multi-label classification in fault diagnosis.2. Compare the mono-Label (mL) classification with Multi-Label (ML) classification fault diagnosis, and the both algorithms are based on Least squares support vector machines (LS-SVM). The result shows that the multi-label classification is more accurate to diagnose simultaneous faults, and it simplifies calculation.3. The paper proposes the multi-label based Relevance vector machine (RVM) and Bayesian extreme learning machine (BELM) to manage simultaneous fault diagnosis. And both algorithms are compared with multi-label LS-SVM on simulation model and experimental platform. The results show that, RVM does the best performance, especially in terms of precision. The second is BELM which is good at learning speed. And the LS-SVM plays worst in precision, recall, error control and accuracy.4 In the fuel cell system simulation model, experimental results show that with the increase of the capacity of the test sample data, three algorithms’ precision are all reducing, and the RVM is slowly falling off. But to BELM, it is sensitive to the sample size, and poor at stability.