A novel fault diagnosis method based on optimal relevance vector machine

作者：He, Shiming; Xiao, Long; Wang, Yalin; Liu, Xinggao^*; Yang, Chunhua; Lu, Jiangang; Gui, Weihua; Sun, Youxian

来源：Neurocomputing, 2017, 267: 651-663.

DOI：10.1016/j.neucom.2017.06.024

• Fault diagnosis
• Relevance vector machine
• Particle swam optimization
• Differential evolution
• Tennessee Eastman process
• Ethylene cracking furnace process

Fault diagnosis is always a crucial and challenging technology in industry, which contains huge amount of variables need to be measured and analyzed. A high-efficiency fault diagnosis method can reduce the economic loss drastically. This paper presents a new fault diagnosis method based on relevance vector machine (RVM) to deal with the small sample data. Particle swam optimization (PSO) algorithm and differential evolution (DE) algorithm are employed together to optimize the parameters of RVM, which strengthen the classification ability and provide strong potential in prediction fault. Then, a novel fault diagnosis method DEPSO-RVM is obtained, which has the advantages of high accuracy and low false positive rate. Furthermore, the proposed method DEPSO-RVM is accomplished with two representative test data TE process and ethylene cracking furnace process. Besides, in order to compare the performance of DEPSO-RVM, SVM, DEPSO-SVM and PSO-RVM are applied to Tennessee Eastman process and ethylene cracking furnace process as well as the basis of comparison. The results show the validity of the proposed method on TE process and ethylene cracking furnace process and give a reference on industry process.

• 出版日期2017-12-6
• 单位中南大学; 浙江大学