Research On Fault Diagnosis Of TE And FPSO Production Process Based On Data-Driven

The fault diagnosis of productive process is one of key problems in the process industry.Further study and application of fault diagnosis can make a great improvement on industrialprocess in aspect of keeping the working safety and raising the economic benefits. The faultdiagnosis based on data-driven has become a focal point in the process industry at present.The methods and the development trend of fault diagnosis are introduced in this paper.The process of fault diagnosis is taken by the research of principal component analysis (PCA)and two statistics. The two statistics consists of the Hotelling T2 statistic and SPE statistic. Atthe same time, some kernel functions are analyzed. The kernel principal component analysis(KPCA) is proposed. The steps of the fault detection method based on KPCA are also given.The Tennessee Eastman (TE) production process which can verify the fault diagnosisalgorithm and the Floating Production Storage & Offloading (FPSO) production processwhich is running on the sea are analyzed and simulated. The simulation of TE productionprocess uses a PID control algorithm. The required data of fault diagnosis is obtained by thetwo simulations.The algorithms of fault detection by PCA and KPCA are tested in TE production processand applied in FPSO production process. By comparing the detection efficiency and theundedetected error rate of PCA and KPCA, the result of KPCA is better than that of PCA inthe strong nonlinear production processes.At last, a fault identification method which uses Decision Directed Acyclic Graph(DDAG) algorithm is proposed and applied to the TE and FPSO processes.The result shows itcan meet the requirements of the two production processes.