Research On Fault Detection Of FPSO Production Process Based On The Kernel Function

The method of fault diagnosis is extremely important in the process industries. Especially for the modern industrial production,research and application of the production process fault diagnosis plays an important role for the protection of the safety of production, improve production efficiency and economic benefits.As an important part of the fault diagnosis, fault detection method is the focus of the study of fault diagnosis.This paper analyzes the application of the current situation and development trend of the kernel function.Machine learning theory, statistical learning theory and support vector machine theory are studied.The typical production process of the floating storage and offloading production system is analyzed, and18important process variables in the oil, gas and water separation systemare are selected as the objects.We obtained the date that validation fault detection method by the normal operating state of the production process as well as runing the state of the simulation after man-made failure.Fault detection metheds of kernel principal component analysis (KPCA) and kernel independent component analysis (KICA) are applied in the FPSO typical production process.First, establish the mathematical model of the normal operating conditions to calculate the normal operating conditions T2statistic and SPE statistics, and then make the test data preprocessing with standardization and calculate the test data T2statistics and SPE statistic.By monitoring the change in the amount of data with normal statistical data to determine whether there is a fault occurs.By comparing the PCA and KPCA of fault detection methods and ICA and KICA of fault detection methods, the small sample size and non-linear problems are drawn.The kernel function combined with multivariate statistical method has obvious advantages in fault detection.