Study on Support Vector Machine-Based Fault Detection in Tennessee Eastman Process

Abstract EN

This paper investigates the proficiency of supportvector machine (SVM) using datasets generated by TennesseeEastman process simulation for fault detection.

Due to its excellentperformance in generalization, the classification performanceof SVM is satisfactory.

SVM algorithm combined with kernelfunction has the nonlinear attribute and can better handle thecase where samples and attributes are massive.

In addition, withforehand optimizing the parameters using the cross-validationtechnique, SVM can produce high accuracy in fault detection.

Therefore, there is no need to deal with original data orrefer to other algorithms, making the classification problemsimple to handle.

In order to further illustrate the efficiency,an industrial benchmark of Tennessee Eastman (TE) process isutilized with the SVM algorithm and PLS algorithm, respectively.

By comparing the indices of detection performance, the SVMtechnique shows superior fault detection ability to the PLSalgorithm.