Reconstruction based sensor fault identification in chemical processes

This paper discusses the reconstruction based sensor fault identification in a chemical process. Many approaches to monitor chemical processes have been developed and modified to isolate various abnormal operating conditions. One of those methods is the reconstruction based approach which has been developed from the missing data estimation to treat incomplete data obtained from a chemical process. We focus on the sensor fault identification using the principal component analysis based on reconstruction and propose new identification indices, FII_M and FII_R, which can distinguish an abnormal sensor from the others in both of model and residual spaces. In addition, a problem of reconstruction based sensor fault identification and identification performances using different reconstruction methods are discussed in this paper. The proposed indices are applied to a simulated nonisothermal continuous stirred tank reactor.