Fault Diagnosis for Nonlinear Biological Processes Based on Machine Learning Models

Kernel-based learning techniques have been widely used to monitor and detect faults in biological systems. However, it is well known that the data used in the training phase must be stored and used for validation purposes. This results in a high computation cost when the training data set is very large. To address the above issue, we propose in this paper a novel approach to jointly enhance the detection accuracy and reduce the execution time required for fault detection. The developed approach, so-called, reduced kernel PLS (RKPLS)-based generalized likelihood ratio test (GLRT) aims to reduce the number of training samples to build a new KPLS model. Then, it consists to apply a GLRT to the evaluated residuals obtained from RKPLS model for fault detection purposes. A simulation using a Cad system in E.coli (CSEC) is performed to show how the reduction of the training data set affects the computation time and fault detection accuracy.