Fault Detection Method based on Auto-associative Kernel Regression and Interval Type-2 Fuzzy Logic System for Multivariate Process
Seunghwan Jung, Minseok Kim, Joungguen Kim, Sungshin Kim · 2021
In multivariate processes (e.g., mechanical, chemical process and power plant, etc.), because several systems are complexly connected and operated, equipment faults can lead to failure or malfunction of the entire process. To improve the safety and reliability of the process, there is a need for a fault detection technology capable of preventing an accident by detecting the fault of an object in advance. In this paper, we detect the faults in multivariate processes using auto-associative kernel regression(AAKR), which is mainly used to detect faults due to benefits applicable to multiple-mode processes. AAKR is a nonparametric multivariate method that compares the similarity between the training data stored in memory and the query vector (testing data), and then assigns high weights to the high similarity vectors to compute the estimated vector. However, since there is a problem that you need to set the proper bandwidth parameter of the weighting function to calculate the weight, it is very important to reduce the parameter uncertainty by properly adjusting the bandwidth parameter according to the properties and characteristics of the input data. The conventional methods for setting of bandwidth parameter is used by trial and error. This method takes a lot of time and computation power. In this paper, we set the proper bandwidth parameter of weighting function in AAKR algorithm using interval type-2 fuzzy logic system (IT2FLS) that can deal with uncertainty. Dataset simulated from a multivariate dynamic process model was used to experiments, and the fault data was generated by applying two types of disturbance (e.g., bias and drift) to the simulated data. Experimental results show that the proposed fault detection method has better performance than the conventional method for setting of bandwidth parameter.