Sensor Diagnosis System Combining Immune Network and Learning Vector Quantization

Abstract

In this paper, we propose a distributed diagnosis system combining Immune Network (IN) and Learning Vector Quantization (LVQ) for detecting fault sensors accurately in industrial plants. It has two execution modes, namely, its training mode, where LVQ extracts correlation between each two sensors from their outputs when they work properly, and its diagnosis mode, where LVQ contributes to testing each two sensors using the extracted correlation, while IN contributes to determining fault sensors by integrating these local testing results obtained from LVQ. How to define a threshold of each quantization vector for judging whether testing data satisfy the correlation or not, is also discussed to improve diagnosis capability of the developed system. It is shown that the thresholds can be determined effectively by the constraint that the hyperregion corresponding to the normal sensor outputs in each quantization vector space is a single region.
Diagnosis capability of the developed system is evaluated using a proto-type system for detecting fault sensors of a reheating furnace plant. By the proposed method, abnormal sensors, such as aged deteriorated ones, which have been difficult to be detected only by checking each output of sensor independently, are possible to be specified.