Hybrid Classification Method for Image-based Anomaly Detection in Manufacturing Processes

Abstract

In this paper, a hybrid classification method for image based anomaly detection is proposed to improve the detection rate from industrial high-dimensional process data. The method involves feature selection with clustering based classification to discover failure patterns for marginal datasets to improve detection accuracy. The proposed hybrid classification method is tested with a real industry data sets. Results show that the proposed hybrid classification method is superior to the conventional classification methods such as multilayer perceptron (MLP) and decision tree in term of anomaly detection accuracy.