Improved SPC chart pattern recognition using statistical features

Abstract

Increasingly rapid changes and highly precise manufacturing environments require timely monitoring and intervention when deemed necessary. Traditional Statistical Process Control (SPC) charting, a popular monitoring and diagnosis tool, is being improved to be more sensitive to small changes and to include more intelligence to handle dynamic process information. Artificial neural network-based SPC chart pattern recognition schemes have been introduced by several researchers. These schemes need further improvement in terms of generalization and recognition performance. One possible approach is through improvement in data representation using features extracted from raw data. Most of the previous work in intelligent SPC used raw data as input vector representation. The literature reports limited work dealing with features, but it lacks extensive comparative studies to assess the relative performance between the two approaches. The objective of this study was to evaluate the relative performance of a feature-based SPC recognizer compared with the raw data-based recognizer. Extensive simulations were conducted using synthetic data sets. The study focused on recognition of six commonly researched SPC patterns plotted on the Shewhart X-bar chart. The ANN-based SPC pattern recognizer trained using the six selected statistical features resulted in significantly better performance and generalization compared with the raw data-based recognizer. Findings from this study can be used as guidelines in developing better SPC recognition systems.