Control chart pattern recognition under small shifts based on multi-scale weighted ordinal pattern and ensemble classifier

Production orientated toward high-quality products makes the manufacturing process highly accurate and precise with minimal variations in its parameters. Accurately identifying small variations in critical control points is extremely important in quality control. The existing control chart pattern recognition (CCPR) methods usually focus on the moderate and large shifts to detect anomalies of pattern parameters, yet ignore the effects of small shifts. Moreover, traditional features only consider the magnitude characteristics of the control chart pattern (CCP) data, and ignore all other characteristics. To solve these problems, a new feature and multi-scale ensemble classification model is proposed in this paper. First, the ordinal pattern (OP) features based on the sequential characteristics of CCP data and the weighted OP (WOP) features combining sequential and magnitude characteristics are proposed. Second, an ensemble classification model (WOP-EC) based on multi-scale WOP features is constructed. Simulation results show that the correct recognition rates of WOP-EC are 94.31% and 99.88% under small and large shifts, respectively. Compared with other traditional features, WOP-EC has better recognition performance under different window sizes and separability levels and can greatly reduce the two types of error.