A Hierarchical System for Recognition of Control Chart Patterns

Monitoring and controlling the manufacturing process is crucial for high quality production. Control charts are commonly used tools which are efficiently applied in this field. Control chart patterns are divided into two categories: natural and unnatural where recognition of the type of an unnatural pattern is of a great importance in diagnostic search procedure. In this study a hybrid system based on statistical and shape features and multilayer perceptron neural network (MLPNN) is proposed. The hierarchical MLPNN system consist of three decision making layers performing a coarse to fine recognition. While in the literature, 6 patterns including normal (NOR), cyclic (CYC), increasing trend (IT), decreasing trend (DT), upward shift (US) and downward shift (DS) are mainly discussed, in this study stratification (STR) and systematic (SYS) are also included to construct a comprehensive system. Experimental results confirm that the proposed algorithm provides significant recognition accuracy comparable to other approaches.