Recognition of Control Chart Patterns Using Feature-Based Artificial Neural Network Approach

Control charts usually exhibit one of the eight types of patterns. These patterns can be classified as normal and abnormal. Recognition of abnormal patterns in control charts can provide clues to reveal potential quality problems in the manufacturing processes. Neural network approaches (with features extracted from the pattern data as input vector representation) have been successfully applied by the researchers in recent years for recognition of control chart patterns. Usage of features leads to smaller network size and results in faster training and generally more effective and efficient recognition of control chan patterns. The reported feature-based approaches can only recognize six principal control chart patterns (CCPs). In this paper a new set of features is proposed and a multilayered perceptron (MLP) neural network trained by back-propagation algorithm is presented that can recognize stratification and systematic patterns in addition to the other six patterns as mentioned above. Extensive performance evaluation of the developed pattern recognizer is carried out using simulated data. Numerical results indicate that the artificial neurai network based pattern recognizer developed using the proposed set of features can perform well in real time process control applications.