Statistical process control using optimized neural networks: a case study

Area of Science:

• Industrial Engineering
• Machine Learning
• Statistical Process Control

Background:

• Control charts are essential statistical process control (SPC) tools for detecting process changes.
• Recognizing control chart patterns (CCPs) accurately is crucial for effective process monitoring and intervention.

Purpose of the Study:

• To design an accurate and efficient system for control chart pattern (CCP) recognition.
• To investigate and select the best neural network classifier for CCP recognition.
• To enhance the generalization performance of the classifier using a hybrid heuristic approach.

Main Methods:

• Developed an efficient recognition system with feature extraction (shape and statistical features) and classifier modules.
• Investigated several neural networks (MLP, PNN, RBF) to identify the optimal classifier.
• Implemented a hybrid heuristic system using the cuckoo optimization algorithm (COA) to improve classifier performance.

Main Results:

• The proposed system achieved high recognition accuracy for control chart patterns.
• The feature extraction module effectively identified discriminative characteristics of CCPs.
• The cuckoo optimization algorithm enhanced the generalization capabilities of the chosen neural network classifier.

Conclusions:

• The developed system provides an accurate method for CCP recognition.
• The hybrid heuristic approach, integrating COA, offers superior performance in identifying process deviations.
• This research contributes to more reliable statistical process control through advanced pattern recognition.