Control chart pattern recognition using RBF neural network with new training algorithm and practical features.

Area of Science:

• Industrial Engineering
• Artificial Intelligence
• Statistical Process Control

Background:

• Statistical Process Control (SPC) relies on control charts to monitor manufacturing processes.
• Recognizing control chart patterns (CCPs) is crucial for identifying process deviations.
• Distinguishing between similar patterns like Normal, Stratification, and Systematic poses a significant challenge.

Purpose of the Study:

• To propose a novel method for enhanced control chart pattern recognition.
• To improve the accuracy of identifying eight distinct CCPs, including challenging similar patterns.
• To develop a robust system for continuous production process monitoring and control.

Main Methods:

• A four-module approach: feature extraction (shape and statistical), feature selection (Association Rules), classification (Radial Basis Function Neural Network - RBFNN), and a novel learning algorithm (Bees Algorithm).
• Utilized Association Rules (AR) for optimal feature selection from shape and statistical features.
• Employed a Bees Algorithm-based learning module to optimize RBFNN performance.

Main Results:

• The proposed RBFNN method demonstrated very good performance in recognizing eight CCPs.
• Successfully distinguished between complex patterns such as Normal, Stratification, and Systematic.
• Tested on a dataset of 1600 samples (200 per pattern) with high accuracy.

Conclusions:

• The optimized RBFNN offers a powerful and accurate solution for CCP recognition.
• This method significantly advances the capability of real-time production process monitoring.
• The approach effectively addresses limitations of previous studies by including and distinguishing more complex patterns.