Process monitoring for quality - A multiple classifier system for highly unbalanced data

Area of Science:

• Data Science
• Machine Learning
• Industrial Engineering

Background:

• Big data analyses face challenges with hyper-dimensional feature spaces, necessitating diverse machine learning algorithms (MLAs).
• Multiple classifier systems (MCSs) enhance pattern recognition by combining diverse algorithms, improving robustness to high-dimensional and noisy data.
• Quality 4.0 initiatives, like process monitoring for quality, grapple with highly imbalanced datasets for defect detection.

Purpose of the Study:

• To present a novel MCS designed for analyzing imbalanced manufacturing data in Quality 4.0 contexts.
• To develop a meta-learning algorithm that identifies reliable classifiers within an MCS for improved defect detection.
• To demonstrate the superiority of the proposed MCS over existing fusion rules in identifying rare quality events.

Main Methods:

• The study employs eight well-established machine learning algorithms (MLAs).
• An ad hoc fitness function and a novel meta-learning algorithm are utilized for classifier selection and decision fusion.
• Performance is evaluated using multiple publicly available, imbalanced datasets relevant to quality monitoring.

Main Results:

• The proposed MCS demonstrates superior performance compared to extensively used fusion rules.
• The meta-learning algorithm effectively identifies and utilizes reliable classifiers for improved prediction accuracy.
• The system shows significant potential for detecting rare quality events in imbalanced big data.

Conclusions:

• The novel MCS offers a powerful solution for defect detection in Quality 4.0, particularly with imbalanced datasets.
• The meta-learning approach enhances the reliability and robustness of multiple classifier systems.
• This research advances the capabilities of Industry 4.0 in addressing critical quality monitoring challenges.