Mathematical and Algorithmic Advances in Machine Learning for Statistical Process Control: A Systematic Review.

Area of Science:

• Industrial Engineering
• Data Science
• Manufacturing Systems

Background:

• Industry 4.0 manufacturing generates complex data (high-dimensional, autocorrelated, non-stationary, imbalanced).
• Classical Statistical Process Control (SPC) methods struggle with these data characteristics.
• Machine learning (ML) offers potential solutions for advanced SPC.

Purpose of the Study:

• To systematically review and synthesize ML techniques for SPC in Industry 4.0.
• To link specific data challenges in manufacturing to appropriate ML methodologies.
• To provide guidance on selecting and deploying ML-based SPC systems.

Main Methods:

• Systematic literature review following PRISMA 2020 guidelines.
• Problem-driven synthesis categorizing ML approaches by data challenges (dimensionality, autocorrelation, imbalance).
• Review of mathematical rationales and industrial applications of representative algorithms.

Main Results:

• ML approaches for high-dimensional data include dimensionality reduction and feature selection.
• Time-series and state-space models address autocorrelated and dynamic processes.
• Cost-sensitive learning, generative modeling, and transfer learning tackle data scarcity and imbalance.

Conclusions:

• Structured guidance is provided for selecting ML techniques for complex manufacturing data.
• The review highlights open issues in interpretability, thresholding, and real-time deployment of ML-SPC.
• This work facilitates the design of reliable online monitoring pipelines for Industry 4.0.