High Dimensional Process Monitoring Using Robust Sparse Probabilistic Principal Component Analysis

TL;DR

This paper introduces a robust sparse probabilistic PCA method for high-dimensional process monitoring, improving dimensionality reduction and interpretability in complex manufacturing data.

Contribution

It develops a novel Bayesian variational inference-based approach for robust sparse probabilistic PCA tailored for correlated high-dimensional process data.

Findings

Effective dimensionality reduction while maintaining interpretability

Successful change detection in Raman spectroscopy data

Validated robustness and efficiency through simulations

Abstract

High dimensional data has introduced challenges that are difficult to address when attempting to implement classical approaches of statistical process control. This has made it a topic of interest for research due in recent years. However, in many cases, data sets have underlying structures, such as in advanced manufacturing systems. If extracted correctly, efficient methods for process control can be developed. This paper proposes a robust sparse dimensionality reduction approach for correlated high-dimensional process monitoring to address the aforementioned issues. The developed monitoring technique uses robust sparse probabilistic PCA to reduce the dimensionality of the data stream while retaining interpretability. The proposed methodology utilizes Bayesian variational inference to obtain the estimates of a probabilistic representation of PCA. Simulation studies were conducted to verify the efficacy of the proposed methodology. Furthermore, we conducted a case study for change detection for in-line Raman spectroscopy to validate the efficiency of our proposed method in a practical scenario.