Related Experiment Video
Updated: Jun 4, 2026
05:47
Simulation of a Scaled Assembly Process with Collaboration of a Robotic Arm and Monitoring through a Vision System for Quality Control
Published on: August 29, 2025
A process-guided uncertainty-aware deep learning framework for reliable and interpretable industrial fault diagnosis
Babar Hayat1, Shabeer Ahmad2, Muhammad Asfandyar Shahid3
1School of Information Engineering, Xi'an Eurasia University, Xi'an, Shaanxi, China.
Plos One
|June 2, 2026
Summary
This study introduces an advanced fault detection framework (SAU-PGA-CNN-BiLSTM) that enhances industrial process safety and efficiency by intelligently using sensor reliability and process structure for accurate fault diagnosis.
Area of Science:
- Industrial Process Engineering
- Machine Learning for Process Control
Background:
- Effective fault detection is crucial for industrial safety, quality, and energy efficiency.
- Existing methods often lack robustness due to insufficient use of process structure and sensor reliability.
Purpose of the Study:
- To develop an improved fault diagnosis framework (SAU-PGA-CNN-BiLSTM) that addresses limitations of current methods.
- To enhance the robustness and practical applicability of fault detection in industrial processes.
Main Methods:
- Coupling Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) for temporal and spatial data analysis.
- Implementing a process-guided, sensor-aware attention mechanism incorporating process centrality and sensor reliability.
- Utilizing Monte Carlo dropout with sensor prior-conditioning for calibrated confidence estimates.
Main Results:
- Achieved 93.6% multiclass diagnosis accuracy and 94.0% F1 score on the Tennessee Eastman Process benchmark.
- Demonstrated improved model calibration with reduced negative log-likelihood, Brier score, and expected calibration error.
- Attention visualizations confirmed focus on process-relevant and reliable sensors.
Conclusions:
- The SAU-PGA-CNN-BiLSTM framework offers a robust and reliable solution for industrial fault detection and diagnosis.
- The sensor-aware attention mechanism significantly improves model performance and interpretability.
- This approach supports enhanced safety, quality, and efficiency in advanced industrial operations.