Big Data Approach to Batch Process Monitoring: Simultaneous Fault Detection and Diagnosis Using Nonlinear Support...

Area of Science:

• Industrial Engineering
• Data Science
• Chemical Engineering

Background:

• Batch processes are critical in manufacturing, requiring robust monitoring for safety and efficiency.
• Accurate fault detection and diagnosis are essential to prevent productivity loss and ensure safe operations.
• Existing methods often struggle with high-dimensional, nonlinear data common in batch processes.

Purpose of the Study:

• To develop a novel data-driven framework for simultaneous fault detection and diagnosis in batch processes.
• To identify the most informative process measurements using a nonlinear Support Vector Machine-based feature selection algorithm.
• To evaluate the framework's performance across different time horizon approaches.

Main Methods:

• A data-driven framework utilizing nonlinear Support Vector Machine-based feature selection was developed.
• High-dimensional batch process data, including 22,200 batches and 15 fault types, was used for evaluation.
• Three time horizon approaches (one-step rolling, two-step rolling, evolving) were employed for model training.

Main Results:

• The proposed framework successfully achieved simultaneous fault detection and diagnosis.
• The two-step rolling and evolving time horizon approaches outperformed the one-step rolling method.
• The framework demonstrated effectiveness across various fault types and batch data trajectories.

Conclusions:

• The developed data-driven framework offers a promising decision support tool for online monitoring of batch processes.
• The two-step rolling and evolving time horizon strategies are recommended for improved fault detection and diagnosis.
• This approach enhances process safety and minimizes economic losses in industrial batch operations.