Detecting Coal Pulverizing System Anomaly Using a Gated Recurrent Unit and Clustering.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Coal pulverizing systems are critical for thermal power generation, directly impacting safety and efficiency.
• Ensuring the reliability of these systems is vital, with prognostics and health management offering an effective approach.
• Traditional mathematical modeling for anomaly detection is challenging due to the dynamic, nonlinear, and high-dimensional nature of these systems.

Purpose of the Study:

• To propose a novel data-driven integrated framework for anomaly detection in coal pulverizing systems.
• To address the limitations of traditional modeling in complex industrial systems.
• To enhance the safety and economic viability of power generation through improved system monitoring.

Main Methods:

• Development of a neural network model using gated recurrent unit (GRU) networks to capture temporal dynamics and predict system states.
• Implementation of a novel unsupervised clustering algorithm to analyze prediction errors for anomaly identification.
• Validation of the proposed framework using a real-world industrial coal pulverizing system case study.

Main Results:

• The proposed data-driven framework successfully identified anomalies in the industrial coal pulverizing system.
• The GRU network effectively modeled the temporal characteristics of the high-dimensional system data.
• The unsupervised clustering algorithm proved effective in detecting deviations from normal operating conditions.

Conclusions:

• The integrated framework offers a robust solution for anomaly detection in complex industrial systems like coal pulverizers.
• This approach enhances the reliability and safety of thermal power generation.
• The data-driven methodology provides a viable alternative to traditional model-based techniques for system prognostics and health management.