Natural-language-processing and safety-engineering-based fault identification technique for electrochemical ESSs

Area of Science:

• Energy Storage Systems
• Artificial Intelligence
• Safety Engineering

Background:

• Electrochemical energy storage systems (ESSs) are critical for grid stability and renewable energy integration.
• Increasing scale and complexity of ESSs heighten safety risks like thermal runaway and fire.
• Existing fault diagnosis methods struggle with the complexity and unstructured data of ESS failures.

Purpose of the Study:

• To develop an integrated fault diagnosis framework for ESSs.
• To enhance the safety, reliability, and resilience of large-scale energy storage.
• To provide real-time monitoring, intelligent diagnosis, and proactive risk management.

Main Methods:

• Compiled a global ESS fault log, augmented with synthetic data generated by large language models.
• Utilized natural language processing (NLP) for analyzing unstructured fault reports.
• Employed Bow-Tie and failure mode analyses to identify fault pathways and risk factors.
• Developed a self-attention augmented convolutional neural network (CNN) with dynamic learning rate for fault classification.

Main Results:

• Achieved 94.93% accuracy and a macro F1-score of 0.9427 in fault classification, outperforming benchmarks.
• Successfully linked identified faults to comprehensive process solutions, including prevention and emergency response.
• Revealed hidden associations among fault categories using keyword networks and hierarchical clustering.

Conclusions:

• The proposed framework offers a robust and practical pathway for intelligent fault diagnosis in ESSs.
• The integrated approach enhances system resilience by reducing downtime and enabling proactive risk management.
• Actionable insights derived from data analysis support targeted preventive strategies for ESS safety.