Leveraging Active Learning for Failure Mode Acquisition.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Identifying failure modes is crucial for product reliability and predictive maintenance sensor selection.
• Current methods rely on experts or simulations, which are resource-intensive.
• Automating failure mode identification from maintenance records using Natural Language Processing (NLP) faces challenges due to data quality and tool maturity.

Purpose of the Study:

• To propose and evaluate a framework using online active learning for identifying failure modes from maintenance records.
• To demonstrate the efficiency of a semi-supervised machine learning approach with human-in-the-loop annotation compared to unsupervised methods.

Main Methods:

• Developed a framework utilizing online active learning for semi-supervised machine learning.
• Incorporated human annotation to train the model on a subset of maintenance records.
• Evaluated the model's performance in identifying failure modes from unstructured maintenance data.

Main Results:

• The active learning model achieved 90% accuracy and an F-1 score of 0.89 in identifying failure modes.
• Effective model training was accomplished by annotating less than ten percent of the total available data.
• The framework's effectiveness was validated through both qualitative and quantitative analyses.

Conclusions:

• The proposed active learning framework offers an efficient and accurate method for extracting failure modes from maintenance records.
• Human-in-the-loop active learning significantly outperforms unsupervised methods for this task, requiring minimal data annotation.
• This approach addresses key challenges in NLP-based failure mode identification, enhancing product design and predictive maintenance strategies.