Hybrid Lexical and Machine Learning Feature Framework for Automated Classification of Industrial Equipment Alarm Logs

I. INTRODUCTION

Manually triaging free-text equipment alarm logs lags true fault onset by minutes to hours, during which maintenance teams remain unaware of developing issues, motivating classification approaches that assess a log message structural and metadata characteristics without needing deep domain parsing.

II. METHODOLOGY

A dataset of 180,000 alarm log entries, balanced between confirmed critical faults and routine notifications, was processed to extract 27 lexical features including message length, token entropy, presence of numeric fault codes, and character n-gram statistics, combined with equipment-metadata features such as machine age and maintenance-history frequency, and used to train a random forest classifier with 300 trees.

III. RESULTS AND EVALUATION

The classifier achieved 97.4 percent accuracy and an F1-score of 0.973 on a held-out test set, with a false-alarm misclassification rate of 1.1 percent, and equipment age together with message-token entropy ranked as the two most important features by mean decrease in impurity. Average per-log inference time was under 5 milliseconds on commodity hardware.

IV. CONCLUSION

Lexical and metadata-based features enable fast, structure-independent alarm log classification suitable for real-time maintenance dashboards. Future work will evaluate robustness against inconsistently formatted log entries from legacy equipment.

V. REFERENCES