Cost-Effective Failure Mode Approach for User-Reported Faults Using LSTM and Word2Vec

With the advancement of technology and industry, improving production efficiency and continuity has become both important and necessary. Today, while there are various maintenance methods for systems, predictive maintenance methods based primarily on sensor data are predominantly used. However, predictive maintenance methods are often not accessible for small and medium-sized enterprises and fail to consider user-reported problems. In this study, an LSTM-based approach is proposed to classify user-reported fault indications. The proposed approach incorporates noise reduction, balancing, and normalization processes to enable its application on datasets with insufficient and imbalanced data. Additionally, Word2Vec and GloVe methods, which possess higher generalization capabilities for limited datasets, are utilized in the classifier component. The proposed approach is trained on a dataset with a small sample size and imbalanced distribution. As a result, the performance of the proposed approach has been validated.