Production Line Fault Diagnosis Based on Time-Lag Data Mining and Machine Learning

With the widespread application of intelligent control system, industrial production lines are advancing toward high automation. However, improving the accuracy and efficiency of auto fault diagnosis remains a critical challenge. This study proposes a novel fault diagnosis mechanism that identifies potential issues by monitoring time-lag, the duration of individual metric stagnation, and the intervals between stagnations of related metrics. First, we calculate time-lag data using the mentioned method and employ KPCA to handle nonlinear data and suppress noise. Then, key components for training and prediction are extracted to perform the classification task using CatBoost. The proposed approach achieved an average Accuracy of 97% and F1-Score of 96 % across multiple validation sets, demonstrating promising application potential.