Multisensory Data-Driven Fault Prediction Model using Conformer-Attention Model

Abstract

For large-scale industrial machine the advancement of Artificial Intelligence (AI) and sensor technology have transformed predictive maintenance approaches. To overcome these problems, this paper offers an AI-Driven Predictive Maintenance Framework that uses multisensory operational data for early defect detection and performance optimization. The raw input dataset fed to data exploration, cleaning and dimensionality reduction. Significant operational features are extracted from multisource sensor inputs via feature engineering and pattern recognition, while data imbalance is addressed with the SMOTE technique. The improved dataset is then sent through a hybrid Conformer-Attention model, which combines convolutional and transformer-style attention processes to capture both local and global temporal relationships. Finally, evaluation metrics are computed to ensure model’s dependability and performance which have the accuracy, precision, recall and F1-score of 99%. This AI-driven strategy increases machinery uptime, decreases maintenance costs and allows for proactive decision-making, all of which contribute to creation of intelligent and sustainable industrial systems.