A Reliable Data Fusion and Predictive Maintenance Framework of Industrial Internet of Things Using Explainable Artificial Intelligence: Improving Resilience and Fast Recovery in Future Manufacturing Systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

INTRODUCTION: Due to the development of the Industrial Internet of Things (IIoT), it has been possible to make the predictive maintenance of the manufacturing industry using data. Nevertheless, recent developments in manufacturing systems are prone to higher levels of disruptions caused by sensor degradation, data anomalies and equipment failures, which threaten production continuity and supply chain resilience. Conventional fault predictor models have no clear data quality evaluation and readability rendering them ineffective in aiding quick recovery. OBJECTIVES: In order to overcome these challenges, the present paper introduces TEF-Net as a reliable data fusion and predictive maintenance approach based on explainable artificial intelligence, in particular, to improve the resiliency of manufacturing systems and their ability to recover quickly in the case of data quality issues. METHODS: The method starts by building four data quality indicators (completeness, stability, consistency across time, correlation consistency), and produces sensor trust measures through TrustNet. Such trust measures are included in dynamic graph calibration and graph attention fusion to reduce low-quality data and combine multiple source characteristics. Fault probabilities are then extracted by a Transformer-GRU architecture that uses degradation-related temporal features, and explainability is provided using trust scores, attention weights, and feature contributions. RESULTS: The experimental findings prove that TEF-Net has an accuracy of 0.947, a recall of 0.932, a F1-score of 0.936, and an AUC of 0.982, which is better than comparative models. In multi-error disturbance environments which are used to mimic manufacturing disruptions in practice, TEF-Net retains an F1-score of 0.884, which is much less degraded than base approaches, which is a direct indicator of its ability to make manufacturing systems resilient. CONCLUSION: The results support the statement that TEF-Net is much more accurate, strong, and understandable in predicting IIoT maintenance compared to other models and can be considered as a reliable basis of AI when it comes to predicting failures early and recovering them quickly in the future.