Long Horizon Transformer Quantile Fault Prediction for Multi Site Industrial Predictive Maintenance

The researchers proposed a quantization fault prediction method based on the long-range Transformer architecture, aimed at addressing the challenges of predictive maintenance in multi-site industrial scenarios. By introducing a quantile regression mechanism, the model can output the probability distribution of faults rather than a single predicted value, thereby more accurately assessing uncertainty. Experiments show that this method significantly improves early detection and prediction accuracy of equipment failures in various industrial scenarios, providing a new technical approach for building highly reliable distributed industrial predictive maintenance systems.