Intelligent Fault Diagnosis for Elevators Using Temporal Adaptive Fault Network

Abstract

Contemporary cities depend on elevators for vertical mobility in residential, commercial, and industrial buildings. However, elevator system malfunctions may cause operational interruptions, economic losses, and safety dangers, requiring advanced tools for detection. High-dimensional sensor data, temporal interdependence, and fault dataset imbalances are common problems in fault detection algorithms. These restrictions reduce fault diagnostic accuracy and reliability, especially in real-time applications. This paper presents a Temporal Adaptive Fault Network (TAFN) to overcome these issues. The system uses Temporal Convolution Layers to capture sequential dependencies, Adaptive Feature Refinement Layers to dynamically improve feature relevance, and a Fault Decision Head for correct classification. For reliable performance, the Weighted Divergence Analyzer and innovative data processing methods are used for feature selection. Experimental findings show that the TAFN model outperforms state-of-the-art fault classification approaches with an F1-score of 98.5% and an AUC of 99.3%. The model’s capacity to handle unbalanced datasets and complicated temporal patterns makes it useful in real life. The paper also proposes the Fault Temporal Sensitivity Index (FTSI) to assess fault prediction temporal consistency. The results demonstrate that TAFN may revolutionize elevator problem detection, improving reliability, downtime, and safety. This technique advances predictive maintenance tactics for critical infrastructure.