Fault Diagnosis of Elevator Based on Adaptive Transfer Network

The early fault diagnosis of elevators is crucial for passenger safety and the riding experience. If a fault occurs but is not detected in time, it may lead to serious accidents. However, with a wide variety of elevator specifications and models, and highly variable operating conditions, constructing a generalizable diagnostic model presents a significant challenge. Moreover, obtaining labeled fault data is difficult, which greatly increases the complexity of early fault diagnosis and poses a challenge to ensuring the safe operation of elevators. To address this issue, a novel adaptive transfer network model for elevator fault diagnosis is proposed. The model is designed to process vibration data during elevator operation, thereby achieving better adaptation to equipment and working condition domains under conditions of label scarcity and sample imbalance. The model extracts and fuses rich vibration signal features from multiple sensors.. Additionally, cross-attention based on the exchange of classification information at different scales is designed to complement multi-scale fault features between the two branches. Finally, through fine-tuning of the model parameters, fault classification of target domain samples is achieved.