Fault diagnosis method for hydraulic robots based on digital twin and Transformer-LSTM-XGBoost

Hydraulic robots are increasingly used in industrial systems, which raises the need for reliable health diagnosis and maintenance under complex operating conditions. An intelligent fault diagnosis method was developed by integrating digital twin technology with deep learning and combining a Transformer, a long short-term memory networks (LSTM), and extreme gradient boosting (XGBoost). A diagnosis architecture based on digital twins was built, and cyber-physical synchronization was made possible by a digital twin with an attribute model and a three-dimensional model. Calibration improved twin accuracy and virtual-physical consistency. Fault mechanism analysis and fault evolution simulation were conducted for four typical hydraulic system faults: leakage, valve sticking, damping orifice blockage, and filter blockage. The simulations generated a dataset covering normal and fault conditions for data-driven modeling and diagnosis. A Transformer-LSTM feature extractor captured global dependencies and temporal dynamics in multi-dimensional time-series data, and XGBoost performed multi-fault classification. The experimental verification results show that the proposed method demonstrate steady performance under various noise disturbances and 96.6% diagnostic accuracy across many problems, suggesting high resilience and generalization and supporting hydraulic robot fault detection and maintenance in industrial applications.