NEURAL NETWORK BASED DIGITAL TWIN FOR PERFORMANCE PREDICTION OF WATER BRAKE DYNAMOMETER

Water brake dynamometer, the core component of aerospace engine testing facilities, is widely used in turbine component testing and turboshaft engine testing. Since the status of the water brake dynamometer system cannot be monitored in detail, equipment maintenance is performed solely based on the operator's experience, resulting in high risks in the dynamometer equipment operation. Post-test inspections may even reveal damage to the dynamometer bearings and key components. Cavitation and other phenomena require urgent technical solutions to improve health monitoring of key equipment and experimental safety. This paper proposes a performance prediction method for water brake dynamometers based on machine learning. By conducting physical correlation analysis of key parameters, the characteristics of water brake dynamometer operation are captured. Subsequently, a performance prediction model for water brake dynamometer is built based on digital twin technology and experimental data, enabling an accurate mapping of the dynamometer's operational state. After turbine test, the digital model is verified by dynamometer operation data set. Predicted operating parameters of the digital model show that the dynamic mean error between predicted values and actual values of multiple core component temperatures is less than 1%. Considering the sensitivity of data changes, these prediction error values are acceptable, which provide valuable reference information.