Hydraulic response enhancement in brake valve anomaly monitoring: an integrated hardware-in-the-loop and cyclic generative adversarial network

Low sampling rates in flight recorder data limit the development of digital twins for remote brake control valve anomaly monitoring, as they obscure internal valve dynamics. To address this issue, this study introduces a domain-adaptive, cyclic Generative Adversarial Network (GAN) that integrates physical simulations with hardware-in-the-loop (HIL) experiments to enhance valve response modeling. An equivalent solenoid valve model is established and calibrated using HIL data, with simulated samples serving as the source domain. The resulting model, named CausalStarGAN, incorporates causal relationships between valve excitation signals and pressure responses to achieve high-fidelity data migration to target domains. Through the use of a domain classifier, cycle reconstruction loss, and a causality-aware mechanism, the model aligns features while preserving macroscopic physical characteristics such as response frequency and accurately generating microscopic pressure traits including fluctuation and delay. Experimental results show that synthetic data improve the causal representation of excitation-response dynamics, which enhances model generalization under unseen conditions. With small sample sizes, the method achieves reductions in Mean Absolute Error, Root Mean Square Error, and Mean Absolute Percentage Error by 21.3%, 23.1%, and 8.8%, respectively, compared to pre-augmentation data. A case study of a surrogate digital twin using the augmented dataset demonstrates the engineering potential of the proposed approach for remote brake monitoring. The valve simulation data is available at:https://github.com/chenyin7255-eng/solenoid-valve-response-data.git.