Thrust estimation in limited ground test data scenarios: A digital twin-driven method for gas turbines with performance variability

The accuracy of thrust estimators decreases in practical applications due to performance variations among individual gas turbine engines. Correcting the thrust estimator is challenging due to the limited availability of engine test data. To address this issue, this study proposes a digital twin-driven thrust estimation framework, which comprises an individual performance difference identification (IPDI) module, a model deduction module, and a fine-tuning module. The IPDI module identifies the individual difference characteristics of the target engine, which are quantified as the individual quantification parameter (IQP). First, the IPDI module uses the underdetermined feature expansion layer to expand available features. Subsequently, it estimates the IQPs using a physics-informed multi-layer perceptron (PIMLP). Finally, the perturbation-based post-adjustment layer refines the PIMLP's output by integrating all available ground test data. Using the estimated IQPs, the model deduction module generates virtual full-envelope data for the target engine, which is then used to fine-tune the benchmark thrust estimator. Simulation results indicate that the fine-tuned thrust estimator achieves a mean relative error of 0.0548 %, compared to 0.9741 % for the benchmark estimator. Micro turbojet engine experiments demonstrate that the fine-tuned estimator achieves a mean relative error of 0.9117 %, significantly lower than 6.1164 % of the benchmark estimator.