PPO-PF: An aero-engine performance degradation state calibration algorithm for turbine blade service loadings assessment

To address the problem of calibrating the degraded state of engine components under the condition of limited number of sensors, this paper proposes a calibration framework based on deep reinforcement learning and particle filtering. Firstly, the degraded state is inherited from the previous flight cycle through imitation learning, and combined with an improved PPO algorithm to output the a priori distribution of the degraded state; then a filtering algorithm generates a posteriori distributions that consider model error; finally, an improved reward and advantage function guides the agent to calibrate health parameters online. Ablation experiments show a maximum inference error of performance degradation parameters is only 0.048 in multiple failure modes, comprehensive consideration of model and sensor uncertainties, and enhanced robustness to performance models with varying accuracies. In addition, the full process from on-board sensor signals to turbine blade load assessment is demonstrated on turbine-failure flight data, capturing degradation impacts with an average assessment time of 46.13 s. It is demonstrated that this fusion strategy can link complete engine health monitoring with structural-load monitoring of critical components, supporting refined damage assessment of these components during the operation and maintenance phase.