An Explainable Artificial Intelligence Framework for Analyzing Health Degradation of Aero-Engines

航空发动机作为复杂且关键的机械设备，其健康状态直接影响到飞行安全与运营效率。鉴于目前健康状态监测技术在可解释性和准确性上的局限性，本文提出了一种基于Shapley值的可解释人工智能分析框架。该框架通过解析机器学习模型的预测结果，揭示了发动机健康衰退的根本原因。首先，运用Shapley值分析法评估各传感器数据特征的边际贡献，识别了影响航空发动机健康的关键因素及其衰退路径。接着，应用HDBSCAN聚类算法识别潜在故障风险的发动机。最后，利用SkopeRules从聚类中提取易于理解且准确度高的规则集合，作为复杂模型决策的替代方案，增强了模型输出结果的透明度和可靠性。实验结果表明，该框架能够揭示航空发动机健康衰退的演变路径，为实现精准维护提供可靠的理论支撑。As a complex and critical mechanical equipment, the health status of aircraft engines directly affects flight safety and operational efficiency. Given the limitations of current health status monitoring technologies in terms of interpretability and accuracy, this paper proposes an interpretable artificial intelligence analysis framework based on Shapley values. This framework reveals the fundamental cause of engine health decline by analyzing the prediction results of machine learning models. Firstly, the Shapley value analysis method was used to evaluate the marginal contribution of each sensor data feature, identifying the key factors affecting the health of aircraft engines and their degradation paths. Next, the HDBSCAN clustering algorithm is applied to identify engines with potential fault risks. Finally, SkopeRules is used to extract a set of rules from clustering that are easy to understand and highly accurate, as an alternative to complex model decision-making, enhancing the transparency and reliability of the model output results. The experimental results indicate that the framework can reveal the evolutionary path of health decline in aircraft engines, providing reliable theoretical support for achieving precise maintenance.