基于数据驱动的涡轮发动机剩余寿命预测

Prediction of remaining life of turbine engine based on data drive

涡轮发动机系统工况复杂,经常在极端环境下工作,容易发生故障造成不可挽回的损失。通过建立物理模型的方法来进行寿命预测不仅建模困难、适用性不好,而且十分依赖先验知识。为建立适用于高维度特征的发动机剩余寿命预测模型,以及更加合理的对发动机的剩余寿命进行预测,使用改进后的梯度提升决策树(GBDT)和进行归一化处理后的涡轮发动机性能数据进行实验。在通用数据集上开展测试比较,结果表明改进GBDT模型适用于不同工况下涡轮发动机的剩余使用寿命预测,预测效果优于现有支持向量回归(SVR)、卷积神经网络(CNN)、长短期记忆网络等方法(LSTM),尤其是在运行时间上有较大幅度的提升,对于涡轮发动机的健康管理与运维决策能够提供保证。

Turbine engine systems have complex operating conditions and often work in extreme environments.They are prone to fail and cause irreparable losses.The prediction method based on physical model relies too much on prior knowledge,which makes it difficult to establish the model and has poor applicability.In order to establish an engine remaining life prediction model suitable for high-dimensional features and predict the engine remaining life more reasonably,the improved gradient lifting decision tree(GBDT)and normalized turbine engine performance data were used for experiments.The results show that the improved GBDT model is suitable for predicting the residual service life of aircraft engines under different operating conditions.The prediction effect is better than existing support vector regression(SVR),convolution neural network(CNN),Long Short-Term memory neural network(LSTM),etc.In particular,the running time has been greatly improved.It can provide guarantee for health management and operation and maintenance decisions of turbine engine.