改进的军用飞机涡轮叶片预测性维修方法

An Improved Predictive Maintenance Method of Military Aircraft Turbine Blade

为了对军用飞机发动机涡轮叶片进行维修性预测,论文首次提出多变量分析与随机森林算法结合的预测方法。首先,利用多变量的方法进行归一化及变量选取,将待预测时刻最近一段时间的预测值与实际值构成融合矩阵,并与增量学习结合进行二次学习。然后,通过多变量与维修频率相关性分析,对结果进行排序筛选出特征值;接下来,构建融合矩阵,将训练数据集和测试数据集输入到融合矩阵决策模型中进行预测,得到预测结果。最后,实验结果显示:几种不同方案的平均绝对百分误差(MPEG)均大幅下降。结果表明,论文方法能够有效提高预测精度,对于保障飞行安全、降低叶片损伤和报废率,从而降低维护成本有着重大意义。

In order to carry out maintenance prediction on the turbine blades of military aircraft,this paper first proposes a prediction method combining multivariate and random forest algorithms.Firstly,the multivariate method is used for normalization and variable selection.The predicted value and the actual value are recently used to be predicted constitute a fusion matrix,and are combined with incremental learning for secondary learning.Then,the correlation analysis is performed on the multivariate and the maintenance frequency,and the results are sorted to select the characteristic values.Next,the fusion matrix is constructed,and the training data set and the test data set are input into the fusion matrix decision model for prediction,and the result is obtained.The experimental results show that the average absolute percentage error(MPEG)of several different schemes has dropped significantly.This method can effectively improve the prediction accuracy,which is of great significance for ensuring flight safety,reducing blade damage and scrap rate,and reducing maintenance costs.