基于卷积神经网络与滤波融合算法的某惯导系统剩余寿命预测模型建立

Establishment of Remaining Life Prediction Model for an Inertial Navigation System Based on Convolutional Neural Network and Filtering Fusion Algorithm

在对产品中具备大量运行观测性能数据的关键系统部件进行剩余寿命预测的过程中,因寿命数据稀少难以建立寿命分布模型。而对产品性能观测数据进行退化建模,传统退化过程分析模型对于产品性能观测数据适应性差导致产品寿命预测精度低、有效性弱的问题,充分挖掘部件退化数据信息,依据相关退化分析技术,基于统计模型中的滤波预测方法与机器学习技术中的回归卷积神经网络(regressive convolutional neural networks,RCNN)预测方法建立产品剩余寿命预测融合模型。融合模型结合了滤波预测模型对产品退化状态的挖掘能力、不确定表达能力与RCNN网络模型良好的数据适应性、预测的准确性,提高了产品退化数据分析的准确性及有效性,可对产品关键部件的寿命进行有效预测,为产品中具备大量运行观测数据的关键系统部件健康管理提供辅助参考。

In the process of predicting the remaining life of key system components with a large amount of operational observation performance data in the product,it is difficult to establish the life distribution model due to the scarcity of life data,and traditional degradation process analysis models have poor adaptability of product performance observation data,which leads to low accuracy and weak validity of product life prediction.Fully excavating component degradation data information,based on relevant degradation analysis techniques and the filtering prediction method in the statistical model and the regression convolutional neural network prediction method in the machine learning technology,a fusion model of product remaining life prediction is established.The fusion model combines the filtering forecasting model’s ability to mine product degradation status,the ability to express uncertainty,and the data adaptability and forecasting accuracy of the regression convolutional neural network model,which improves the accuracy and effectiveness of product degradation data analysis,and can effectively predict the life of key product components,and provides auxiliary reference for health management of key system components with large amount of operational observation data in the product.