基于数据驱动的机车车轮剩余寿命预测技术

Remaining Useful Life Prediction Technology of Locomotive Wheel Based on Data-driven

为优化机车车轮运维管理,提高机车车轮剩余寿命的预测精度和稳定性,提出基于数据驱动的机车车轮剩余寿命预测模型。依托机车车轮全寿命周期的造修和运维数据,分析影响车轮服役寿命的主要因素,包括轮径、历史和环境影响因素。构建基于数据驱动的机车车轮剩余寿命预测模型,模型以轮径区间作为寿命计算单元,并引入轮径历史和环境损失率,综合考虑3类因素对寿命预测的影响。现场测试中,因某轮径区间的数据不符合计算要求,导致轮径环境损失率存在数据缺失问题,于是采用多层感知机算法对缺失项进行预测和补充。基于测试集中的车轮旋修数据,对比数据驱动预测算法和传统预测算法的预测效果。由结果可知,数据驱动预测算法的预测精度和稳定性均优于传统预测算法。

In order to optimize the O&M management of locomotive wheels and improve the prediction accuracy and stability of their remaining useful life(RUL),this paper proposes a data-driven prediction model for RUL of locomotive wheels.Relying on the manufacturing,repair and O&M data of locomotive wheels in their full life cycle,this paper analyzes the main factors affecting the service life of wheels,including wheel diameter,historical and environmental impact factors.The model is built using the wheel diameter interval as the life calculation unit,and introducing the historical and environmental loss rates of wheel diameter to comprehensively consider the influence of three types of factors on life prediction.In the field test,the missing items in the environmental loss rate of wheel diameter caused by the data of a wheel diameter interval that fails to meet the calculation requirements are predicted and supplemented by multi-layer perceptron algorithm.Based on the wheel reprofiling data in the test set,the prediction effect of data-driven prediction algorithm is compared with that of traditional prediction algorithm.The results show that the prediction accuracy and stability of data-driven prediction algorithm are better than those of traditional prediction algorithms.