基于RFTL-1DNet的矿用挖掘机发动机故障诊断方法  

Fault diagnosis method of mining excavator engines based on RFTL-1Dnet

作　　者：顾清华[1,2] 银璐阳子 王丹 骆家乐 GU Qinghua;YIN Luyangzi;WANG Dan;LUO Jiale(School of Resources Engineering,Xi'an University of Architecture and Technology,Xi'an 710055,China;Xi'an Key Laboratory for Intelligent Industrial Perception,Calculation and Decision,Xi'an 710055,China;School of Management,Xi'an University of Architecture and Technology,Xi'an 710055,China)

机构地区：[1]西安建筑科技大学资源工程学院,西安710055 [2]西安市智慧工业感知计算与决策重点实验室,西安710055 [3]西安建筑科技大学管理学院,西安710055

出　　处：《有色金属(矿山部分)》2025年第2期51-58,79,共9页NONFERROUS METALS(Mining Section)

基　　金：国家自然科学基金面上项目(52374135,52074205);陕西省金属矿智能开采理论及技术创新团队(2023-CX-TD-12);陕西省智能开采理论与技术创新团队高校青年创新团队;陕西省矿产资源低碳智能高效开采技术创新引智基地。

摘　　要：针对矿用挖掘机发动机故障数据集较少、诊断准确率低等问题,提出了一种基于一维卷积核、池化核的残差网络与迁移学习策略的故障诊断方法。通过随机森林(Random Forest,RF)分类器对初始数据集进行维度筛选,去除掉重要性低的特征以提高模型的学习效率和分类精度,使用筛选后的10维数据集对一维残差网络(ResNet18_1D)模型进行预训练,并保留训练结果;添加随机噪声扩充数据集,将一维残差网络训练结束参数作为迁移学习(Transfer Learning,TL)初始参数,使用扩充后数据集进行五倍交叉验证训练,保存并输出训练模型;调用训练效果最佳的模型进行测试,并输出分类结果。利用河南某矿山挖掘机发动机故障数据集对上述RFTL-1DNet模型进行诊断实验,实验结果表明,所提出方法的故障诊断性能明显优于其他方法,对矿山挖掘机发动机状态诊断精度超过99%。该模型对发动机常见故障的高分类准确度可快速诊断出维修计划外的故障。研究结果为智慧矿山设备管理提出了新方法。To address the issues of limited datasets and low diagnostic accuracy,a fault diagnosis method of mining excavator engines was proposed.The method was based on a one-dimensional convolutional kernel,pooling kernel residual network,and transfer learning strategy.A Random Forest(RF)classifier was used to perform dimension selection on the initial dataset,removing features with low importance to improve the model's learning efficiency and classification accuracy.The filtered 10-dimensional dataset was then used to pre-train a one-dimensional residual network(ResNet18_1D),and the training results were retained.Random noise was added to expand the dataset,and the parameters from the end of the one-dimensional residual network training were used as the initial parameters for transfer learning(TL).Five-fold crossvalidation training was conducted using the expanded dataset,and the trained model parameters were saved and outputted.The best-performing model was called for testing,and the classification results were output.The proposed RFTL-1DNet model was tested on a fault dataset from a mining excavator engine in Henan,and the experimental results showed that this method significantly outperformed other methods in terms of fault diagnosis performance with over 99%diagnostic accuracy for mining excavator engine states.The model had a high classification accuracy for common engine faults and could quickly diagnose faults beyond the maintenance plan.The research results propose new methods for the equipment management of smart mines.

关 键 词：矿用挖掘机发动机 故障诊断 深度学习 残差网络 迁移学习 

分 类 号：TD631[矿业工程—矿山机电] TD67[动力工程及工程热物理—动力机械及工程] TK429