基于一维卷积神经网络构建医用直线加速器高价值零件故障预测模型的应用效果

The Application Effect of Fault Prediction Model for High Value Parts of Medical Accelerator Based on One-Dimensional Convolutional Neural Network

目的构建医用直线加速器高价值零件故障预测模型,以实现对高价值零件故障的预判。方法选取2013年1月至2017年12月医院在用医科达Synergy医用直线加速器的60组共381个维修记录数据,按照7:3比例随机分配为训练集(42组)和测试集(18组),采用一维卷积神经网络进行二分类建模,随机选取30组数据作为验证集评估模型性能,并采用测试集数据检测模型预测效果。结果设定最大训练学习次数为120次,实际训练次数超过80次时数据趋于稳定,训练集和验证集的准确率均稳定于90%左右,测试集数据准确率均在96%以上,表明模型收敛较好。结论该模型预测医用直线加速器高价值零件的故障次数与实际情况接近,为预防性维修和保修服务采购提供了可靠的数据支持。

Objective To construct a high-value component fault prediction model for medical accelerators to achieve prediction of high-value component faults.Methods With the selection of 60 sets of 381 maintenance record data from January 2013 to December 2017 for the Elekta Synergy medical accelerator,these data were randomly assigned to the training set(42 groups)and the testing set(18 groups)in a 7:3 ratio.With a onedimensional convolutional neural network used for binary classification modeling,30 sets of data were randomly selected as the validation set to evaluate the performance of the model,and the prediction effect of the model was tested by using the test set data.Results With a maximum training and learning times of 120 times,when the actual number of training exceeds 80 times,the data tended to be stable.In addition,the accuracy of the training and validation sets remained stable at around 90%,while the accuracy of the test set data was above 96%,indicating good model convergence.Conclusion The predicted failure times of high value parts of medical accelerators were close to the actual situation,which provided reliable data support for preventive maintenance and warranty service procurement.