摘要
目的:基于文本挖掘技术深度解析内窥镜维修数据,构建维修等级预测模型,识别关键特征以优化维修管理决策。方法:收集2005年至2020年西南地区19 676例奥林巴斯内窥镜维修数据,采用Jieba分词-TF-IDF联合特征提取法处理故障文本,基于Xgboost算法建立维修等级分类预测模型,结合特征选择算法进行关键特征识别。结果:该模型预测准确率达90%,受试者工作特征曲线下面积(AUC)值为0.85,灵敏度为76%,特异度为95%,F1值为80%。重要度评分前10的特征分别为:图像异常、磨损、漏水、插入管、设备类别、导光软管、按钮、角度故障、CCD玻璃、橡皮。结论:基于文本挖掘的内窥镜维修等级预测模型可精准预测内窥镜维修的等级。建议工程师重点关注内窥镜插入管、CCD玻璃等关键部件的预防性维护,降低维修成本,提升设备使用效能。
Objective:To conduct an in-depth analysis of endoscope maintenance data using text mining techniques,establish a maintenance level prediction model,and identify key features to optimize maintenance management decisions.Methods:A total of 19676 maintenance data of Olympus endoscope from 2005 to 2020 in Southwest China were collected.The Jieba segmentation-TF-IDF combined feature extraction was adopted to process fault text.A classification model about maintenance grade was constructed on the basis of Xgboost algorithm.And then,the feature selection algorithm was combined to identify key feature.Results:The model achieved an accuracy of 90%,AUC of 0.85,sensitivity of 76%,specificity of 95%,and F1-score of 0.80.The top 10 features ranked by importance were:image abnormality,wear,leakage,insertion tube,device category,light guide tube,button,angle malfunction,CCD glass,and rubber.Conclusion:The text mining-based prediction model for maintenance grade of endoscope can accurately predict the grade of repair.The paper provides suggestions for engineers to pay key attention to preventive maintenance for insertion tube,CCD glass of endoscope,so as to reduce cost and enhance usage efficiency of equipment.
作者
卓义轩
杜江
杨罗宽
叶朋鑫
刘麒麟
Zhuo Yixuan;Du Jiang;Yang Luokuan;Ye Pengxin;Liu Qilin(Department of Medical Engineering,West China Hospital,Sichuan University,Chengdu 610041,China;Digestive Endoscopy Center,West China Hospital,Sichuan University,,Chengdu 610041,China)
出处
《中国医学装备》
2025年第9期155-157,162,共4页
China Medical Equipment
关键词
电子内窥镜
文本挖掘
机器学习
特征选择
预测模型
Electronic endoscope
Text mining
Machine learning
Selection of feature
Prediction model
