粗糙神经网络数据挖掘技术在大型医疗设备故障预警中的应用研究  

Discussion and research of data mining technique based on RSNN in pre-alarm on the fault of large-scale medical equipment

作　　者：陈艳[1] 王琪 王佳庆 蒋佳旺 李林献[3] CHEN Yan;WANG Qi;WANG Jia-qing(Department of Equipment,The Second People’s Hospital of Huai’an,Huai’an 223002,China;不详)

机构地区：[1]淮安市第二人民医院设备科,江苏淮安223002 [2]淮安市第五人民医院设备科,江苏淮安223300 [3]无锡市妇幼保健院医学工程处,江苏无锡214002

出　　处：《中国医学装备》2020年第9期144-148,共5页China Medical Equipment

摘　　要：目的:探讨粗糙神经网络的数据挖掘技术故障预警方法,为大型医疗设备智能预警提供参考。方法:采用数据挖掘技术基本原理、粗糙集理论及其约简方法,建立粗糙神经网络,收集2017年10月至2019年5月医院呼吸机使用中的267次故障报警事件资料,制定故障因素采集方法,从设备使用环境因素、电气因素和气路因素进行数据采集,搭建粗糙神经网络,并使用训练后的粗糙神经网络搭建设备故障预警平台。结果:使用训练后的反向传播(BP)神经网络对测试集进行测试,测试集潮气量异常、空压机故障、自检报警、氧浓度偏差过大、无法送气、低压报警和密闭性无法通过识别率分别为81.7%、76.2%、89.6%、90.2%、83.6%、93.2%和82.1%,故障总识别率为85.2%;使用训练后的粗糙神经网络搭建预警平台,当预警值取0.41时医疗设备故障预警准确率为91.0%。结论:通过挖掘故障模式与故障因素的潜在关系,设备故障模式能较好的得到识别,同时可准确对各故障进行预警,为医疗设备故障监测与智能预警提供参考。Objective: To discuss the pre-alarm method on the fault of data mining technique based on rough sets neural network(RSNN) so as to provide reference for the intelligent pre-alarm of large-scale medical equipment. Methods: We adopted the basic principles of data mining technique, rough set theory and its reduction method to establish a RSNN. This article collected the data of alarm events of 267 faults of using ventilator in hospitals from October 2017 to May 2019. We formulated a collection method of failure factors, and the data collection was implemented and RSNN was established from environmental factors of using equipment, electrical factors and gas path factors. And then, the trained RSNN was used to establish pre-alarm platform for the fault of equipment. Results: The trained back propagation(BP) was used to test the test set. And the results of test indicated that recognition rates of abnormal tidal volume of test set, the fault of air compressor, self-check alarm, excessive deviation of oxygen concentration, inability to aspirate, low pressure alarm, and airtightness fault were 81.7%, 76.2%, 89.6%, 90.2%, 83.6%, 93.2% and 82.1%, respectively. The total recognition rate of fault was 85.2%. When the pre-alarm value was 0.41 in using trained RSNN to establish pre-alarm platform, the accuracy rate of pre-alarm for faults of medical equipment was 91.0%. Conclusion: The fault model of equipment can be favorably recognized by the potential relationship between the fault model of mining and the factor of fault. And it can accurately implement pre-alarm for various faults, and can provide reference for monitoring the fault of medical equipment and intelligent pre-alarm.

关 键 词：粗糙神经网络 数据挖掘 故障预警 智能预警 故障监测 

分 类 号：R197.324[医药卫生—卫生事业管理]