基于多标签集成学习的螺旋CT机故障诊断研究  

Research on Fault Diagnosis of Spiral CT Machine Based on Multi Label Ensemble Learning

作　　者：闫小如[1,2] YAN Xiaoru(Department of Clinical Medical Engineering,Lianyungang First People's Hospital,Lianyungang 222000,China;Colleage of Computer,Central South University,Changsha 410083,China)

机构地区：[1]连云港市第一人民医院临床医学工程部,江苏连云港222000 [2]中南大学计算机学院,长沙410083

出　　处：《计算机测量与控制》2024年第11期48-55,共8页Computer Measurement &Control

摘　　要：医学应用领域计算机X线断层摄影螺旋机由于复杂的结构和较高的集成度在实际故障定位和检测中具有极高的难度;为解决这个问题,研究对螺旋CT机故障定位与检测问题进行了分析,提出一种多标签集成学习方法;该方法采用了折半查找算法获取螺旋CT机的故障数据,同时有效结合现有的卷积神经网络和循环神经网络的文本表征网络,通过自适应标签关系增强方法找出标签间的依赖关系,并利用加权约简标签集的不平衡学习能有效杜绝模型可扩展性低和模型泛化性弱等问题;经损失值、准确度、运行时间、精准率、灵敏度5个指标的实例测试结果表明,研究所给出的方法均相对于其他3种较为创新的多标签集成学习方法更具优势,且提升数值均超过2%,训练集的各个指标数据均比测试集相应数值更高;训练集和测试集中空时网络聚类约简的多标签集成学习方法的精准率分别为93.12%和87.26%,召回率分别为86.35%和84.25%;该方法能精准快速查找螺旋CT机的故障类型和故障部位,极大程度降低维修成本和延长设备的使用年限。Computed tomography camera(CT)spiral machines in the field of medical applications face extremely high difficulties in actual fault localization and detection due to their complex structure and high integration.To address this issue,an analysis was conducted on the fault localization and detection of CT spiral machines,and a multi label ensemble learning method was proposed.This method uses a half search algorithm to obtain fault data of CT spiral machines,while effectively combining existing convolutional neural networks and recurrent neural networks for text representation.Through an adaptive label relationship enhancement method,the dependency relationships between the labels are identified,and the imbalanced learning of weighted reduction label sets can effectively eliminate problems such as low model scalability and weak model generalization.The test results of five indicators,including loss value,accuracy,running time,accuracy,and sensitivity,show that the proposed methods have more advantages over the other three innovative multi label ensemble learning methods,and the improvement values all exceed 2%.The various indicators of the training set are higher than those of the test set.The accuracy of the multi label ensemble learning method for spatiotemporal network clustering reduction in the training set and test set is 93.12%and 87.26%,respectively,with the recall rates of 86.35%and 84.25%.This method can accurately and quickly identify the types and locations of faults in CT spiral machines,greatly reducing maintenance costs and extending the service life of equipment.

关 键 词：螺旋CT机 多标签集成学习 故障检测 折半查找算法 空时网络聚类约简 

分 类 号：TH77[机械工程—仪器科学与技术]