实验室力值类设备故障预测及算法比较  

Predictive Analysis and Algorithmic Comparison for Faults in Laboratory Force-Measuring Equipment

作　　者：张雪梅[1] 孔祥吉 ZHANG Xuemei;KONG Xiangji(Jiangsu Institute of Metrology,Nanjing 210023,China;Nanjing Institute of Environmental Science,Ministry of Ecology and Environment of the People’s Republic of China,Key Laboratory of Pesticide Environmental Assessment and Pollution Control,Ministry of Ecology and Environment of the People’s Republic of China,Nanjing,210042,China)

机构地区：[1]江苏省计量科学研究院,南京210023 [2]生态环境部南京环境科学研究所国家环境保护农药环境评价与污染控制重点实验室,南京210042

出　　处：《计量科学与技术》2023年第8期54-60,共7页Metrology Science and Technology

基　　金：国家科技基础条件平台重点项目“国家计量基标准体系资源共享平台”(KY(2016)QT04H)。

摘　　要：为提高设备管理的数字化水平,节约实验室管理成本,建立实验室设备故障预测模型,实现设备故障数据的有效利用。选取力值类设备的运行故障数据作为分析对象,基于统计方法对各影响量进行相关性分析,分别采用RidgeCV、XGBoost、LightGBM三种回归模型对该数据集进行拟合,比较、选择适宜预测设备首次故障前时间的算法。以r2、均方误差、可解释方差和平均绝对误差为模型精度衡量指标,经网格搜索-交叉验证优化后的LightGBM算法预测精度、运行速度最优,设备已服役时间和设备原值是确定首次故障前时间最为重要的特征。通过对设备故障数据的有效统筹,结合大数据分析技术,可针对不同类型设备建立符合自身规律的故障预测模型,探索出一条实验室管理的提质增效之路。This study aims to elevate digital equipment management and reduce laboratory management costs by developing a predictive model for laboratory equipment faults,thus making effective use of fault data.Operational fault data from force-measuring equipment was selected for analysis.A correlation analysis of influencing factors was conducted using statistical methods,and three regression models—RidgeCV,XGBoost,and LightGBM—were employed to fit the dataset.These models were compared to select the most appropriate algorithm for predicting the time before the first equipment fault.Model accuracy was evaluated using r2,mean squared error,explained variance,and mean absolute error.The LightGBM algorithm,optimized through grid search and cross-validation,demonstrated the best predictive accuracy and operational speed.Key features for determining the time before the first fault included the equipment's service time and its original value.By effectively managing equipment fault data and leveraging big data analysis techniques,a tailored fault prediction model for various equipment types can be established,paving the way for enhanced laboratory management efficiency and quality.

关 键 词：计量学 力值类设备 故障预测 首次故障前时间 相关性分析 回归算法 

分 类 号：TB931[一般工业技术—计量学]