基于深度学习的油液磨粒智能检测与分割  

Intelligent Detection and Segmentation of Wear Debris Based on Deep Learning

作　　者：任松[1] 涂歆玥 朱倩雯 李眉慷 REN Song;TU Xinyue;ZHU Qianwen;LI Meikang(State Key Laboratory of Coal Mine Disaster Dynamics and Control,Chongqing University Chongqing,400044,China)

机构地区：[1]重庆大学煤矿灾害动力学与控制国家重点实验室,重庆400044

出　　处：《振动．测试与诊断》2024年第6期1068-1075,1243,共9页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(52074048,51774057)。

摘　　要：针对机械系统磨损状态监测与故障诊断中油液磨粒识别难度大、时间与人力成本高等问题,提出了基于深度学习的油液磨粒智能检测与分割方法。首先,基于滤膜谱片技术制备油液磨粒谱片并采集图像,构建了含6类不同金属磨粒的优质数据集;其次,根据数据集特点与算法优缺点,搭建单阶段实例分割模型YOLACT与两阶段实例分割模型Mask-RCNN对磨粒进行智能检测与分割。实验结果表明:Mask-RCNN模型平均检测精确率为93.8%,召回率为92.7%,适用于磨损颗粒智能分析的精准检测;YOLACT模型平均检测精确率为84.7%,召回率为83.3%,检测速度快,边缘分割精细,适用于磨损颗粒快速检测与智能分割;两种模型均有效提高了油液磨粒的检测效率。In view of the difficulties in high cost of time and labor in the identification of oil wear debris in mechanical system wear state monitoring and fault diagnosis,an intelligent detection and segmentation method for oil wear debris based on deep learning is proposed.Based on the filter spectrum technology,this method prepares spectrum of oil debris to collect images,and constructe a high-quality dataset containing 6 different types of wear debris;According to the characteristics of the dataset and the advantages and disadvantages of the algorithm,the one-stage instance segmentation model YOLACT and the two-stage instance segmentation model Mask-RCNN are respectively constructed for Intelligent detection and segmentation of oil debris.The experimental results show that the average detection accuracy of the Mask-RCNN model is 93.8%,and the recall rate is 92.7%,which is suitable for the precise detection of wear debris intelligent analysis.The average detection accuracy of the YOLACT model is 84.7%,the recall rate is 83.3%,with advanced detection speed and detailed edge segmentation,which is suitable for rapid detection and intelligent segmentation of wear debris.Both models effectively improve the detection efficiency of oil wear debris.

关 键 词：深度学习 卷积神经网络 分割模型 油液磨粒分析 金属磨粒检测 

分 类 号：TH17[机械工程—机械制造及自动化] TP389.1[自动化与计算机技术—计算机系统结构]