基于Mask R-CNN网络的磨损颗粒智能识别与应用

针对设备磨损故障诊断中磨粒识别技术难度高、工作主观经验影响大等问题，采用深度学习技术开展了磨粒智能识别的研究，提出了基于Mask R-CNN卷积神经网络的磨粒数字化表征方法. 该方法利用迁移学习训练基于Mask R-CNN网络的磨粒识别模型对图像中磨粒进行识别和实例分割，然后使用Suzuki85算法、迭代算法、等比例计算方法计算出磨粒的真实尺寸，解决了磨粒分析中难定量分析的问题. 结果表明：基于Mask R-CNN网络(采用R-101-FPN骨干网络)训练的磨粒识别模型可以对图像中多个异常磨损颗粒进行识别，综合准确率和召回率达到当前图像识别领域的主流水平. 辅以上述Suzuki85等算法，成功实现磨粒图像的定量评价分析，对促进设备故障诊断技术的自动化发展和工业应用具有一定的实际应用价值.

To address the challenges of high technical difficulty in wear debris recognition and great influence of subjective experience in equipment wear fault diagnosis, this study conducts research on intelligent wear debris recognition using deep learning technology, and proposes a digital characterization method for wear debris based on the Mask R-CNN convolutional neural network. This method uses transfer learning to train a wear debris recognition model based on the Mask R-CNN network to identify wear debris and perform instance segmentation on them in images. Then, the Suzuki85 algorithm, iterative algorithm and equal-proportion calculation method are used to calculate the real size of wear debris, solving the problem of difficult quantitative analysis in wear debris analysis. The results show that the wear debris recognition model trained based on the Mask R-CNN network (adopting the R-101-FPN backbone network) can recognize multiple abnormal wear particles in images, and its comprehensive accuracy and recall rate reach the mainstream level in the current field of image recognition. Combined with the aforementioned algorithms including Suzuki85, quantitative evaluation and analysis of wear debris images are successfully realized, which has certain practical application value for promoting the automated development and industrial application of equipment fault diagnosis technology.

搜集汇总

背景与挑战

背景概述

该数据集基于Mask R-CNN网络，专注于磨损颗粒的智能识别与定量分析，解决了磨粒识别中的技术难题，并实现了高准确率和召回率。数据集由广州机械科学研究院有限公司的研究团队开发，适用于摩擦学领域的研究和应用。