基于机器视觉的3D打印异常诊断方法

Abnormal Diagnosis Method of 3D Printing Based on Machine Vision

为解决3D打印过程中出现的诸如堵头、断丝、翘曲等异常情况导致打印失败的问题,搭建检测平台并提出一种融合Xception的改进YOLOv5算法,完成异常实时检测,达到及时处理、提高打印成功率的目的。通过对YOLO算法头部、躯干部以及瓶颈块进行轻量化改进,提高识别帧率并减小参量;然后对输出部分进行改进,使特征相似的异常图像被收集后输入至Xception算法中,提升异常识别分类的准确率;最后利用Qt跨平台开发框架设计打印异常诊断系统人机交互界面软件。结果表明:改进的融合算法在自建3D打印异常数据集中识别准确率为88.75%,较原YOLOv5算法提高3.22%,同时识别平均帧率为28帧/s,提高了40.0%,可以满足实际打印中对识别准确率及实时性的要求。

In order to solve the problem of printing failure caused by abnormal conditions such as plug,broken wire and warping in 3D printing process,a detection platform was built and an improved YOLOv5 algorithm with Xception was proposed to complete real-time anomaly detection,achieving the goal of timely processing and printing success rate improvement.The YOLOv5 algorithm was re-constructed by improving the head,trunk and bottleneck block of YOLO algorithm,improving the identification frame rate and reducing the parametes.Then the output part was improved so that the abnormal images with similar features were collected and input into Xcep-tion algorithm to improve the accuracy of abnormal recognition and classification.Finally,the Qt cross-platform development framework was used to design a printing abnormal diagnostic system human-computer interaction interface software.The results show that the accu-racy rate of the improved fusion algorithms in self-built 3D printing abnormal data set recognition is 88.75%,which is 3.22%higher than the original YOLOv5 algorithm,and the average recognition frame rate is 28 f/s,which is increased by 40.0%.It can meet the actu-al printing recognition accuracy and real-time requirements.