應用YOLO模型於焊道系統品質改善-以W公司鋼捲退火酸洗線電焊機為例

產業界不銹鋼冷軋廠中的連續退火酸洗線(Annealing Pickling Line；APL)，為了達成不間斷的連續生產模式，皆設立一台電弧焊接機，作為鋼捲對鋼捲之間的連接。焊接過程中時常因為焊接機參數調整不良，造成焊槍位置偏移、能量不穩定等狀況，產生焊道缺陷。常見缺陷類別如焊道偏移、焊道破孔、焊道滲透不良等，這些缺陷使得焊道強度減弱，導致斷帶進而迫使產線停機。現階段產線電焊人員以目視進行焊道品質管控，決定放行與否，然而視線不佳與視覺疲勞所發生的人為誤判為不可避免的因素，存在漏檢之風險。為改善焊道品質不良所造成的風險，本研究提出YOLOv4(You Only Look Once；YOLO)物件偵測型深度學習應用於自動焊道缺陷檢測與分類。透過網路攝影機記錄焊道表面品質，經影像處理與資料補強(Data Augmentation；DA)後，以YOLOv4技術建立焊道缺陷篩檢與焊道缺陷分類兩種模型，其中焊道缺陷篩檢模型用來檢查該焊道是否存在缺陷，若檢出缺陷則由焊道缺陷分類模型負責判斷缺陷種類。最後整合至後端資訊資料庫系統，並達到異常警示目的，提醒電焊人員焊道品質異常，參數調整重新焊接至符合規定後才可放行，以降低漏檢風險，達焊道品質改善之目的。研究成果表明，焊道缺陷篩檢模型異常召回率達95.80%、精確率達100%。焊道缺陷分類模型中，焊道偏移召回率100%、精確率100%；焊道破孔召回率31.74%、精確率89.66%；焊道滲透不良召回率94.69%、精確率100%。由此可知，焊道缺陷篩檢模型已具備提升焊道品質管控能力；焊道缺陷分類模型部分，除破孔類別表現不理想，其餘缺陷亦可達到自動分類標準。因此，YOLOv4導入自動焊道缺陷篩檢可以獲得不錯效果，為鋼捲製造業者大幅降低焊接不良所產生的風險，有利於提升製程品質。

關鍵字

電弧焊接；焊道偏移；焊道破孔；焊道滲透不良； YOLOv4

並列摘要

In order to achieve the uninterrupted continuous production mode in Annealing Pickling Line(APL) in cold-rolled stainless steel plant in industry, arc welding machine has been set up for connection between steel coils. During the poor setting of parameters in welding process, the defects generated in torch shifted and unstable energy. Common defect categories such as the welding bead shifted, welding bead hole, and poor welding bead penetration, etc. Band steel breakage by the weak welding bead strength resulting in shutting down in the processing. At the present stage, welding crew decide whether to pass or not by quality control in visual inspection. However, inevitable happened human error in judgement caused by poor vision and visual fatigue, there is a risk of missing inspection. This study proposes application of YOLOv4 (You Only Look Once ; YOLO) in automatic defect detection and classification of welding process to improve welding quality and reduce risk. The surface quality of the welding bead has been recorded by IPcamera. After image processing and data augmentation, two models of welding defect filter and welding defect classification build with YOLOv4. The model of welding defect filter used to check whether there is a defect in welding bead, then classified by model of welding defect classification. Finally, the abnormal information integrated into the back-end information database system, and reminding welding crew to adjust parameters to meet the requirements before release, then reduce risk of missing inspection and improving quality of welding bead. The results show that recall rate is 95.80% and precision rate is 100% in model of welding defect filter. In model of welding defect classification, both recall rate and precision rate are 100% in welding bead shifted, recall rate is 31.74% and precision rate is 89.66% in welding bead hole, recall rate is 94.69% and precision rate is 100% in poor welding bead penetration. Obviously, model of defect filter has ability to improve the quality control of the welding bead. In model of defect classification, got good results except for classfication in welding bead hole. Therefore, application YOLOv4 of automatic defect detection and classification show good results to help reducing risk and improving quality of the process for steel coil manufacturers.

並列關鍵字

arc welding ； welding bead shifted ； welding bead hole ； poor welding bead penetration ； YOLOv4

參考文獻

1. Y.S. Tarng, S.C. Juang, C.H. Chang. (2002). The use of grey-based Taguchi methods to determine submerged arc welding process parameters in hardfacing. Journal of Materials Processing Technology, 128(1-3), 1-6. https://doi.org/10.1016/S0924-0136(01)01261-4

Google Scholar

2. S.-K. Cho, Y.-S. Yang, K.-J. Son, J.-Y. Kim. (2004). Fatigue strength in laser welding of the lap joint. Finite Elements in Analysis and Design, 40(9-10), 1059-1070. https://doi.org/10.1016/j.finel.2003.08.010

Google Scholar

3. L.J. Yang, R.S. Chandel, M.J.Bibby. (1993). An analysis of curvilinear regression equations for modeling the submerged-arc welding process. Journal of Materials Processing Technology, 37(1-4), 601-611. https://doi.org/10.1016/0924-0136(93)90121-L

Google Scholar

4. L.J. Yang, M.J. Bibby, R.S. Chandel. (1993). Linear regression equations for modeling the submerged-arc welding process. Journal of Materials Processing Technology, 39(1-2), 33-42. https://doi.org/10.1016/0924-0136(93)90006-R

Google Scholar

5. J I Lee, S Rhee. (2014). Prediction of process parameters for gas metal arc welding by multiple regression analysis. Proceedings of the Institution of Mechanical Engineers Part B Journal of Engineering Manufacture, 214(6), 443-449. https://doi.org/10.1243/0954405001517757

Google Scholar

主題瀏覽