中圖分類號:TQ110.5 文獻標志碼:A 文章編號:1001-5922(2025)07-0033-04
Abstract:In order to improve theaccuracyof image detection of crack welding defects of large wind turbine impellers,a detection method based on the improved YOLOv5model was proposed.Based on the YOLOv5 model,this method introduced the CBAMatention mechanism after the backbone network of the YOLOv5 model,CSPDarknet53,to enhance the learning of important features,and directlycalculated the diference between the widthand heightof the predictionboxandthereal boxto replace the distance lossof theaspectratio,soas toavoid the failure of the model to converge,and realized the improvementof the YOLOv5 model.Finaly,the improved YOLOv5 model was used to detect thecrack welding defect image ofthe impellerof the large wind turbine,and the detection accuracyof the crack welding defect image of the impellerof the large wind turbine was improved.The simulation results showed that theaverage accuracy,precision,recalland F1valueof the proposed method for the detection of impeller crack welding defect images of large wind turbines reached 96.30% , 96.77% , 94.72% and 96.27% ,respectively, which had higher accuracyand faster detection speed of 22.38 frames/s compared with the standard YOLOv5 model,CNN model,SSD model and RESNET50 model.
Key Words : wind turbines ;impeller cracks ; welding defects ;image detection ; YOLOv5 model
大型風電機組是風能發電的關鍵設備,對風能發電效率和運行狀態具有重要意義。然而由于大型風電機組的葉輪在焊接過程中,會受到焊接應力的影響,導致葉輪產生裂紋,不利于其平衡,進而影響發電效率。基于此,本研究結合YOLOv5模型在目標檢測領域的應用,如王書坤等利用輕量型YOLOv5模型,實現了對電網線路絕緣子的缺陷檢測[1];孫麗萍等以YOLOv5模型為檢測模型,實現了對林業中的有害物檢測與識別[2];豐玉華等在YOLOv5的基礎上,引入多頭自注意力,實現了對跌倒行人的檢測[3],發現YOLOv5模型具有優異的目標檢測性能。本研究選擇YOLOv5模型為基礎框架,通過在YOLOv5模型的基礎上引入注意力機制,并改進其損失函數,以提高模型的性能,提出一種改進YOLOv5的大型風電機組葉輪裂紋焊接缺陷圖像檢測方法。
1 基本算法
1.1 YOLOv5模型簡介
YOLOv5模型是一種目標檢測算法,主要包括輸入端、主干網絡、頸部網絡和輸出端四個部分[4-5]。其中,輸入端是由3個不同的輸入層組成,負責對大、中、小3種不同尺度的目標進行輸入;主干網絡使用CSPDarknet53網絡,負責對特征輸入圖像特征進行提取;頸部網絡為特征圖金字塔(FPN)網絡,可實現不同特征圖層次信息的融合;輸出端使用Focalloss損失函數解決目標檢測中類別不平衡問題,并使用非極大抑制對重疊目標框進行處理,可提高模型性能[6-7]
YOLOv5模型具有結構簡單的特點,并通過使用CSPDarknet53、FPN等技術和策略,具有較高的性能和魯棒性。本研究選用YOLOv5模型作為大型風電機組葉輪裂紋焊接缺陷圖像檢測的基本框架。……