| 摘要: | 本研究建立一套鑄造金屬表面瑕疵的自動光學檢測系統，藉由鑄造金屬高爾夫球桿頭一號桿的表面瑕疵進行研究，表面瑕疵包括反應氣孔、卷氣氣孔、脫皮、縮孔、凸點等五種主要瑕疵。首先，將桿頭表面依據不同曲率角度劃分21個區域，設計各區域最佳的取像角度和一套影像拍攝流程，我們的影像拍攝流程包括在左、右側的單側光源下進行照明，配合旋轉平台的旋轉角度調整以及升降平台的上下移動，使CCD能適應不同區域表面的曲率，拍攝出瑕疵特徵最明顯且最小瑕疵在6像素值以上的影像。再來利用LabVIEW來做瑕疵檢測影像處理流程的工具，透過四大步驟階段影像處理流程來達到瑕疵檢測。第一步使用開放運算來初步去除圖像中金屬上的噪點和黑點，第二步透過灰階標準差校正的方法，成功克服了影像中由於對比度變化提升瑕疵特徵而導致的亮度不均勻問題，將灰階標準差為45以上的原圖，降低至22~23之間。第三步，使用拉普拉斯濾波器消除背景雜訊並且銳化瑕疵特徵。第四步，採用了二值化分割技術，有效地將瑕疵從背景中分離出來，並利用形態學中的閉合運算，將影像中瑕疵特徵進行修補與強化，同時使用粒子濾波器將過篩率從2.74%降至1.8%。最後利用Matlab程式將劃分的不同拍攝區域的影像進行合併，將瑕疵的位置標示在一個平面坐標系統上，提供了更容易理解的視覺位置。
本研究針對每一個桿頭樣本提取了21張影像，總共有5個樣本桿頭，因此提取了105張影像，自動光學瑕疵檢測系統總共檢測到了944個瑕疵，在將辨識結果與人眼檢測結果進行比對之後，漏篩的瑕疵數量為0個，表示我們的瑕疵檢測系統成功辨識了所有人眼辨識的瑕疵，過篩的瑕疵數量為17個，過篩率為1.8%，檢測成功率為98.2%，證實此自動光學檢測系統能有效對鑄造金屬表面瑕疵進行檢測。
;This study develops an automatic optical detection system for cast metal surface flaws and investigates surface defects on the cast metal golf club head. Surface defects include reaction pores, air entrainment pores, peeling, shrinkage holes, lumps, and so on. Five main flaws. First, the club head′s surface is separated into 21 parts based on different curvature angles, the best imaging angles for each area are determined, and an image shooting technique is developed. Our image-capturing approach incorporates unilateral light sources on the left and right sides. Illumination, combined with the adjustment of the rotation angle of the rotating platform and the up and down movement of the lifting platform, allows the CCD to adapt to the curvature of the surface in different areas and capture images with the most obvious defect features as well as the smallest defect above 6 pixels. Next, use LabVIEW as a tool for image processing to find defects using a four-step procedure. The first phase involves using open operations to eliminate the noise and dark spots on the metal in the image. The second stage employs the grayscale standard deviation correction approach to successfully overcome the uneven brightness problem in the image produced by contrast variations that accentuate fault features, decreasing the original image with a grayscale standard deviation of over 45 to between 22 and 23. The third step involves using the Laplacian filter to reduce background noise and sharpen fault details. The fourth phase employs binary segmentation technology to successfully distinguish faults from the background, followed by the closure operation in morphology to fix and enhance the image′s defect features. At the same time, particle filters are utilized to minimize the screening frequency. decreased from 2.74% to 1.8%. Finally, the Matlab application is used to combine photos from several shooting regions and identify the location of the defect on a plane coordinate system, resulting in a more intelligible visual position.
This investigation obtained 21 pictures for each club head sample. There were 5 sample club heads in all, therefore 105 pictures were obtained. The automatic optical defect detection system identified 944 faults. The identification findings were compared to human eye detection results. After comparison, the number of faults overlooked by the screen was zero, indicating that our defect detection method correctly identified all errors seen by the human eye. There were 17 faults examined, with a 1.8% screening rate and a 98.2% detection success rate. It has been confirmed that our automatic optical inspection system can successfully detect faults on the surface of cast metal. |
