球狀工件在工業上有許多用途,特別是在精密軸承方面之應用,由於軸承的品質會影響到機器運轉的狀況,因此軸承裡的球狀工件品質便顯得十分重要。傳統的球狀工件表面瑕疵檢測是採用人工目視檢驗,檢驗人員容易因長期檢測造成的疲勞或主觀認定而誤判,因此本研究提出自動化檢測系統以取代人工檢驗。本研究所提之視覺檢測系統使用不同以往的拍攝方法,透過一次拍攝可同時獲得球狀工件三個不同面之影像,每張待測影像再以不重疊之小區塊為單位進行區塊離散餘弦轉換(Block Discrete Cosine Transform, BDCT),計算DCT頻譜上不同區域之能量特徵值,後續搭配主成份分析(Principal Component Analysis, PCA)計算各主成份貢獻率,將主成份貢獻率做為支援向量機(Support Vector Machine, SVM)輸入區塊特徵值的權重值,經由SVM方法即可判斷是否為瑕疵區塊,最後使用簡單閥值切割法即可確定瑕疵位置。 實驗結果顯示本研究所提方法之瑕疵面積檢測率為90.02%,誤判率為0.26%。本研究除了進行鋼珠表面瑕疵檢測外,並將所提之方法應用在其它三種不同材質之球狀工件,實驗結果說明本研究之方法確實能適用於不同材質的球狀工件表面瑕疵檢驗,並具有不錯之檢測效果。
Spherical workpieces are widely used in various industries, especially in the applications of precision bearings. Quality of precision bearings is important in many industrial applications, because surface defects of bearings will affect the stability of machine operations. Currently, most of the inspection tasks of surface defects in spherical workpiece are conducted by human eyes. Human visual inspection is costly, time-consuming, and prone to making errors due to inspectors’ lack of experiences, eye fatigues, bad moods, and so on. Therefore, this research explores the automated defect inspection for spherical workpieces. The proposed vision system with one CCD can capture three surface images from three different ways of a spherical workpiece at one shot. The image blocks are transformed to the DCT (Discrete Cosine Transform) domain and five representative energy features of each DCT block are extracted. Then, we applied principle component analysis (PCA) to calculate contribution rates of the principle components. And, the contribution rates are used as the weights of the feature values of the blocks for the inputs of the SVM (Support Vector Machine) model. The SVM network can judge the block regions of the surface defects in a spherical workpiece image. Finally, a simple thresholding method can determine the defect locations. Experimental results show the defect detection rate of the proposed method is 90.02% better than those of the traditional spatial and frequency domain techniques. We also applied the proposed approach to detect surface defects of other spherical workpieces with different types of materials. The results show that the proposed method can detect surface defects well not only on steel balls but also on different types of spherical workpieces.