鉚釘電氣接點在各式的開關、斷路器和繼電器中扮演了重要的角色,且在使用期間,每個接點要工作上百萬次,為了能承受這樣嚴格的工作要求,接點必須耐磨,性能可靠。但在接點沖壓、打造的過程中,會因為機械設定參數不佳而產生各式不良缺陷,進而影響接點的導電性、導熱性和外觀賣相,因此本研究目的在運用影像處理技巧,建立一套有效之鉚釘電氣接點之視覺檢測系統,以機器視覺代替人力來做嚴格的品質把關及瑕疵分類。研究內容主要為擷取六種表面瑕疵的空間域特徵資訊來做處理,並依據空間域特徵資訊區別瑕疵種類。 另外,本研究在真圓度量測上共使用三種不同之最佳化方法(粒子群最佳化,(PSO),結合全域學習之模式搜尋法(PSO-HJ法),模擬退火結合模式搜尋法(SA-HJ)來求解真圓度量測之數學模型: 最大內切圓(Maximum Inscribing Circle)和最小外接圓(Minimum Circumscribing Circle),而量測物件為接點之正、反面邊緣。實驗結 果顯示,本研究提出之結合全域學習之模式搜尋法能以較快速度求得真圓度量測結果,其精確度與粒子群最佳化法所得之解相當,而模擬退火結合模式搜尋法則在求解速度和求解品質上都有尚待改善的空間。
Electric contacts for switches, breakers and relays play a very important role in electric industry. In this paper, we propose an automated visual Inspection system for electric contact rivets using computer vision .The system is developed based-on spatial domain information of defects, and classifies the surface defects including cracks, breaks, and scratches. Furthermore, this study proposes a machine vision-based roundness measuring method that applies Particle Swarm Optimization Algorithm (PSO) , modified Hooke –Jeeves Pattern Search and Simulated annealing algorithms (SA) for computing the roundness measurement of maximum inscribing circle (MIC), minimum circumscribing circle (MCC). Experimental results show the modified Hooke –Jeeves Pattern Search-based method outperforms PSO-based method and SA-based method in both accuracy and the efficiency.