本研究為工件表面粗糙度與加工副產物顏色之影像視覺檢測,主要探討利用影像檢測的方式,檢視在不同的光源與加工條件下之金屬表面,並針對表面粗糙度進行分級,以及銑削時所產生之切屑顏色分析。光源的分析為光譜照度分析分別在暗室、日光燈、LED白光與LED紅光光源下進行。物件特徵辨識主要針對金屬加工件的表面紋路與切屑顏色,進行特徵的檢測分析,並配合田口方法進行L9直交表的建立得出最佳加工參數,而本研究的最佳加工參數為轉速以1000rpm、進給速度以40m/min、鍍鈦兩層與切深0.4mm。其中分析的物件主要以中碳鋼鐵塊,在使用不同鍍層的高速鋼銑刀配合不同加工參數下銑削之表面,與其產生的切屑。表面粗糙度檢測方式是利用影像中灰階的梯度變化,進行邊緣搜尋計算edge數,其邊緣搜尋的基礎理論為影像傅立葉轉換的一階導函數,利用其計算結果成功辨識出標準試片銑削紋路特徵位置,並統計其特徵數量以建立資料庫進行粗糙度分級,且與非接觸式表面粗糙儀進行交叉比對,最後使用田口方法進行實際加工的望大特性確認實驗,其實驗結果使用本研究的影像檢測系統量測數值為Rz3.2,並使用表面粗糙儀進行量測結果一致而其準確率為100%,而為確認系統穩定性,進行不同銑削表面之再現性實驗其結果準確率為98%。本研究檢測切屑溫度範圍為100至600度並分為6個等級,藉由加工溫度上升產生不同切屑顏色之特性,建立不同溫度鐵屑顏色的資料庫。顏色辨識分級使用樣本比對的方式成功辨識出不同切屑顏色,並比較樣本與實際影像之HSL色域強度,則發現樣本與檢測物色相、飽和度與亮度誤差率皆在3%之內,此趨勢為顏色檢測時的重要依據,以證明本研究顏色檢測系統的可靠度。
This study investigates the vision image inspections of surface roughness and byproduct color for cutting parts. In general, the image inspections for object recognition usually depend on metal surfaces and processing conditions. Therefore, the roughness classification and chip color analyses of different light sources are analyzed for image inspections. In addition, the LED red light, LED white light and fluorescent of spectral irradiance analysis are in a darkroom. The objects for image inspections have the roughness and byproduct color of the medium-carbon steel using different coatings and HSS end mill of different processes. The experimental method of the larger-the-better is Taguchi method L9 orthogonal arrays and finds out the optimum processing parameter. The edge detection of the object is search by gradient grayscale basic on Fourier transform of the first derivate. The color detection of byproduct is compared by the database of temperature color. Results showed that the position of the roughness standard test piece marks identifies and standard of the different roughness are successfully to create the database, and the database match of the non-contact surface roughness tester is also successfully. In addition, the result trends of response graph matches with sample, and surface roughness tester showed the result roughness is Rz3.2.The roughness consistency of vision inspections is 100%, and the system stability of different surface roughness vision inspections is 98%. Color classification of byproduct detection is successfully and the trends byproduct of HSL and HSV intensity are close. Finally, the database of different chip temperature color is successfully to create.