應用機械手臂進行物件三維表面掃描、重建及形貌誤差分析

Abstract

本研究於六軸機械手臂進行高精度之物件三維表面重建並利用形貌誤差分析驗證重建精度。利用空間姿態校正架之特殊排列的參考球定位物件在空間中的絕對位置，其定位精度取決於光學量測探頭，本研究使用國立台灣大學機械系精密量測實驗室所開發之雙相機光學量測探頭進行掃描，空間姿態校正架定位物件在空間中絕對位置的定位精度最小誤差可達0.123 mm，平均誤差為0.195 mm，並使用變異值最近點迭代法(Variants of the Iterative Closest Point)將兩鄰近之點雲精密的擬合(Registration)，可彌補定位精度不足的問題。此外利用適當之信心指數(Confidence)篩選點雲品質可降低31.4%的重建誤差，利用德勞內三角化(Delaunay triangulation)可有效的降低資料量並保留原始三維形貌特徵，相較於等量法降低資料量，可減少7.7%因為降低資料量而產生的形貌誤差。完成物件三維表面重建後，可利用形貌誤差分析驗證重建精度。誤差分析的精度取決於重建後的結果與CAD模型之間的對位誤差，以本研究為例，誤差分析的精度為5 μm。利用上述點雲處理相關的演算法並搭配空間姿態校正架可於六軸機械手臂上進行高精度之物件三維表面重建。

This study utilizes a six-axis robotic arm to achieve accurate three-dimensional object surface reconstruction. To position the object, this study designed the calibration target. The positioning accuracy of the calibration target depends on the optical measurement probe. This study utilizes the dual CCD optical measurement probe from National Taiwan University Precision Metrology Laboratory to reconstruct three-dimensional object surface. The minimum positioning accuracy of the calibration target is 0.156 mm and the mean positioning accuracy of the calibration target is 0.2 mm. To compensate the low positioning accuracy, this study implements the Variants of the Iterative Closest Point (ICP) algorithm to register two close point clouds fine. Furthermore, estimating the quality of point cloud by Confidence could reduce 31.4% of the reconstruction error. Comparing the unit downsampling and Delaunay triangulation downsampling, Delaunay triangulation downsampling could reduce 7.7% of the form error. Using object form error analysis could verify the object surface reconstruction accuracy. To analyze the form error, the CAD model of the object needs to be aligned with the surface reconstruction model of the object. The accuracy on the form error analysis depends on the alignment error. In this study, the accuracy on form error analysis is 5 μm. Integration between the calibration target and point cloud processing algorithm proposed by this study could achieve accurate three-dimensional object surface reconstruction using a 6-axis robotic arm.