Abstract: Owing to lacking of prior knowledge of the profile,the reconstruction accuracy of the measured surface heavily depends on the flexibility and intelligence of the sampling strategy that is how to allocate the next sample appropriately and duly during the measuring process.The geometrical feature responses of the smooth surface to the candidate locations were predicted by means of the nonlinear mapping capability of the Radial Basis Function Neural Network (RBFNN),whose uncertainties were estimated and introduced into the proposed new informative criterion MaxCWVar weighting profile curvature's effect for selecting the Next Best Point (NBP).Taking the inspection of a freeform curve of the blade cross-section as an example,the adaptive sampling outperformance of the proposed method was validated.The comparison with the other two process-based adaptive sampling strategies showed that the RBFNN-based geometric response prediction was well guidable to assign sample points.Furthermore,contrasted to the other three NBP selection criteria,the distribution of the sample points directed by MaxCWVar follows the geometric feature changes unexceptionably,which was verified by the correlation analysis between sample density and curvature variation characteristic.Especially in the case of the higher real-time requirement on the sampling process,the proposed method had superior performance in the reconstruction accuracy and modeling efficiency.This study can enlighten us to explore favorable methods for re-modeling complex non-model smooth surfaces fast and intelligently.

Key words: uncertainty model, adaptive sampling, radial basis function neural network, MaxCWVar informative criterion, next best point

摘要： 由于缺少关于廓形的先验知识,具有不确定性被测表面的重构精度取决于采样方法的自适应程度,即在测量过程中对下一采样点的实时合理设置。利用径向基函数神经网络(RBFNN)的非线性映射能力预测被测光滑表面备选采样点的几何特征响应,并将其不确定度估计代入提出的考虑轮廓曲率影响的MaxCWVar信息标准中用于选择下一最优测点(NBP)。以叶片截面自由曲线为例,验证了该方法自适应采样性能的优越性。与其他自适应采样策略的对比表明,基于RBFNN的响应预测对于采样点位置确定具有很好的指导作用;与其他三个常用的NBP选择标准相比,根据MaxCWVar标准得到的采样点分布更为合理,能及时准确地跟随轮廓的几何特征变化,经样本密度与曲率之间的相关性分析得以验证。特别是对采样实时性有较高要求的情况下,所提出方法具有更好的重构精度和建模效率。研究成果对于探索快速、智能的复杂无模型光滑曲面重构方法具有启发意义。

关键词: 不确定模型, 自适应采样, 径向基函数神经网络, MaxCWVar信息标准, 下一最优测点