自由曲面在航空航天、模具、汽车等重点领域具有广泛的应用。由于产品不断快速迭代升级，提高自由曲面的加工精度、加工效率成为数控加工领域的重要研究课题。在曲面的精加工之前测量，通过获取加工过程中的测量数据指导多轴加工的加工路径生成是提高曲面加工精度的有效方法之一，这种方法称为几何自适应加工。然而，由于引入测量会占用一定的时间，因此几何自适应加工的效率较低。为解决传统的几何自适应加工过程中加工精度与加工效率不能兼顾的问题，提出了一种基于原位检测的自由曲面几何自适应加工方法，该方法通过规划原位检测系统的测量过程，提升自由曲面的加工效率与加工精度。首先，提出了一种基于特征的自由曲面采样方法。该方法以曲率和弧长为依据定义了曲线特征。通过密集选取曲面上的曲线，将曲面采样问题转化为曲线采样问题。通过动态调整曲率和弧长的权重，优化采样点在各条曲线的分布位置，从而实现自由曲面采样。实验结果表明，基于特征的采样方法能够在保证采样精度的前提下减少所需的采样点数量，从而提高减少测量所需时间，提高自适应加工效率。然后，提出了一种原位检测系统五轴测量路径规划方法。采用了粗、精干涉检查相结合的方法避免了测量中测头与工件发生碰撞，并获得了采样点处测头候选方向集合。提出了一种测头姿态预筛选方法获取最小的测头候选方向集合，简化后续计算。该路径规划方法的核心在于改进了遗传算法(GA)的个体结构，增加了代表测头姿态的基因组，并相应地修改了GA的各个操作过程，实现了在路径优化过程中为采样点动态匹配测头姿态，缩短了测量路径长度。实验结果表明，该方法能够有效缩短五轴测量路径的长度，节省后续测量所需时间，提高自适应加工效率。接下来，提出了一种测量数据驱动的原位检测系统测量误差补偿方法。该方法基于三坐标测量机(CMM)、原位检测系统对相同曲面的测量数据获得原位检测系统测量误差。测量数据用于训练卷积神经网络(CNN)模型,实现误差的预测与补偿。提出了一种曲面重构方法用于评价原位检测系统曲面测量的精度。实验结果表明，该方法能够提高自由曲面的测量与评价精度，为自适应加工提供了准确的数据支撑，进而提高自适应加工精度。最后，生成了五轴自适应加工刀具路径。将以上功能集成开发了Thu-OMI自适应加工软件。该软件在Window操作系统环境下采用C++语言开发。基于该软件实现的几何自适应加工能够有效提高自由曲面的加工精度与加工效率。

Freeform surfaces have wide-ranging applications in key areas such as aerospace, molds, and automobiles. Due to the rapid iteration and upgrading of products, improving the machining accuracy and efficiency of freeform surfaces has become an important research topic in the field of numerical control machining. Measuring before the precision machining of surfaces and using the measurement data obtained during the machining process to guide the generation of machining paths for multi-axis machining is an effective method for improving the machining accuracy of surfaces. This method is known as geometric adaptive machining. However, introducing measurement takes up a certain amount of time, resulting in lower efficiency in geometric adaptive machining. To solve the problem of trade-off between machining accuracy and efficiency in traditional geometric adaptive machining processes, a method based on on-machine measurement for freeform surfaces is proposed. This method improves the machining efficiency and accuracy of freeform surfaces by planning the measurement process of an on-machine measurement system.Firstly, a feature-based sampling method for freeform surfaces is proposed. This method defines curve features based on curvature and arc length. By densely selecting curves on the surface, the surface sampling problem is transformed into a curve sampling problem. By dynamically adjusting the weights of curvature and arc length, the distribution of sampling points on each curve is optimized, thereby achieving freeform surface sampling. Experimental results show that the feature-based sampling method can reduce the number of required sampling points while ensuring sampling accuracy, thereby reducing the measurement time and improving the efficiency of adaptive machining.Next, a five-axis measurement path planning method for the on-machine measurement system is proposed. A combination of rough and fine interference checks is used to avoid collisions between the measurement probe and the workpiece during measurement, and a candidate direction set for the probe at the sampling points is obtained. A probe attitude pre-selection method is proposed to obtain the smallest candidate direction set for the probe, simplifying subsequent calculations. The core of this path planning method lies in the improvement of the individual structure of the genetic algorithm (GA) by adding a genome representing the probe attitude and modifying various operations of GA accordingly. This enables dynamic matching of probe attitudes for sampling points in the path optimization process, thereby shortening the measurement path length. Experimental results show that this method can effectively shorten the length of the five-axis measurement path, save subsequent measurement time, and improve the efficiency of adaptive machining.Furthermore, a measurement data-driven error compensation method for the on-machine measurement system is proposed. This method utilizes the measurement data of the same surface obtained from a coordinate measuring machine (CMM) and the on-machine measurement system to obtain the measurement error of the on-machine measurement system. The measurement data is used to train a convolutional neural network (CNN) model to achieve error prediction and compensation.A surface reconstruction method is proposed to evaluate the measurement accuracy of the on-machine measurement system. Experimental results show that this method can improve the measurement and evaluation accuracy of freeform surfaces, providing accurate data support for adaptive machining and thereby improving machining accuracy.Finally, five-axis adaptive machining tool paths are generated. The above functions are integrated and developed into the Thu-OMI adaptive machining software. This software is developed in C++ language under the Windows operating system. Geometric adaptive machining implemented based on this software can effectively improve the machining accuracy and efficiency of freeform surfaces.