高精度薄壁零件加工的动力学特性分析与工艺在线精化方法研究

Nowadays, the demand for the high precision thin-walled shell parts is especially large. However, the strict demand for high precision of these part, the poor rigidity of the process system and the outstanding vibration problem, which lead to low passing rate, the long processing technology research and development cycle, and the unstable processing quality, all have become the main concerns and urgent problems to be resolved on manufacturing such parts. So this research is focused on the scientific challenges of the modeling and analysis of the characteristics of the thin-walled process system and the online learning method under the cutting state. The mechanism of interaction of the contact interface between the thin-walled workpiece and the weak rigid tool is studied and proposed. This research then analyzes the influence of contact behavior on the dynamic characteristics of the workpiece and the tool and establishes a machining stability model considering the dynamic characteristics of the workpiece and the tool and the contact behavior. The relative vibration and deformation between the tool and the workpiece are further considered in the study on design method of theoretical process window. Then on this basis, this research proposes a method to obtain the perceptual information of the actual machining condition. The relative vibration between the tool and the workpiece is indirectly predicted using the model of the vibration transfer mechanism between the measuring points and the measured points. Then the design and optimization of process parameters for reward function combining Bayes reinforcement learning method are researched. The optimization direction are determined by evaluating the advantages and disadvantages of the machining process based on information of the relative vibration and the process window refinement is realized based on perception of online condition according to the iterative optimization of the actual result of the machining process.

目前，高精度薄壁壳体类零件的需求量大，但是其精度要求高，由于工艺系统刚性差，振动问题突出，导致合格率不高，加工工艺研发周期长，且加工质量不稳定，已成为当前此类零件制造急需解决的难题。为此，本项目围绕切削状态下的薄壁工艺系统特性建模、分析以及工艺在线学习方法的科学挑战，提出通过研究加工状态下薄壁工件与弱刚性刀具的接触界面行为机理，分析接触界面行为对工件端、刀具端的动力学特性影响规律，建立考虑刀具端、工件端动力学特性及其接触界面行为的加工稳定性模型，进而研究考虑刀具与工件相对振动、变形的理论工艺窗口的设计方法，在此基础上，提出通过获取实际加工工况的感知信息，利用测点与被测点的振动传递机理模型，间接测量刀具和工件的相对振动，并结合贝叶斯与强化学习，研究面向工艺参数优化的奖罚函数设计，根据相对振动信息评估工艺的优劣，确定优化方向，根据实际的加工结果迭代优化，实现基于在线工况感知的工艺精化。

航空航天和3C领域，存在大量的如叶片、机匣和舱段等薄壁零件，由于零件刚性差，导致加工过程中易发生颤振和变形，且加工质量不稳定，已成为当前此类零件制造急需解决的难题。为此本课题深入开展了如下研究：.1) 针对加工过程中刀具和工件的接触行为对系统动力学特性的影响，开展了考虑刀具接触条件的薄壁工件动力学参数的分析研究，揭示了刀具接触行为对工件动力学特性的影响规律。.2) 考虑刀具耦合效应对工件模态参数的影响，使用奇异谱分析去除功率谱谐频并用PolyMAX算法辨识得到模态参数，研究了加工条件下薄壁工件模态参数相对静态条件下的变化规律。.3) 围绕薄壁零件加工过程容易发生颤振的问题，提出了动力学参数在线辨识方法，结合OMA辨识得到的模态参数求解薄壁工件铣削稳定性动力学模型，利用改进的数值积分法得到稳定性lobe图。.4) 将切削加工信号结合模态分析确定颤振与频谱数据分布的关系并提取了四种对颤振敏感的特征，选用了能够实现数据和特征并行、对海量数据进行快速训练和测试的LGBM模型，针对颤振样本和正常样本数目不平衡的问题提出了基于网格搜索最优类别权重参数的方案，最后分别在域内和域外切削参数验证了LGBM模型的适用性和可靠性。.5) 针对壁薄零件加工过程易变形的问题，开展了基于负载电流和机器学习的加工误差建模以及预测研究，建立了基于前向传递网络的壁薄零件航轴加工误差预测模型，显著提高了零件加工质量；.6) 围绕薄壁零件加工过程刀具磨损对加工质量的影响，提出了基于稀疏堆叠自编码的刀具磨损特征提取方法，提取了对刀具磨损的敏感特征，建立了基于堆栈式稀疏自编码器的刀具磨损预测方法，并在手机壳体等薄壁零件加工过程中进行了验证。.通过本课题的研究，完成了课题预期目标，取得了一系列成果，发表论文8篇，其中SCI收录论文5篇，申请发明专利2项，培养博士生1名，硕士生4名。