【作者】 豆尚成；

【导师】 赵万生； 奚学程；

【作者基本信息】 上海交通大学 ， 机械制造及其自动化， 2013， 博士

【摘要】 电火花线切割加工作为电火花加工的一种类型，在模具、成形电极、难加工材料和精密复杂零件的加工中具有重要的地位。随着技术的发展，对电火花线切割加工的表面质量和加工精度提出了更高的要求。在用线切割加工变厚度工件时，断丝和效率低下是两个普遍存在的问题。当加工工件的厚度由厚变薄时，由于放电能量集中，易于断丝；反之，则效率低下。本文针对线切割加工变厚度工件时的断丝和效率低下问题，开展了离线与在线厚度识别、3D工件模型高度信息提取、加工过程模型建立、以及模型预测控制等方面的研究，以期提高加工的稳定性和加工效率。本文首先建立了基于Linux的线切割加工机床全软数控系统，采用RTAI实时内核满足了系统对实时性的要求。该数控系统包括人机界面、代码解释器、任务管理器、运动控制、机床I/O控制五个模块。运动控制模块产生的控制信号通过多功能I/O卡来驱动机床工作运动。在分析线切割放电电压和电流波形的基础上，根据法拉第电磁感应定律开发了放电频率检测电路。以上即为研究变厚度线切割加工的软硬件平台。为了避免加工变厚度工件时易于断丝和加工效率低下的问题，首先需要获得工件在加工路径上的厚度，然后才能根据厚度选取合适的加工参数。本文针对电火花线切割加工时有无工件的3D模型提出了采用“黑盒法”和“白盒法”分别处理。当缺少工件三维模型时，利用“黑盒法”建立工件厚度的辨识模型。当已知工件三维模型数据时，则利用“白盒法”从工件三维模型中直接提取加工路径上的工件厚度信息，经过数控系统处理，用于在线控制加工过程。鉴于支持向量机在非线性系统建模方面的优异特性，本文利用支持向量机，根据线切割加工过程中采集的放电频率和进给速度以及控制输入量伺服电压和脉冲间隔，建立工件厚度辨识模型，即所谓的“黑盒法”。输出量是待辨识的工件厚度，而模型输入量为放电加工参数(伺服电压和脉冲间隔)以及放电频率和机床的进给速度。在建立了工件厚度辨识模型之后，在加工中应用该模型在线实时地辨识工件厚度。以工件厚度为依据选取合适的加工参数控制加工过程。当加工具有三维模型数据的工件时，从工件三维模型中直接提取加工路径上的工件厚度，并应用到数控系统中在线控制加工过程，即所谓的“白盒法”。随着工件厚度范围的不断扩大，加工过程中所采集的加工信息也不断增加。这些数据不但可以在线指导加工过程，而且还可以用于在线修正工件厚度辨识模型，目的是使模型包含的加工状况更丰富、模型更准确。本文利用最小二乘支持向量机的在线算法实时修正工件厚度辨识模型。当新的数据产生后，在线算法根据投影法稀疏性处理判断新的数据是否需要参与模型计算。如果新的数据需要参与模型计算，则删除支持向量中对模型影响最小的向量。该算法确保参与建模的数据量不会超过预先设定的数量，这不仅减少了计算量和存储量，而且解决了最小二乘支持向量机固有的稀疏性问题。无论是通过“黑盒法”还是“白盒法”获得的工件厚度，最终目的是根据工件厚度，选取合适的加工参数，控制加工过程，达到避免断丝和提高加工效率的目的。为了控制加工过程，基于标准支持向量机建立了加工过程模型，设计了适合变厚度线切割加工的模型预测控制器，当工件厚度变化时，由检测到的工件厚度实时计算出放电频率和加工速度参考值，模型预测控制器根据参考值实时调整输入量伺服电压和脉冲间隔，控制加工过程在不断丝的前提下保持高的加工效率。更多还原

【Abstract】 As one of the non-traditional machining approaches, wire electrical dischargemachining(WEDM) plays an important role in the manufacture of molds, shaped elec-trodes, hard-to-cut materials and precise parts with intricate shapes. With the devel-opment of technologies, the requirements for quality and precision of WEDM arebecoming more demanding. When machining workpieces with a variable height, wirebreakage and low machining efficiency are the two major issues. When the workpieceheight along a machining path changes from high to low, the discharge energy willbe concentrated on a short range of the wire. As a result, wire breakage is likely tooccur. On the contrary, when the workpiece height varies from low to high, the densi-ty of discharge along the wire becomes low, thus resulting in a low material removalrate(MRR).In order to avoid wire breakage and improve machining efficiency when machin-ing workpieces with variable heights by WEDM, this dissertation investigates severaltopics which are crucial for variable height machining by WEDM, those include of-fline and online workpiece height estimation, extraction of workpiece heights from3D models, modelling of WEDM processes, as well as model predictive control forWEDM processes.In this dissertation, a CNC (computer numerical control) system for a WEDMwas developed based on Linux operating system. The RTAI kernel is attached to theLinux to meet the real time requirement for CNC systems. The CNC system consists offive modules, which include GUI module, code interpreter module, task managementmodule, motion control module and I/O control module. The control signals generatedby the motion control module drive the machine through a multi-function I/O card.After analyzing the discharge current and voltage wave forms collected during the machining，on the basis of Farad’s induction theory, a discharge frequency monitorsystem was developed to detect the discharge frequency.For variable height WEDM, the workpiece height must be provided to CNC sys-tem, so that appropriate machining parameters can then be determined according toit. In this dissertation, depending on the availability of a3D model, two cases areprocessed separately. If there is no3D model of a workpiece, a black box methodis adopted to build up a workpiece height estimation model. When the3D model isavailable, a white box method is applied to extract the workpiece height informationalone the machining path from the3D model. The workpiece height data is finally fedinto the CNC system.Due to its good performance in modelling nonlinear systems, support vector ma-chine (SVM) is used to build up a workpiece height estimation model with the ma-chining data collected. This is the so-called black box method. The input and outputof the model are chosen from those machining parameters which have no noticeableimpacts on the surface quality. The output variable is the workpiece height that is tobe estimated, and the input variables are a set of machining parameters, discharge fre-quency and feed rate. The estimated workpiece height will be then used as a referencefor appropriately determining a set of machining parameters.When the3D model of a workpiece is available, the workpiece height extractedfrom the3D model can be directly applied to the CNC system. This is the so-calledwhite box method. As the range of workpiece heights becomes larger, the machiningdata collected is increasing. Not only can this data be used to control the machiningprocess in progress, but also can be used to update the workpiece height estimationmodel online. By learning the new dynamics which have not been excited before, theheight estimation model can be refined, thus improving its prediction accuracy. Theupdate of the workpiece height estimation model was achieved by using the on lineleast square support vector machine (LS-SVM). When a new data pair is available, ajudgment must be made whether the new data pair should be selected as a new basicvector (BV) based on the projecting method. If the new data pair is selected as a newBV, the old BV which has the least significant impact on the model should be removedfrom the BV set. By keeping the number of BVs within a predefined value, this online algorithm can not only reduce the computational load and the requirement for memoryspace, but also overcome the non-sparsity of LS-SVM.No matter how the workpiece height is obtained either by the black box methodor by the white box method, the workpiece height will be used to determine an appro-priate set of machining parameters so as to avoid wire breakage and improve MRR.A model predictive controller is designed in such a way that the reference dischargefrequency will be followed to maintain a stable machining. The optimal control inputsare then generated by solving an online optimization problem.更多还原