Model-based in-process determination of the tool wear at high performance turning

尽管使用的涂层硬质合金刀具种类繁多，但对刀具磨损潜在机制的理解仍然显示出相当大的缺陷。在本项目中，基于切屑形成模拟的（白盒）模型将与人工神经网络（ANN，黑盒模型）相结合，形成灰盒模型，以显著提高对高性能车削涂层硬质合金刀具磨损形成和开发的理解。白盒模型旨在根据测量的热-机械负荷谱近似确定当前工具磨损参数。然后，估计的磨损将与过程中的测量数据一起作为训练后的黑匣子模型的输入，以精确预测工具磨损。特别是，除了热-机械载荷谱中包含的信息外，还将考虑结构噪声信号、巴克豪森噪声幅值、工件尺寸和表面粗糙度中包含的刀具磨损状况信息。通过这种研究方法，优先计划应该成功地绘制和识别以前未知的涂层降解和刀具磨损机制，从而有助于改进基于知识的高性能切削刀具涂层资格。白盒模型是在现有的有限元切屑形成模型的基础上发展起来的。阐明了涂层性能的有效嵌入和逆模型的用法。黑盒模型采用人工神经网络实现。它的训练需要快速直接检测刀具磨损，从而实现和使用基于光学原理的两种测量程序，从而能够精确地确定刀具几何形状和涂层厚度。对于黑箱模型，研究了人工神经网络的结构和传感器数据，以便以最小的不确定性确定磨损。为了最大限度地减少黑箱模型的输入数量，阐明了如何对进程中采集的数据进行信号预处理，使信息损失可以忽略不计。建立了灰盒模型，并对调质42CrMo4的外纵向车削进行了验证。此外，还计划对优先方案C45的截面材料进行扩展验证。在优先计划的第二阶段，灰盒模型将被扩展到磨损发展的预测以及更广泛的驱动和系统变量的应用范围。

Despite the variety of coated carbide tools in use, the understanding of tool wear underlying mechanisms still shows considerable deficits. In this project, a (whitebox) model based on chip formation simulations is to be combined with an artificial neural network (ANN, blackbox model) to form a greybox model in order to develop a significantly improved understanding of the wear formation and development of coated carbide tools for high-performance turning. The whitebox model is intended to enable an approximate determination of current tool wear parameters on the basis of the measured thermo-mechanical load spectrum. The estimated wear will then be used together with in-process measurement data as input for a trained black box model to precisely predict tool wear. In particular, the information on the wear condition of the tool contained in structure-borne noise signals, Barkhausen noise amplitudes, workpiece dimensions and surface roughness will be taken into account, in addition to the information contained in the thermo-mechanical load spectrum. With this research approach, the priority program should succeed in mapping and identifying previously unknown mechanisms of coating degradation and tool wear and thus contribute to an improved knowledge-based qualification of tool coatings for high-performance cutting. The whitebox model is developed on the basis of an existing finite element chip formation model. The valid embedding of the coating properties and the inverse model usage are clarified. The blackbox model is realized by means of an artificial neural network. Its training requires a fast direct detection of the tool wear, so that two measurement procedures based on optical principles are realized and used, which enable a precise in-situ determination of the tool geometry and the coating thickness. For the black box model, the structure of an artificial neural network and the sensor data are investigated to enable wear determination with minimal uncertainty. In order to minimize the number of inputs of the blackbox model, it is clarified how a signal preprocessing of the in-process collected data with negligible information loss can be realized. The resulting greybox model is developed and validated for external longitudinal turning of quenched and tempered 42CrMo4. In addition, an extended validation on the cross-section material of the priority program C45 is planned. In the second phase of the priority program, the greybox model is to be extended for the prediction of wear development as well as for a broader application range of actuating and system variables.