In machining, high temperatures and strain rates impact the flow stress of the workpiece material, making it essential to understand the materials behaviour in these process conditions for meaningful finite element analysis (FEA) of the cutting process. The Johnson-Cook constitutive model, despite being the most widely applied, is reported to struggle in capturing the material behaviour outside of the reference conditions it was calibrated on. However determining these parameters in conventional material tests is challenging. To solve this issue, this study proposes a grey-box approach which aims to increase the accuracy of process force prediction of FEA, employing a Johnson-Cook model determined by experiments conducted on a Split-Hopkins Pressure Bar and compression tests at elevated temperatures on a Gleeble 3800c for AISI 1045, over a variety of cutting parameters. In total 110 cutting experiments and their corresponding simulations were carried out in a fully factorial experimental design with eleven cutting speeds and ten uncut chip thicknesses. Succeeding the white-box model, a black box model is trained to capture the non-linear behaviour between the simulation and the cutting experiments. Among the tested algorithms, XGBoost and Support Vector Regression outperformed Random Forests and Neural Network for predicting cutting force and feed force. The proposed grey-box approach showed an improved capability of predicting cutting force and feed force, reducing the mean absolute error and mean squared error compared to the white-box model by 97.9 % and 99.9 % for cutting force and by 94.9 % and 99.7 % for feed force, respectively. The grey-box model achieved a mean error of 1.3 % with a standard deviation of 0.1 in process force prediction.
扫码关注我们
求助内容:
应助结果提醒方式:
