Heat flux during machining has received extensive attention due to its importance for understanding the cutting mechanism and promising prospects on intelligent manufacturing. Research on heat flux estimation by inverse heat conduction methods faces many challenges, including measurement error amplification, stability of the methods, and limitations for applications. In this paper, we introduce a long short-term memory (LSTM) based encoder-decoder (ED) scheme in online estimation of the heat flux imposed at the tool-chip region during turning. The math-physical model and finite element model are established to generate training datasets. Numerical tests using simulated heat flux and temperature data representing different machining conditions are carried out to evaluate the method performance. Compared with other artificial intelligence methods such as multilayer perceptron, convolutional neural networks and LSTM, the LSTM-ED model performs better at all tested noise levels ($1\le \sigma \le 20K$) with acceptable time cost for online process. Effects of the location and number of sensors on the accuracy of heat flux estimations are also investigated. Experimental validations based on cutting temperature measurements by five thermocouples located in the insert are conducted. Both numerical and experimental tests indicate the potential of the LSTM-ED method for online heat flux monitoring in scientific research and industrial production.

机加工过程中的热通量因其对理解切削机理的重要性和智能制造的前景而受到广泛关注。通过逆导热方法估算热通量面临许多挑战，包括测量误差放大，方法的稳定性以及应用限制。在本文中，我们介绍了一种基于长短期记忆（LSTM）的编码器-解码器（ED）方案，用于在线估计车削过程中施加在工具芯片区域的热通量。建立数学-物理模型和有限元模型以生成训练数据集。使用模拟热通量和代表不同加工条件的温度数据进行了数值测试，以评估该方法的性能。$\mathrm{1个}\le \sigma \le 20ķ$），并且在线过程的时间成本可以接受。还研究了传感器的位置和数量对热通量估计精度的影响。基于刀片中五个热电偶的切削温度测量结果进行了实验验证。数值和实验测试均表明，LSTM-ED方法在科研和工业生产中用于在线热通量监测的潜力。