Induction Hardening (IH) process contributes to the improvement of the mechanical properties of the steel. In this application, the inverter operates at resonant frequency for high operating performances. However, this is usually accompanied by steel characteristics variation under temperature effect especially the magnetic permeability. This paper proposes two approaches for resonance frequency adaptation. The first is based on maximum efficiency tracking (MET) scheme. In this context, simulation with an algorithm for control is developed to find in real-time the adequate frequency based on current and voltage sensor. This approach is found effective but requires a constant determination of the input and the output power. Hence, it makes it difficult for implementation. To alleviate this drawback, advanced technique based on deep learning (DL) algorithm is proposed. Thus, an experimental prototype is built to determine the experimental Temperature-Frequency data profiles up to the hardening for two metal bars. Hence, a neural network (NN) based model control is elaborated. After a careful selection of NN parameters, a trained model is obtained with satisfactory accuracy leading to the predict model. This was easily implemented in real time on raspberry pi 3 + and allows the system to perform continuously under high efficiency for hardening purposes.

感应淬火 (IH) 工艺有助于提高钢的机械性能。在此应用中，逆变器以谐振频率运行，以实现高运行性能。然而，这通常伴随着钢特性在温度影响下的变化，尤其是磁导率。本文提出了两种谐振频率自适应方法。第一种是基于最大效率跟踪 (MET) 方案。在这种情况下，开发了带有控制算法的仿真，以基于电流和电压传感器实时找到合适的频率。发现这种方法是有效的，但需要不断确定输入和输出功率。因此，实施起来很困难。为了减轻这个缺点，提出了基于深度学习（DL）算法的先进技术。因此，建立了一个实验原型来确定直到两个金属棒硬化的实验温度-频率数据分布。因此，详细阐述了基于神经网络（NN）的模型控制。在仔细选择 NN 参数后，以令人满意的精度获得训练模型，从而生成预测模型。这很容易在 raspberry pi 3 + 上实时实现，并允许系统在高效率下连续执行以达到硬化目的。以令人满意的精度获得训练模型，从而产生预测模型。这很容易在 raspberry pi 3 + 上实时实现，并允许系统在高效率下连续执行以达到硬化目的。以令人满意的精度获得训练模型，从而产生预测模型。这很容易在 raspberry pi 3 + 上实时实现，并允许系统在高效率下连续执行以达到硬化目的。