Solving evolutionary problems using recurrent neural networks

A technique for flexible control of induction baking of electrically non-conductive layers (paints, varnishes, resins, etc.) is presented, based on the temperature prediction. As the numerical solution of the full model of the process takes a long time, it is necessary to approximate it with a suitable equivalent model. In this case, recurrent neural networks (RNNs) prove to be a powerful tool for solving the task practically online and providing the input data to control the field current fast enough. The methodology was first tested to predict the current based on the knowledge of the voltage, which can be determined from the analytical solution of the ordinary differential equation that describes the feeding circuit. Subsequently, the methodology was implemented on a system for baking non-conductive layers.(c) 2023 Elsevier B.V. All rights reserved.