Heat treatment process parameter estimation using heuristic optimization algorithms

We present an approach for estimating control parameters of a plasma nitriding process, so that materials with desired product qualities are created. We achieve this by solving the inverse optimization problem of finding the best combination of parameters using a real-vector optimization algorithm, such that multiple regression models evaluated with a concrete parameter combination predict the desired product qualities simultaneously. The results obtained on real-world data of the nitriding process demonstrate the effectiveness of the presented methodology. Out of various regression and optimization algorithms, the combination of symbolic regression for creating prediction models and covariant matrix adaptation evolution strategies for estimating the process parameters works particularly well. We discuss the influence of the concrete regression algorithm used to create the prediction models on the parameter estimations and the advantages, as well as the limitations and pitfalls of the methodology.