Saving Costly Experiments and Simulations through Machine Learning

The application of machine learning methods has great potential for accelerating technical processes and reducing costs. Based on existing data, neural networks can be trained to predict the results of expensive experimental studies or time-consuming simulations reliably and thus replace them. This article introduces a new neural network that is based on many layers and can be classified into the category of deep learning. Furthermore, it is shown how a network already trained for a data set can be efficiently and time-savingly adapted to a new data set with changed input and/or changed target variables by means of a filter method. The filter method replaces the often very time-consuming training of a neural network by forward and backpropagation. The applicability and efficiency of the presented methods are demonstrated by means of two real technical applications, and the performance is analyzed with regard to the prediction quality and the computing time.