A model for predicting the temperature of a machine tool structure by a neural network using the sliding window method

Abstract

Abstract The paper studies a model for predicting the surface temperature of structural elements of a machine tool using artificial neural networks. A method of forming a training sample by the sliding window method for solving the problem of retrospective forecasting is presented. As applied to a neural network, the sliding window method is an algorithm for forming a training set from an initial set of experimental data necessary to build a forecasting model. Research was carried out for various types of neural networks, namely, generalized regression neural network, radial basis function network and feed forward network. Extrapolation was performed using multistep prediction, in which the predictive system uses the data obtained at the output of the neural network to predict subsequent values. The efficiency and practical suitability of neural network models for predicting the temperature of key heat sources located in certain areas of the machine structure was verified.