he paper presents a procedure for studying the stability of modeling an artificial neural network as applied to the thermal characteristics of machine tools. The topicality of this procedure is dictated by the ambiguity of the results generated by the neural network when constructing the predicted thermal characteristics of machine tools. Therefore, to select one of the possible solutions generated by the neural network, it was proposed to use two criteria. The effectiveness of their use is confirmed by the presented machine experiments. The methodology proposed in this work has made it possible to form a generalized concept for studying the effectiveness of the use of neural network technologies in thermal modeling of machine tools. This concept defines a typical set of variable modeling parameters, a basic mathematical model based on a modal approach, and an architecture of a typical software tool that can be developed to study the effectiveness of artificial neural network modeling. For each variant of the input data of the network, the following parameters were varied: the number of neurons in the hidden layer; the size of the input and output vectors; input vectors error; the size of the training, validation and test sample; functional features of thermal characteristics supplied to the network input or their multimodality; the presence and absence of normalization of the input vector. The paper presents experimental thermal characteristics for two spindle speeds of a vertical CNC machine. The results of the machine experiment are presented for six variants of the variable parameters of the mathematical model. The software tool used to carry out the machine experiment was developed in Matlab.
Hybrid modelling routine for metal‐oxide TFTs based on particle swarm optimisation and artificial neural network