Prediction thermal deformation of machine tool by neural networks - analysis of effective temperature data

摘要

In training neural networks, it is important to reduce training data for saving memory for the training data, reducing network size, and achieving fast training. This paper proposes two kinds of analysis methods for useful training data used in neural networks applied to prediction and diagnosis. One method is to use information of connection weights after training. A summation of absolute value of the connection weights related to each input element. In some case, only positive connection weights are taken into account. The other method is based on correlation coefficients of the input vectors. If some data can be obtained by amplifying and shifting another data, then the former can be absorbed in the latter. These analysis methods are applied to prediction of thermal deformation in machine tools. The training data are well reduced while achieving good prediction performance.