Verification of Determination Possibility using Convolutional Autoencoder for Machine Tool Abnormality Detection

The purpose of this research is to clarify the cause of failure and to improve the accuracy of abnormality detection by predicting the noise added to the machine tool from the results of the convolutional autoencoder.Data obtained from an acceleration sensor mounted on the machine tool are reconstructed using a convolutional autoencoder, and the average absolute error is calculated.The maximum value of the average absolute error of the training data is used as the threshold value for abnormality detection.Multiple simulated data with different amplitude values based on a composite sine wave with a specific frequency and random numbers within a specified amplitude value were verified.In this paper, the characteristics of each type of noise and the parameters of the optimal model were examined from error rate and error distribution.