LAUSR

Health monitoring systems play a key role inside smart factories. To enhance the real-time capability and reliability of health monitoring systems, we propose a fully automatic method for machine diagnosis. Firstly, acquired vibration signals are converted into high-resolution images by wavelet packet spectral subtraction. Next, a trained convolutional neural network (CNN) automatically extracts important features and determines the current health of the machine. The performance of the proposed method is demonstrated by employing a diagnosis problem of a bearing system. The result shows an outstanding classification accuracy of 99.64 % even with a small amount of training data (5 % of the data).