Deep-Learning Method Based on 1D Convolutional Neural Network for Intelligent Fault Diagnosis of Rotating Machines

Abstract

Fault diagnosis in high-speed machining centers (HSM) is critical in manufacturing systems, since early detection saves a substantial amount of time and money. It is known that 42% of failures in these centers occur in rotatory machineries, such as spindles, in which, the bearings are fundamental elements for effective operation. Nowadays, there are several machine- and deep-learning methods to diagnose the faults. To improve the performance of those traditional machine-learning tools, a deep-learning network that works on raw signals, which do not require previous analysis, has been proposed. The 1D Convolutional Neural Network (CNN) proposed model showed great capacity of adapting to three types of configurations and three different databases, despite a training set with a smaller number of categories. The network still detected faults at early damage stages. Additionally, the low computational cost shows the Deep-Learning Neural Network’s (DLNN) suitability for real-time applications in industry. The proposed structure reached a precision of 99%; real-time processing was around 8 ms per signal, and standard deviation of repeatability was 0.25%.