CNN-based fault classification using combination image of feature vectors in rotor systems

The advent of 4th industrial revolution technologies has spurred the development of computing technologies such as big data, cloud computing, and the internet of things (IoT). These advancements have facilitated the application of automated systems across various industrial domains, including the innovative application of these technologies in rotating machinery diagnostics. In this field, vibration data measured at various locations can be utilized for fault diagnosis by analyzing key feature parameters derived from time, frequency, entropy, and cepstrum signals, which are crucial for vibration signal analysis. This study proposes a novel image processing method that constructs diagnostic images by combining feature vectors extracted from these signals. To evaluate the efficacy of this method, simulated vibration signals representing 7 different operational states were acquired using a lab-scale gearbox. The classification performance of the proposed method was assessed using a CNN algorithm, known for its superior performance in image classification tasks. The results demonstrate that combining feature vectors from multiple domains enhances classification performance compared to using feature vectors from a single domain.