Vibration-Based Condition Classification of ASTM C93900 Bronze Plain-Bearing Bushings Using Deep Learning

Abstract

Plain-bearing bushings are essential components in rotating machinery, directly influencing reliability, efficiency, and maintenance costs. While vibration-based condition monitoring has been widely applied to rolling-element bearings, experimental studies involving sliding bushings under real fault conditions remain limited. This work proposes an integrated experimental and deep learning framework for the vibration-based condition classification of ASTM C93900 bronze plain-bearing bushings. An experimental test bench was developed to simulate realistic operating conditions and acquire vibration signals under three scenarios: the normal condition, clearance variation, and geometric eccentricity. Data were collected using low-cost MEMS accelerometers and processed through signal segmentation, MinMax normalization, and a compact $8\times 8$ grayscale matrix representation for convolutional neural network (CNN) classification. The proposed framework was evaluated under multiple fault combinations and bearing positions and was compared with baseline dense neural network (DNN) models. Results showed classification accuracies above 90% in all evaluated scenarios, demonstrating the capability of the CNN to identify subtle vibration patterns associated with different fault conditions. Furthermore, the compact signal representation reduced computational complexity while preserving relevant diagnostic information. The proposed methodology demonstrates the potential of combining tribological experimentation, vibration monitoring, and artificial intelligence (AI) techniques for predictive maintenance applications, and it provides an experimental dataset for future research involving sliding bushings.