Abstract:

To address the lack of adaptability in traditional signal processing methods for vibration feeder fault diagnosis, an adaptive 1D CNN-based fault diagnosis algorithm was introduced. Through an in-depth analysis of the fault characteristics of the vibration feeder, the original vibration signals of fault states were preprocessed to construct training and testing samples. Each training sample was partitioned over time and served as the input for the 1D CNN, enabling adaptive hierarchical extraction of vibration signal features at different time intervals. The results show that by introducing adaptability, the algorithm can adapt more flexibly to complex dynamic patterns in vibration signals. This adaptability results in higher diagnostic accuracy, especially when facing frequent changes in vibration characteristics or the presence of noise interference. After selecting appropriate neural network dimensions, the algorithm outputs diagnostic results for fault states in the vibration feeder. With an overlap ratio of 49%, the accuracy reaches 90.2%.

Key words: vibration feeder, one-dimensional convolutional neural network (1D-CNN), fault diagnosis, adaptive extraction, overlap ratio