Intelligent measurement system for vibrating screen faults based on accelerometers and deep learning

Abstract

An end-to-end intelligent measurement system based on multi-accelerometers and deep learning is proposed to address the issues of insufficient accuracy and poor real-time performance in fault diagnosis of vibration screens under complex working conditions. The system utilizes an array of accelerometers to achieve collaborative perception and real-time acquisition of multi-dimensional vibration signals. An improved Savitzky–Golay filter is utilized for signal smoothing, maintaining the amplitude measurement deviation within 2.3%, combined with an adaptive wavelet algorithm to effectively eliminate trend item interference. The method converts time-domain acceleration signals into two-dimensional grayscale images and constructs an improved LeNet-5 convolutional neural network model to achieve end-to-end intelligent diagnosis directly from signals to fault categories. Experimental results demonstrate that the proposed model achieves an accuracy of 99.28% after 20 training epochs, with its excellent feature separability validated via t-SNE algorithm visualization. Finally, a system developed on the NI Vision and LABVIEW platform enables full-process real-time measurement, monitoring, and diagnosis, providing a reliable solution for the intelligent operation and maintenance of heavy-duty equipment.