Simulated Centrifugal Fan Blade Fault Diagnosis Based on Modulational Depthwise Convolution–One-Dimensional Convolution Neural Network (MDC-1DCNN) Model

Existing intelligent fault diagnosis methods have been widely developed and proven to be effective in monitoring the operating status of key mechanical components. However, centrifugal fans, as important equipment in energy and manufacturing industries, have been used for a long time in complex and harsh environments such as boiler plants and gas turbines. Therefore, the vibration signals they generate show complex and diverse characteristics, which brings great challenges to the monitoring of centrifugal fan operation status. To solve this problem, this paper proposes a centrifugal fan blade fault diagnosis method based on a modulational depthwise convolution (DWconv)–one-dimensional convolution neural network (MDC-1DCNN). Specifically, firstly, a convolutional modulation module (CMM) with strong local perception and global modeling capability is designed by drawing on the Transformer self-attention mechanism and global context modeling idea. Second, multiple DWconv layers of different sizes are introduced to capture high-frequency shocks and low-frequency fluctuation information of different frequencies and durations in the signal. Next, a DWconv layer of size 11 is embedded in the multilayer perceptron to enhance spatial information representation while saving computational resources. Finally, to verify the effectiveness of the method, this paper simulates and analyzes the actual working state of centrifugal fan blades, constructs a simulation dataset, and builds a centrifugal fan experimental bench to obtain a real dataset. The experimental results show that the MDC-1DCNN framework significantly outperforms the existing methods in both simulation and experimental bench datasets, fully proving its versatility and effectiveness in centrifugal fan blade fault diagnosis.