基于CBAM-CNN的高速列车制动闸片摩擦块偏磨状态监控

为了解决高速列车制动闸片偏磨状态特征提取困难的问题，提出了一种基于二维图像识别的多尺度卷积和卷积注意力模块(CBAM)结合的状态监控模型. CBAM分别对数据的通道和空间给予不同的关注度以提取出关键的信息，使模型能够准确的对偏磨的状态特征进行提取. 模型的多尺度卷积模块则是增加模型对卷积核尺度的适应性，以提取到更加丰富的特征. 此外，为了防止模型训练过程中梯度爆炸和梯度消失的现象，在模型中加入残差连接. 最后，将所提方法应用于自行研制的高速列车自制试验台得到的制动闸片偏磨数据集，实验结果表明，该模型不仅能够准确有效的识别制动闸片各种偏磨状态，平均准确率达到99.89%，而且也具有较强的稳定性.

To address the challenge of feature extraction for the partial wear state of high-speed train brake pads, this paper proposes a condition monitoring model that combines multi-scale convolution and Convolutional Block Attention Module (CBAM) based on two-dimensional image recognition. CBAM assigns different attention weights to the channel and spatial dimensions of the data respectively to extract key information, enabling the model to accurately capture the feature characteristics of the partial wear state. The multi-scale convolution module of the model enhances the model's adaptability to convolution kernel sizes, thereby extracting more abundant features. Additionally, residual connections are added to the model to prevent gradient explosion and gradient vanishing during the training process. Finally, the proposed method is applied to the brake pad partial wear dataset collected from a self-developed high-speed train test bench. The experimental results demonstrate that the model can not only accurately and effectively identify various partial wear states of brake pads, with an average accuracy of 99.89%, but also exhibits strong stability.

搜集汇总

背景与挑战

背景概述

该数据集聚焦于高速列车制动闸片摩擦块的偏磨状态监控，提供基于二维图像识别的偏磨数据，用于训练和验证结合多尺度卷积、卷积注意力模块（CBAM）及残差连接的深度学习模型。数据集旨在解决偏磨特征提取困难的问题，模型实验结果显示平均准确率达99.89%，具有高稳定性和实用性，适用于摩擦学领域的状态监控研究。数据集以开放获取方式发布，便于学术和应用开发使用。