摘要
列车转向架传动系统的高效安全运行对确保城市轨道交通和乘客安全至关重要。针对现有故障诊断模型计算复杂度高、可解释性不足的问题,提出了一种轻量级可解释深度诊断框架——SAKNet。首先,设计一种稀疏自注意力模块,通过聚焦序列关键元素实现注意力权重稀疏化,减少计算量和参数规模。其次,设计了一种基于Kolmogorov-Arnold Network的可解释分类器模块,基于B样条参数化的可学习激活函数动态调整函数形态,增强模型内在可解释性。实验表明,所提方法在保持轻量化的同时展现出了更好的可解释性,同时达到了98.55%的准确率。
The efficient and safe operation of train bogie transmission systems is crucial for ensuring the safety of urban rail transit and passengers.To address the problems of high computational complexity and insufficient interpretability in existing fault diagnosis models,a lightweight and interpretable deep diagnostic framework,termed SAKNet,is proposed.Firstly,a sparse self-attention module is designed to achieve attention weight sparsification by focusing on key elements in sequential data,thereby reducing computational load and parameter scale.Secondly,an interpretable classifier module based on the Kolmogorov-Arnold Network is developed,wherein dynamically adjustable function forms are realized through B-spline parameterized learnable activation functions,enhancing the intrinsic interpretability of the model.Experiment shows that the proposed method exhibits superior interpretability while maintaining lightweight characteristics,achieving an accuracy of 98.55%.
作者
杨超
陈俊富
冷贝贝
YANG Chao;CHEN Junfu;LENG Beibei(Department of Artificial Intelligence and Software Engineering,Nanyang Normal University,Nanyang,Henan 473061,China)
出处
《自动化应用》
2025年第16期82-86,共5页
Automation Application
基金
南阳市科技发展计划项目(23KJGG039)
南阳师范学院青年项目(2024QN003)。
关键词
故障诊断
自注意力机制
故障诊断框架
fault diagnosis
self-attention mechanism
fault diagnosis framework
