基于振动信号图像编码与深度学习的柴油机故障识别方法

车用发动机 ›› 2026, Vol. 0 ›› Issue (1) : 88-94.

李祥利¹，臧建淋²，徐冉³，马杰¹，于涛¹，陈东峰¹

作者信息 +

Fault Identification Method of Diesel Engine Based on Vibration Signal Image Encoding and Deep Learning

LI Xiangli¹,ZANG Jianlin²,XU Ran³,MA Jie¹,YU Tao¹,CHEN Dongfeng¹

Author information +

文章历史 +

摘要

柴油机振动信号具有典型的非线性、非平稳特征,基于振动信号的柴油机故障精准识别存在诸多挑战。为此，提出一种基于振动信号的端到端的深度学习网络框架（ResNet-ConvLSTM-Self-Attention），该框架结合了多种模型提取鲁棒特征的优势，无需手动提取特征，通过将振动数据流图像编码输入模型，可实现柴油机三种典型故障的有效识别。对该框架模型进行了超参数优化、故障识别评估以及有无注意力机制比较，获得了98.38%的识别准确率，验证了该柴油机故障识别模型的有效性。

Abstract

Due to the typical nonlinear and non-stationary characteristics of diesel engine vibration signals, there are many challenges in the precise identification of diesel engine faults based on vibration signals. An end-to-end deep learning network framework (ResNet-ConvLSTM-Self-Attention) was proposed based on vibration signals. This framework combined the advantages of multiple models in extracting robust features without manual design. By encoding the vibration data stream images into the model, the framework can effectively identify three typical faults of diesel engines. The framework model underwent hyperparameter optimization, fault identification evaluation, and comparison with and without attention mechanisms, and an identification accuracy of 98.38% was achieved. The research results verified the effectiveness of the diesel engine fault identification model.

导出引用

李祥利, 臧建淋, 徐冉, 马杰, 于涛, 陈东峰. 基于振动信号图像编码与深度学习的柴油机故障识别方法[J]. 车用发动机. 2026, 0(1): 88-94

LI Xiangli, ZANG Jianlin, XU Ran, MA Jie, YU Tao, CHEN Dongfeng. Fault Identification Method of Diesel Engine Based on Vibration Signal Image Encoding and Deep Learning[J]. Vehicle Engine. 2026, 0(1): 88-94