基于复Morlet变换和改进AlexNet神经网络的柴油机气门间隙异常故障诊断方法

A fault diagnosis method for diesel engine valve clearance abnormality based on a complex Morlet transform and an improved AlexNet neural network

针对柴油机缸盖振动信号非线性、非平稳的特点,以及传统故障诊断方法需要先验知识且特征提取费时费力的缺点,提出了一种基于复Morlet变换和改进AlexNet神经网络的柴油机气门间隙异常故障诊断方法。首先通过复Morlet小波将柴油机缸盖振动信号转换为时频图,该变换包含了信号的时频域信息,比单一的时域或频域信号更适合分析柴油机缸盖振动这种非平稳信号;其次将时频图输入至AlexNet神经网络进行特征自动提取并建立故障诊断模型,解决了传统手工提取特征费时费力且需要专家经验的问题;然后通过Batch Normalization和Dropout技术改进网络结构,并优化神经网络超参数以提高模型的准确度和计算效率;最后将本文方法与传统的故障诊断方法应用于柴油机气门间隙异常故障诊断并进行对比,发现其诊断准确率最高,验证了所提方法的优越性。

In the light of the non-linear and non-stationary characteristics of diesel engine cylinder head vibration signals,as well as the shortcomings of time-consuming and laborious traditional fault diagnosis methods that require prior knowledge and feature extraction,a diesel engine valve clearance abnormal fault diagnosis method based on a complex Morlet transform and an improved AlexNet neural network is proposed.Firstly,the diesel engine cylinder head vibration signal is converted into a time-frequency map by a complex Morlet wavelet.This transformation contains the time-frequency domain information of the signal,which is more suitable for analyzing the non-stationary signal of diesel engine cylinder head vibration than a single time-domain or frequency-domain signal.Secondly,the time-frequency graph is input into the AlexNet neural network for automatic feature extraction,and a fault diagnosis model is established,which solves the problems of traditional manual feature extraction that is time-consuming and laborious as well as requiring expert experience.Thirdly,the network structure is improved through Batch Normalization and Dropout technology,and the neural network is optimized using hyperparameters to improve the accuracy and computational efficiency of the model.Finally,it is shown that the diagnosis accuracy of the method proposed in this paper is higher than the traditional fault diagnosis method,which verifies its advantages.