吴劲松，徐志京

(上海海事大学 信息工程学院，上海 201306)

基于BP_Adaboost神经网络的船舶桨叶故障预警模型*

吴劲松，徐志京

(上海海事大学 信息工程学院，上海 201306)

为了提高船舶桨叶故障识别率，构建了一种基于BP_Adaboost神经网络的船舶桨叶故障预警模型。该模型首先采用数字水听器采集二进制船舶桨叶静水噪声信号，并利用MATLAB编程将二进制信号转换成WAV音频信号，通过梅尔频率倒谱系数法(Mel-Frequency Cepstral Coefficient, MFCC)提取特征值，得到36维桨叶静水噪声信号，最后运用BP_Adaboost神经网络进行分类识别预警。实验结果表明，基于BP_Adaboost神经网络的船舶桨叶故障预警模型能够高效分类预警船舶桨叶故障，与BP神经网络识别率对比，分类识别预警率高达96%。

船舶桨叶故障预警模型；数字水听器；MFCC；BP_Adaboost

Abstract: In order to improve the fault recognition rate of ship blades, a fault prediction model based on BP_Adaboost neural network is proposed in this paper. The model uses digital hydrophone acquisition to collect binary data of ship hydrostatic blade noise, and uses MATLAB to convert binary signal into WAV audio format. Then the model uses Mel-Frequency Cepstral Coefficient (MFCC) feature extraction method to extract features and gets the water noise signal in the 36 dimension, and finally uses BP_Adaboost neural network to classify and warn the fault of ship blades. The experimental result shows that the fault prediction model based on BP_Adaboost neural network can effectively classify and predict the failure of the ship blades compared with the BP neural network. The classification and recognition warning rate is as high as 96%.

Key words:ship blades faults warning model; digital hydrophone acquisition ;MFCC；BP_Adaboost

0 引言

船舶是重要的水上交通工具，螺旋桨是船舶将发动机力矩转换成动能的主要装置[1]，主机输出转矩通过轴系带动螺旋桨旋转，螺旋桨产生的推力推动船舶航行。由于各种不同原因造成螺旋桨折断、缺损，桨叶故障会导致主机故障，严重影响了船舶的航行安全[2]，因此研究船舶桨叶故障具有理论与现实意义。传统的船舶故障诊断法有四种：直接测量法、数字测量法、故障树分析法、专家诊断法[3]。近年来神经网络算法的兴起，为船舶桨叶故障预警提供了新的思路。相对于传统的船舶桨叶故障诊断法，神经网络能够及早分类识别预警船舶桨叶故障，为船舶故障诊断排除提供了可靠依据，对于船舶运输以及国防科研具有重要意义。……