针对舰船轮机设备故障信号监测中存在的运算量大、缺少故障数据、模型训练复杂、检测效率低、准确度不高等问题，设计了基于机器学习的舰船轮机设备多发故障信号监测方法。通过多种传感器采集舰船轮机设备振动信号，经小波变换降噪后，通过EMD经验模态分解提取舰船轮机设备振动信号特征，将其作为孤立森林算法输入进行异常信号检测，以异常信号检测结果为依据，构建决策二叉树支持向量机故障信号分类模型识别故障信号，实现舰船轮机设备多发故障信号监测，实验表明，该方法可以高效、准确地检测并识别舰船轮机设备的故障信号，而且适应性广泛，在舰船轮机设备的各种工况下，监测性能都十分良好。
In response to the problems of high computational complexity, lack of fault data, complex model training, low detection efficiency, and low accuracy in the monitoring of fault signals of ship turbine equipment, a machine learning based monitoring method for multiple fault signals of ship turbine equipment is studied. The vibration signal of marine engine equipment is collected through a variety of sensors. After denoising by wavelet transform, the characteristics of the vibration signal of marine engine equipment are extracted by EMD empirical mode decomposition, which is used as the input of the isolated forest algorithm for abnormal signal detection. Based on the abnormal signal detection results, a decision Binary tree support vector machine fault signal classification model is constructed to identify fault signals, so as to monitor multiple fault signals of marine engine equipment. The experiment shows that this method can efficiently and accurately detect and identify fault signals of ship engine equipment, and has wide adaptability. The monitoring performance is very good in various working conditions of ship engine equipment.
2023,45(16): 100-103 收稿日期：2023-3-24
DOI：10.3404/j.issn.1672-7649.2023.16.020
分类号：U676.4+2
作者简介：杨双齐(1983-),男,实验师/工程师,研究方向为船舶轮机工程技术
参考文献：
[1] 孙晋锰, 于忠清, 王海崖. 基于机器学习算法的滚动轴承在线故障诊断[J]. 青岛大学学报(自然科学版), 2021, 34(2): 15–22
[2] 余永华, 陈育成. 基于瞬时转速和机器学习的船用柴油机健康状态评估[J]. 内燃机工程, 2020, 41(6): 101–106
[3] 彭道刚, 裴浩然, 尹德斌, 等. 基于机器学习的燃气轮机控制系统分层故障诊断[J]. 热能动力工程, 2022, 37(12): 165–173
[4] 俞晓丰, 帅长庚, 杨雪, 等. 船舶水润滑轴承振动状态的光纤传感监测及信号分解[J]. 船舶工程, 2021, 43(5): 22–28
[5] 孙原理, 宋志浩. 基于多物理场信号相关分析与支持向量机的离心泵故障诊断方法[J]. 振动与冲击, 2022, 41(6): 206–212
