为了提高船舶蒸汽动力辅助系统的运行安全性和可靠性，减少系统故障导致的船舶停航和潜在安全事故，提出一种结合长短期记忆网络（Long Short-Term Memory network, LSTM）和阈值法的故障预警方法。首先，利用LSTM模型处理船舶蒸汽动力辅助系统的历史运行数据、学习系统的动态行为和潜在故障模式。其次，通过设定阈值法，结合LSTM模型的预测输出，实现对系统状态的实时监控和故障预警。最后，基于MINIS一体化仿真平台开发的蒸汽动力辅助系统模型，以汽轮给水泵转速故障为例，进行故障预警实验，验证方法的有效性。实验结果显示，结合LSTM的预测能力和阈值法的决策效率，提出的预警模型能有效识别并预警汽轮给水泵的转速故障。通过对比实际故障数据，模型在预测准确性和预警及时性方面均表现出色。该方法不仅提高了故障预警的准确性，而且为船舶维护和安全管理提供了有力决策支持，为类似工业系统的故障预警提供了新的研究思路。
In order to enhance the operational safety and reliability of marine steam power auxiliary systems, and to reduce ship downtime and potential safety incidents caused by system failures, a fault warning method combining Long Short-Term Memory (LSTM) networks and threshold methods is proposed. Initially, the LSTM model is utilized to process the historical operation data of marine steam power auxiliary systems, learning the system's dynamic behaviors and potential failure patterns. Subsequently, by setting threshold methods in conjunction with the predictive outputs of the LSTM model, real-time monitoring and fault warning of the system state are achieved. Finally, fault warning experiments are conducted using a steam power auxiliary system model developed on the MINIS integrated simulation platform, taking the example of a steam turbine feedwater pump speed fault, to verify the effectiveness of the method. The experimental results demonstrate that the warning model, integrating the predictive capabilities of LSTM with the decision-making efficiency of the threshold method, can effectively identify and warn against speed faults of the steam turbine feedwater pump. By comparing with actual fault data, the model exhibits excellent performance in terms of prediction accuracy and timeliness of warnings. This method not only improves the accuracy of fault warnings but also provides robust decision support for marine maintenance and safety management. Furthermore, it offers a novel research perspective for fault warning in similar industrial systems.
2024,46(13): 150-157 收稿日期：2024-03-18
DOI：10.3404/j.issn.1672-7649.2024.13.027
分类号：U672.9
基金项目：国家自然科学基金资助项目（51576207）
作者简介：梁聚伟(1996-),男,助教,研究方向为模拟训练条件
参考文献：
[1] 杨元龙, 吴金祥, 郑子都, 等. 汽轮给水机组切换工况下船舶蒸汽动力系统的响应特性[J]. 中国舰船研究, 2019, 14(2): 150-155.
YANG Yuanlong, WU Jinxiang, ZHENG Zidou, etal. Response characteristics of ship steam power system under switchover of turbinefeed water pump[J]. Chinese Journal of Ship Research, 2019, 14(2): 150-155.
[2] 王瑞涵, 陈辉, 管聪. 基于机器学习的船舶机舱设备状态监测方法[J]. 中国舰船研究, 2021, 16(1): 158-167.
WANG Ruihan, CHEN Hui, GUAN Cong. Condition monitoring method for marine engine room equipment based on machinelearning[J]. Chinese Journal of Ship Research, 2021, 16(1): 158–167.
[3] 王鑫, 吴际, 刘超, 等. 基于LSTM循环神经网络的故障时间序列预测[J]. 北京航空航天大学学报, 2018, 44(4): 772-784.
[4] KUMAR B R, RAMANA K V, RAO K M. Condition monitoring and fault diagnosis of a boiler feed pump unit[J]. Strosnicky Casopis, 2009,60(1): 19–32.
[5] KURIEN C, SRIVASTAVA A K. Case study on the effectiveness of condition monitoring techniques for fault diagnosis of pumps in thermal power plant[J]. Mechanics & Mechanical Engineering, 2019, 23(1): 70–75.
[6] YOUNG A, WEST G, BROWN B, et al. Symbolic representation of knowledge for the development of industrial fault detection systems[C]//International Congress and Workshop on Industrial AI. Cham: Springer International Publishing, 2021: 307–318.
[7] 苏健, 宋汉江, 宋福元, 等. 基于卷积神经网络的蒸汽动力系统故障诊断[J]. 中国舰船研究, 2022, 17(6): 96-102.
SU Jian, SONG Hanjiang, SONG Fuyuan, et al. Fault diagnosis of steam power system based on convolutional neural network[J]. Chinese Journal of Ship Research, 2022, 17(6): 96–102.
[8] 王世威, 甘辉兵, 胡国彤, 等. 基于粒子群优化混合神经网络的船舶辅锅炉故障诊断[J]. 中国航海, 2022, 45(2): 37-42+55.
[9] 郑鑫, 陈辉, 管聪, 等. 基于SVM-BP集成学习的舰船蒸汽动力系统故障诊断[J]. 舰船科学技术, 2023, 45(5): 97-101.
ZHENG Xin, CHEN Hui, GUAN Cong,et al. Fault diagnosis of ship steam power system based on SVM-BP integrated learning[J]. Ship Science and Technology, 2023, 45(5): 97-101.
[10] 李东亮, 程刚, 耿江华. 基于汽机流网一体化的辅助汽轮机建模与仿真[J]. 船电技术, 2019, 39(3): 21-25.
