摘要
设计开发了基于深度学习的海洋复合材料柔性管损伤识别创新实验,以“深度学习驱动的柔性管损伤识别”为核心特色。与传统验证性实验不同,学生需在教学实验的基础上自主设计深度学习模型结构、优化超参数、并探索信号预处理方法对模型性能的影响。教学实践表明:采用VMD优化的深度学习模型使柔性管损伤识别误差最高降低至1.73%,显著改善了损伤程度的识别效果;85%的学生能够准确指出“传感器布局对特征提取具有关键影响”,并成功开发出“VMD+CNN-Transformer”混合模型,使多损伤工况识别准确率达到99.2%。通过实验,不仅让学生掌握深度学习在海洋工程中的实际应用,更培养其从问题定义到模型部署的完整创新能力,为海洋工程复合型人才培养提供了参考。
Marine pipeline is the“lifeline”of marine oil and gas transportation,and it is extremely important to recognize its damage condition and evaluate its safety performance in time.Innovative experimental teaching is of great significance to improve the comprehensive quality of students.An innovative experimental platform was designed for damage detection in marine composite flexible pipes using deep learning technology.This experiment features“deep learning-driven damage identification”as its core characteristic,distinguishes it from traditional verification experiments.Students need to independently design deep learning model structures,optimize hyperparameters,and explore the impact of signal preprocessing methods on model performance,build upon the foundational teaching experiments.The experimental results show that using VMD optimized deep learning model can reduce the maximum error of flexible pipe damage identification to 173%,significantly improving the recognition effect of damage degree.85%of students are able to accurately point out that“sensor layout has a key impact on feature extraction”,and successfully develop the“VMD+CNN Transformer”hybrid model,achieving an accuracy rate of 992%in identifying multiple damage conditions.Through this experiment,students not only master the engineering application ability of deep learning technology,but also increase their complete innovation ability from problem definition to model deployment,providing a reference for the cultivation of composite talents in ocean engineering.
作者
包兴先
郝颖奎
王骏峰
刘猛
BAO Xingxian;HAO Yingkui;WANG Junfeng;LIU Meng(College of Mechanical and Electronic Engineering,China University of Petroleum(East China),Qingdao 266580,Shandong,China)
出处
《实验室研究与探索》
北大核心
2025年第8期144-149,160,共7页
Research and Exploration In Laboratory
基金
国家自然科学基金(51979283)
国家重点研发计划(2016YFC0303800)
中国石油大学(华东)研究生教育教学改革项目(YJG2024047)。
关键词
深度学习
损伤识别
柔性管
创新实验
海洋工程
deep learning
damage identification
flexible pipe
innovative experiment
ocean engineering
