摘要
塔式起重机(塔机)的结构损伤智能诊断是其高效稳定运行的重要保障措施。然而,由于塔机严格的维护保养制度,实际损伤样本相对较少,导致基于数据驱动的模型仅能从中学习到极为局限的诊断知识,致使诊断准确率大幅下降。针对此问题,引入对比学习的特征提取思想,提出塔机小样本条件下结构损伤智能诊断方法。该方法主要通过构建核函数引导的对比学习损失函数和标准分类损失函数实现,其中,核函数引导的对比学习损失函数能通过高斯核将特征映射到无穷维空间中进行样本对度量的学习,并采用超参数控制分类边界的带宽。最后,借助通过塔机物理仿真模型实验台和真实服役塔机采集的损伤样本验证所提方法的有效性,实验结果表明所提方法在小样本条件下诊断塔机损伤相比于深度学习模型和标准对比学习模型具有鲁棒性和优越性。
Intelligent diagnosis of tower crane structure damages is an important guarantee for its efficient and stable operation.However,due to the strict maintenance system of the tower cranes,the actual damage samples are relatively few,leading to that the model based on data-driven can only learn very limited diagnosis knowledge,resulting in a significant decrease of the diagnosis accuracy.In response to this problem,based on the feature extraction concept of comparative learning was introduced,an intelligent diagnostic method of structure damage was proposed under the condition of small samples of tower cranes.This method is mainly performed by constructing a kernel guided contrastive learning loss function and a typical classification loss function.Among them,the kernel guided contrastive learning loss function can map features to an infinite dimensional space through a Gaussian kernel for sample pair metric learning,and use hyper-parameters to control the bandwidth of the classification boundary.Finally,the effectiveness of the proposed method was verified through the physical simulation model experimental platform of the tower crane and damage samples collected from real service tower cranes.The experimental results show that the proposed method has robustness and superiority in diagnosis of tower crane damage under small sample conditions compared to deep learning models and standard contrastive learning models.
作者
杨蕊
安增辉
宋世军
孟祥林
张茹
黄文武
YANG Rui;AN Zenghui;SONG Shijun;MENG Xianglin;ZHANG Ru;HUANG Wenwu(School of Mechanical and Electrical Engineering,Shandong Jianzhu University,Jinan 250101,China;China Railway 14th Bureau Group Corporation Ltd.,Jinan 250101,China;Shandong Dahan Construction Machinery Company Ltd.,Jinan 250200,China)
出处
《噪声与振动控制》
北大核心
2025年第5期131-137,共7页
Noise and Vibration Control
基金
国家自然科学基金资助项目(52005300)
山东省高等学校青创科技支持计划资助项目(2023KJ124)
山东省自然科学基金资助项目(ZR2024QE083)。
关键词
故障诊断
塔式起重机
对比学习
小样本
fault diagnosis
tower cranes
contrastive learning
small sample
