摘要
盾构机(tunnel boring machine, TBM)滚刀在重载、冲击和地质复杂的环境中服役,极易发生偏磨等失效故障,因此,掌握滚刀的磨损状态、实现基于数据驱动的滚刀偏磨故障诊断并指导滚刀的运维尤为重要。提出了一种基于小波时频分析和Inception-BiGRU模型的诊断模型以提高滚刀偏磨故障诊断效率。以滚刀为研究对象,在多功能缩比滚刀试验台上进行直线破岩试验,采集滚刀破岩时产生的振动加速度信号。采用连续小波变换获取反映振动信号时频域特征的小波时频图,进而以Inception模块的不同大小卷积核自适应地提取时频图中的多尺度空间信息,并通过添加双向门控循环单元(bidirectional gated recurrent units, BiGRU)使模型可更为准确地学习到时频图中丰富的时序依赖性关系,模型的超参数由贝叶斯优化算法确定。4种不同偏磨程度滚刀的诊断试验表明所提模型能够有效提取时频图中滚刀的偏磨特征并完成滚刀偏磨状态识别,实现端到端的盾构滚刀偏磨故障诊断。模型平均诊断准确率可达到98.5%,其诊断准确度和稳定性均优于其他常用算法,证明了所提方法的可行性。
Tunnel boring machine(TBM)hobs are easy to have failure faults of eccentric wear,etc.during serving in heavy load,impact and complex geological environments.Therefore,it is particularly important to master wear status of hobs,realize data-driven hob fault diagnosis of partial wear,and guide operation and maintenance of hobs.Here,a diagnosis model based on wavelet time-frequency analysis and Inception-BiGRU model was proposed to improve the efficiency of hob fault diagnosis of partial wear.Taking a hob as the study object,straight line rock-breaking tests were conducted on a multi-function scaled hob test rig to collect vibration acceleration signals generated by hob when it breaking rock.The continuous wavelet transform was used to obtain wavelet time-frequency maps to reflect time-frequency characteristics of vibration signals,and then multi-scale spatial information in time-frequency maps was adaptively extracted using different convolution kernels of Inception model.By adding a bidirectional gated recurrent unit(BiGRU),Inception-BiGRU model could more correctly learn rich time sequences-dependency relations in time-frequency maps.Hyperparameters of the model were determined with Bayesian optimization algorithm.Diagnosis tests for 4 types of hobs with different degrees of eccentric wear showed that the proposed model can effectively extract eccentric wear characteristics of hobs from time-frequency maps and complete identification of hob eccentric wear status to realize end-to-end TBM hob fault diagnosis of partial wear;the average diagnostic accuracy of the model can reach 98.5%,its diagnostic accuracy and stability are superior to other commonly used algorithms to verify the feasibility of the proposed method.
作者
樊翔翔
项载毓
孙瑞雪
张敏
莫继良
FAN Xiangxiang;XIANG Zaiyu;SUN Ruixue;ZHANG Min;MO Jiliang(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China;Tribological Design Laboratory of Shield/TBM Equipment,Southwest Jiaotong University,Chengdu 610031,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2023年第15期232-240,共9页
Journal of Vibration and Shock
基金
国家自然科学基金(51822508)
四川省科技计划项目(2020JDTD0012)
四川省科技重点研发项目(21ZDYF3658)。
