对抗神经深度网络与信号重构技术对混凝土梁应力波损伤信号的模糊聚类

GENERATIVE ADVERSARIAL NETWORKS WITH FUZZY CLUSTERING ON DAMAGE DETECTION OF RC BEAMS USING PIEZOCERAMIC SENSING SIGNALS

作为嵌入式的多功能压电换能器,球形智能骨料具有频带宽、响应迅速、与混凝土互融的优点。目前,基于该球形智能骨料的损伤识别方法已经被逐渐用于混凝土结构的健康监测,通过将损伤工况下获取的信号与基准信号相比来量化结构损伤。为了表征所监测结构在不同阶段的损伤程度,需要利用小波分解法处理采集到的压电信号,从而间接得到相应的损伤指数。该过程将所有时域信号进行n阶小波分解换算成能量,最后通过计算根均方误差来表征结构的损伤因子。然而,整个换算过程需要消耗较大的算力对近百组的时域信号进行换能计算,加上信号采集频率高,减缓了整个计算过程,使得整个监测流程具有滞后性。针对以上问题,该文提出了一种基于对抗神经网络与监测信号重构的半监督监测方法,可以根据重构信号与初始健康信号在频域的差异提取损伤敏感特征,同时优化模糊聚类的伪标签传播,通过读取监测信号间接评估结构的损伤状态。所建立结构状态评估模型的优势在于通过“致损前训练验证、致损后快速预测”的逻辑,将耗时较长的训练、验证环节前置,保障后期预测评估的高效性。

As an embedded multifunctional piezoelectric transducer,smart aggregate has the advantages of wide frequency band,fast response,and integration with concrete.To characterize the damage degree at different stages,it is necessary to use the wavelet decomposition method to process the collected signal.All time domain signals are converted into energy by n-order wavelet decomposition,and finally the damage factor is characterized by calculating the root mean square error.However,the process needs to consume a large amount of computing power for nearly hundreds of groups of signals.Additionally,the high frequency of signal acquisition slows down the calculation process and makes the entire monitoring process lag.To overcome the above problems,this paper proposes a semi-supervised monitoring method based on adversarial neural network and monitoring signal reconstruction,which can extract damage-sensitive features according to the difference between the reconstructed signal and the original health signal in the frequency domain and optimize the pseudo-clustering of fuzzy clustering.The tag propagates,thereby indirectly assessing the damage status of the structure by reading the monitoring signal.The advantage of the established structural state evaluation model is that through the logic of training and verification before the loss,and the rapid prediction after the loss,the time-consuming training and verification links are pre-positioned to ensure the efficiency of the later prediction and evaluation.