基于声发射和神经网络的木材受力损伤过程检测

Damage process testing of wood under loads based on acoustic emission and neural network

对受力木材的声发射信号进行检测与分析,实现木材损伤过程的检测和监控;研究过程中采用时间序列结合神经网络建模方法,对声发射信号的累积能量时间序列和载荷时间序列进行仿真和预测。利用神经网络建模对声发射信号的累积能量时间序列进行预测,模型对44个样本进行检验的最大误差为5.6%,而且误差较大的样本对应声发射振幅参数值的局部极大值,而振铃计数与声发射率则对应局部极小值。对相应载荷时间序列预测分析中,模型输出与目标输出的最大误差不到0.1%。结构为5×5×1的网络能很好地根据序列前5个值准确预测即将发生的声发射累积能量值;结构为6×5×1的网络能很好地根据序列前6个值精确预测木材即将承受的载荷。

Acoustic emission（AE）signals of wood under loads were tested and analyzed to realize nondestructive testing and monitoring.In this study,time serial and neural network（NN） are employed to simulate and predict the cumulative energy of AE parameter and corresponding loads.A neural network model is used to predict the time serial of AE cumulative energy.The maximum error of 44 verification samples is 5.6%,which indicates the good generalization capacity of the trained model.And it appears that the AE amplitude of the samples with relatively large error corresponds to its local maximum,however the ring count and AE rate correspond to the local minimum.In the prediction of the time serial of loads,the maximum error between the neural network model output and object output for 53 verification samples is less than 0.1%,which indicates the perfect generalization capacity of the trained model.The NN models with the structure of 5×5×1 and 6×5×1 can predict the time serial of AE cumulative energy and loads accurately.