基于神经网络算法的复合材料板声发射源定位

摘要

使用Levenberg-Marquardt训练函数的神经网络算法在复合材料板上进行声发射定位,采用声波到达时间作为输入向量,训练得到了到达各个传感器的声波触发时间与板上坐标的良好的映射关系.考虑到波形衰减,用传统的固定门槛提取的到达时间容易与实际情况偏差较大,讨论了一种相对准确的适合工程使用的到达时间提取方法.%Levenberg-Marquardt training function was applied to locate the acoustic emission source on composite plate, in which the arriving time of wave was chosen as the input vector. After training, the position of the AE source is well mapped to the arriving time of each sensor. Besides, a more precise way in attaining the arriving time, which can be easily applied to engineering, is discussed here instead of the conventional way of crossing fixed threshold, which may result in some deviation of time, in view of attenuation of waves.