Multiple ultrasonic-guided wave signals generated by wave propagation model: focused on delamination damage location variations in cross-ply composite beams

&lt;p&gt;The ultrasonic guided-wave data has been used in the paper “&lt;i&gt;Deep learning uncertainty quantification for ultrasonic damage identification in composite structures&lt;/i&gt;” in Composites Structures. &lt;p&gt;The raw ultrasonic signals have had their consistent excitation component removed, leaving only the sections that exhibit the main variations. The amplitude of all waveform signals has been normalized to the range of [-1,1]. Additionally, the damage location has been normalized to the range of [0,1] in increments of 5mm, covering a distance of 30cm to 70cm. This is to avoid any potential effects of scale differences. &lt;p&gt;The datasets used for categorized as OOD (three noise levels in signal-to-noise ratio (SNR). ) and ID datasets: &lt;p&gt;&lt;strong&gt;OOD datasets:&lt;/strong&gt;&lt;br&gt; （1）The 9 waveform signals within each normalized damage location range of [0.2, 0.3] correspond to an SNR of 10 dB (10dB normalized delamination damage position at 0.2 to 0.3 and 5dB at 0.7 to 0.8.csv);&lt;br&gt; （2）The 9 waveform signals within each normalized damage location range of [0.7, 0.8] correspond to an SNR of 5 dB (10dB normalized delamination damage position at 0.2 to 0.3 and 5dB at 0.7 to 0.8.csv);&lt;br&gt; （3）The 81 waveform signals within each normalized damage location range of [0, 1] correspond to a noise-free one (Noise-free normalized delamination damage position at 0 to 1.csv).&lt;br&gt; &lt;p&gt;&lt;strong&gt;ID datasets:&lt;/strong&gt;&lt;br&gt; The 81 waveform signals within each normalized damage location range of [0, 1] correspond to the averaged value of the 30 times at each damage location with random white Gaussian noise SNRs of 10 dB (Averaged 10dB normalized delamination damage position at 0 to 1.csv).&lt;br&gt; The average value of the 30 waveform signals is computed at each damage location after normalization to simulate the real experimental de-noising. &lt;p&gt;Ultrasonic guided-wave data generated from transient ultrasonic guided wave propagation model considering composite beam structures. The parameters of a wave propagation model are generated by performing FEM simulations using Abaqus. FEM notes: solid elements (i.e. C3D8R in Abaqus) with the following dimensions: 0.5mm thick, 0.5mm wide, and 0.2mm long. Damage modes are modelled by node duplication. Ultrasonic tests data: 6-cycle Hanning windowed sinusoid centered at 100 kHz of frequency. Excitation signal at one end of the beam in the vertical direction. Sensor situated at 0.2m from the point of excitation and measuring for 720 \mu s. The first antisymmetric mode (A0) is excited. Cross-ply carbon fibre composite beam (dimensions: 1m x 3mm x 2mm) with [0_{2}/90]_s stacking sequence and delamination (5mm long at &lt;i&gt;L&lt;/i&gt; mm from excitation) in it, where &lt;i&gt;L&lt;/i&gt; is the damage location.<p>本超声导波数据集已应用于发表于《Composites Structures》期刊的论文《复合材料结构中超声损伤识别的深度学习不确定性量化》（Deep learning uncertainty quantification for ultrasonic damage identification in composite structures）。</p> <p>原始超声信号已去除一致的激励分量，仅保留展现主要变化的信号片段。所有波形信号的幅值已归一化至[-1,1]区间。此外，损伤位置以5mm为步长归一化至[0,1]区间，覆盖距离范围30cm至70cm，以消除尺度差异可能带来的潜在影响。</p> <p>本数据集分为分布外（OOD，Out-of-Distribution）数据集与分布内（ID，In-Distribution）数据集：</p> <p><strong>OOD数据集:</strong><br> （1）归一化损伤位置区间[0.2, 0.3]内的9条波形信号，对应信噪比（SNR，Signal-to-Noise Ratio）为10dB，对应数据文件为10dB normalized delamination damage position at 0.2 to 0.3 and 5dB at 0.7 to 0.8.csv;<br> （2）归一化损伤位置区间[0.7, 0.8]内的9条波形信号，对应信噪比为5dB，对应数据文件为10dB normalized delamination damage position at 0.2 to 0.3 and 5dB at 0.7 to 0.8.csv;<br> （3）归一化损伤位置区间[0, 1]内的81条波形信号，对应无噪声场景，对应数据文件为Noise-free normalized delamination damage position at 0 to 1.csv.<br> </p> <p><strong>ID数据集:</strong><br> 归一化损伤位置区间[0, 1]内的81条波形信号，对应每个损伤位置经30次采样后取平均值的结果，叠加随机高斯白噪声，信噪比为10dB，对应数据文件为Averaged 10dB normalized delamination damage position at 0 to 1.csv.<br> 为模拟真实实验的去噪过程，我们先对每个损伤位置的30条波形信号进行归一化，再计算其平均值。 </p> <p>本超声导波数据集基于考虑复合梁结构的瞬态超声导波传播模型生成。波传播模型的参数通过Abaqus软件开展有限元（FEM，Finite Element Method）仿真得到。<br> 有限元设置说明：采用实体单元（Abaqus中型号为C3D8R），尺寸参数为：厚度0.5mm、宽度0.5mm、长度0.2mm。损伤模式通过节点复制法建模。<br> 超声测试参数如下：采用中心频率为100kHz的6周期汉宁（Hanning）窗调制正弦信号作为激励信号，沿垂直方向施加于梁的一端。传感器设置于距激励点0.2m处，采样时长为720μs，用于激励一阶反对称模态（A0模态）。<br> 试验采用正交铺层碳纤维复合梁，尺寸为1m×3mm×2mm，铺层顺序为[0₂/90]_s，梁内设置分层损伤，损伤长度为5mm，距激励点的距离为L mm，其中L为损伤位置参数。</p>

搜集汇总

背景与挑战

背景概述

该数据集包含用于深度学习不确定性量化研究的超声波导波信号，专注于交叉层合复合材料梁中分层损伤位置的变化。数据通过有限元模拟生成，信号经过归一化处理，并分为OOD（不同信噪比水平）和ID数据集，以支持损伤识别和不确定性分析。