This work proposes an electro mechanical impedance based approach for detection of disbonds in a composite lap joint using both drive point and cross mechanical impedance signature. An approximate analytical model has been developed for determination of drive point and cross mechanical impedance of a composite lap joint for given location and size of disbond. Electro-mechanical impedance response of same structure fitted with piezoelectric patches is obtained through finite element simulation using ANSYS and experiment. Mechanical impedance obtained through approximate analytical model is validated with the mechanical impedance extracted from numerically and experimentally obtained electro-mechanical impedance data. Damage features based on root mean square deviation (RMSD) and correlation coefficient (CC) of the mechanical impedance for various damage scenarios with respect to the pristine state mechanical impedance measurements have been extracted. Artificial neural network (ANN) has been trained using damage features extracted from approximate analytical model based simulation. Novelty of this work lies in selection of proper features through extraction of drive point and cross mechanical impedance from electro mechanical impedance data. It has been shown that if a machine learning system is trained through such features obtained from low cost approximate analytical model, it can detect damage status from damage features extracted from finite element simulation and experimental data satisfactorily. This can significantly reduce computational effort in generating adequate training data.

中文翻译：

---

使用基于新型交叉和驱动点机械阻抗的特征，对复合材料搭接接头内的分层/脱粘进行定位和量化


这项工作提出了一种基于机电阻抗的方法，使用驱动点和交叉机械阻抗特征来检测复合材料搭接接头中的脱粘。已经开发了一个近似分析模型，用于确定给定脱粘位置和尺寸的复合材料搭接接头的驱动点和交叉机械阻抗。通过ANSYS有限元模拟和实验，获得了安装压电片的相同结构的机电阻抗响应。通过近似解析模型获得的机械阻抗与从数值和实验获得的机电阻抗数据中提取的机械阻抗进行了验证。相对于原始状态机械阻抗测量，提取了基于机械阻抗均方根偏差（RMSD）和相关系数（CC）的损伤特征，用于各种损伤场景。人工神经网络 (ANN) 使用从基于近似分析模型的模拟中提取的损伤特征进行训练。这项工作的新颖性在于通过从机电阻抗数据中提取驱动点和交叉机械阻抗来选择适当的特征。事实证明，如果通过低成本近似分析模型获得的这些特征来训练机器学习系统，它可以从有限元模拟和实验数据中提取的损伤特征令人满意地检测损伤状态。这可以显着减少生成足够训练数据的计算工作量。