Damage Diagnosis in Semiconductive Materials Using Electrical Impedance Measurements
49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference 16th AIAA/ASME/AHS Adaptive Structures Conference 10t
Recent aerospace industry trends have resulted in an increased demand for real-time, effective techniques for in-flight structural health monitoring. A promising technique for damage diagnosis uses electrical impedance measurements of semiconductive materials. By applying a small electrical current into a material specimen and measuring the corresponding voltages at various locations on the specimen, changes in the electrical characteristics due to the presence of damage can be assessed. An
... be assessed. An artificial neural network uses these changes in electrical properties to provide an inverse solution that estimates the location and magnitude of the damage. The advantage of the electrical impedance method over other damage diagnosis techniques is that it uses the material as the sensor. Simple voltage measurements can be used instead of discrete sensors, resulting in a reduction in weight and system complexity. This research effort extends previous work by employing finite element method models to improve accuracy of complex models with anisotropic conductivities and by enhancing the computational efficiency of the inverse techniques. The paper demonstrates a proof of concept of a damage diagnosis approach using electrical impedance methods and a neural network as an effective tool for in-flight diagnosis of structural damage to aircraft components. 2 Accurate and timely assessment of airframe damage is a critical element of aviation safety, but analytical methods for diagnosing damage are often too computationally intensive to be suitable for in-flight assessment of damage. Direct analytical solutions to damage diagnosis problems can require hours or days of computational time on computers suitable for meeting the weight and power constraints of today's aircraft. Non-analytical (data-driven) techniques offer a computationally efficient alternative to analytical methods when an analytical solution to the problem does not exist or is not feasible. Even when an analytical solution exists, its inverse may not. Consider a simple mathematical function, such as the sine function, which can be applied to an independent variable, x, to compute the dependent variable, y. In this case, there exists an inverse function, sin -1 , which can be applied to y and should yield the initial value: x. However, not all mathematical functions have inverses that can be obtained by direct computation, especially for relationships with many independent and dependent variables. In these cases, approximations must be used. Inverse approaches are effective for applications where exact analytical solutions are not known or are too complex to be practical. One such application is the in-flight identification and characterization of structural damage in airframe components. Techniques such as the finite element method (FEM) provide a forward analytical solution to the computation of physical properties (such as strains, displacements, and thermoelectrical characteristics) given a known damage state. However, there are no analytical techniques for solving the inverse problem: estimating damage from physical properties. In such cases, inverse approaches are warranted. While some inverse methods are too computationally intensive, artificial neural networks (ANN) can be an effective and efficient approach for solving inverse problems. As described later in this paper, neural networks are "trained" to estimate the desired outputs from a set of inputs, while minimizing overall error. Although the training process can be quite arduous and time-consuming, a fully trained network can be quite efficient at estimating solutions to inverse problems. This paper presents an effective and computationally efficient electrical impedance measurement-based method for in situ detection and characterization of airframe damage that is applicable to electrically semiconductive materials. Such materials are often used to provide shielding from sources of electromagnetic interference (EMI). By applying a small electrical current into a material specimen and measuring the corresponding electrical parameters such as voltage and resistance at various locations on the specimen, changes in these electrical characteristics due to the presence of damage can be detected. An artificial neural network uses these changes in electrical properties to provide an inverse solution that predicts the location and magnitude of the damage. The advantage of the electrical impedance method over other damage diagnosis techniques is that it uses the specimen itself as the sensor, requiring only simple voltage measurements instead of using discrete sensors, reducing weight and system complexity. This research paper introduces a damage diagnostic approach using electrical impedance measurements and presents relevant background material on previous work using this technique. Next, the damage diagnosis approach is presented using FEM models to estimate electrical properties based on known damage conditions. The paper describes how the neural network uses these electrical property estimates to compute the inverse relationship: estimating damage characteristics from electrical properties. Next, the paper describes the testing and experimental demonstration of proof of concept for the damage diagnosis methodology. The paper concludes with a presentation of the results of this research effort. Figure 9. Grid method for damage visualization.