Systematic comparison of Artificial Neural Networks for a SHM procedure applied to Composite Structure

The problems related to damage detection represents a primary concern, particularly in the framework of composite structure. In fact, for this kind of structures barely visible damage can occur. Moreover, one of the major in-service damage of composite aircraft strcutures is represented by disbonds between the stiffeners and the skin undergoing dynamic or post-buckling loads. The effective implementation of a SHM system relies on the synthesis of non-destructive technique (NDT), fracture mechanics, sensors technology, data manipulation and signal processing, and it can receive a great improvement through the use of an Artificial Neural Networks. Different architectures of Artificial Neural Networks for a structural damage detection are studied in order to detect damage without any prior knowledge of the model of the structure so as to serve as a real-time data processor for SHM systems. The standard feed-forward Multi Layer Perceptron and the Radial Basis Function ANNs are compared. In the former paradigm each neuron is arranged in a series of layer. There are no precise rules for the choose of the number of hidden neurons, only empirical indications. Three training algorithms are considered: the Levenberg-Marquardt, the Conjugate Gradient backpropagation and the Gradient Descent with momentum. Moreover different complexity of the network are considered, by varying the number of hidden neurons. On the other hand, the RBF can require more neurons than standard MLP networks but they can be often designed to reduce the training time with respect to MLP. Two types of RBF are trained. The first can produce a network with zero error on training vectors but the number of hidden neurons is very high since it must be equal to the number of input vectors. So a second type of RBF is created by adding one neuron at each training input, until the sum-squared error falls beneath an error goal or a maximum number of neurons is reached. For the second type of ANNs instead, assumed that the number of hidden neurons is fixed due to the number of training data used to build and train all networks, a parametric analysis is made by changing the spread parameter. A drop-ply delaminated structure employed to represent a bidimensional simplification of the adhesive joints between composite aircraft fuselage skins and stiffners is here analyzed to study the sensing capability of the SHM system proposed. The analyzed configuration consists of an host delaminated structure, made up by 0° and 90° graphite-epoxy (GE) plies and of a piezoelectric patch, employed to arrange the sensing device. The skin is arranged with unidirectional graphite-expoxy plies. The drop-ply assembly deforms under plane strain conditions and it is clamped at the flange root. The structure undergoes several treansverse shear loads per unit length F acting on the free-edge of the skin. Moreover, for dynamic analyses the load is modeled as a step load to simulate the sudden raise of the shearing postbuckling action. Lastly, a delamination of length a is assumed to occur at the skin-flange 0°/90° interfaces. The piezoelectric patch is assumed to be bonded on the flange top surface. The training data are obtained, in term of a Damage Index ID distribution, from a Dual Reciprocity Boundary Element Method transient analysis of the host damaged structure and the bonded piezoelectric sensor. The BEM model allows to compute the electrical signals that are used to define the ID generated by an array of piezoelectric sensors bonded on a delaminated composite skin-stiffner. Thirty-two delamination length and twenty-one load cases are analyzed. It has been observed that the RBF NN tipically requires larger number of training patterns and also a larger network architecture to achieve the same lavel of desired accuracy as the MLP. On the other hand, an advantage of using RBF NNs is that they require a training time inferior than the MLP NN