ABSTRACT
ABSTRACT The complexity of damage detection, assessment, and prognosis has, in recent years, led to the search for data evaluation tools such as artificial intelligence tools to enable the identification of trends and features of a structural response to damage. This sort of feature in any structural health monitoring (SHM) system has always been desirable because of the robustness characteristics required in any deployable SHM. The work presented in this chapter outlines in detail the research methodology used to investigate the use of artificial neural networks (ANNs) applied to static strain response on a physical structural member. A suite of preprocessing algorithms was developed for the ANN to function adequately. A brief description of the threefold validation technique used to decide on the type of learning rule and network architecture is given. The entire study was conducted on a thick-sectioned glass-reinforced polymer composite T-joint-one similar to that used in full-scale marine hull construction. The finite element analysis was also conducted by placing delaminations of different sizes at various locations in two structures, a composite beam and a T-joint. Glass fiber-reinforced polymer T-joints were then manufactured and tested, thereby verifying the accuracy of the finite element analysis results experimentally. The SHM system was found to be capable of not only detecting the presence of multiple delaminations in a composite structure but also determining the location and extent of all the delaminations present in the T-joint structure, regardless of the load (angle and magnitude) acting on the structure. This SHM system necessitated the development of a novel preprocessing algorithm, Damage Relativity Assessment Technique, along with a pattern recognition tool, ANN, to predict and estimate the damage. Another program developed-the Global Neural Network Algorithm for Sequential Processing of Internal Sub Networks-uses multiple ANNs to render the SHM system
3.4.4 Variable Loading Angle and Magnitude ...................................... 57 3.4.4.1 Training Set ........................................................................ 57 3.4.4.2 Test Set ................................................................................ 59
3.4.5 DRAT Modification .......................................................................... 59 3.4.5.1 Performance of the Modified DRAT ..............................60 3.4.5.2 Anomalies Due to Crack Closure ...................................60 3.4.5.3 ANN Test Results: Using MDRAT..................................60
