Damage detection and identification in fiber reinforced plastic structural members and field bridges using acoustic emission technique

With the increased use of fiber reinforced polymer (FRP) based structural systems for rehabilitation of existing and construction of new bridges there is a requirement for identification of critical components of these structural systems and the determination of critical damage thresholds in them. Of the many available non-destructive techniques (NDT), acoustic emission (AE) monitoring had been identified as one of the most popular techniques applicable for damage discrimination in composites. The current study aimed at using patterns in AE data for the identification of damage modes exhibited by composite structural systems. The extensive experimental program involved testing of two structural systems: (i) Reinforced concrete specimens with CFRP retrofit to study debonding failure mechanism and (ii) GFRP laminates coupon specimens tested under varied load conditions to study critical failure modes such as fiber breakage, matrix cracking, delamination and debonding. Real-time AE monitoring was also conducted for a newly installed FRP deck field bridge subjected to live load tests. The AE data collected from the bridge revealed the overall structural performance of the new bridge and helped establish baseline AE activity for future condition evaluation. The AE data acquired from all the experimental tests conducted in this research were subjected two methods of analysis. The first analysis technique involved subjecting the data to the traditional signal processing techniques and identifying various AE sources by visual observations of trends in correlation plots. Meanwhile the same dataset was analyzed using neural networks to perform pattern recognition. In this work, a methodology based on the use of an unsupervised k-means clustering to generate the learning dataset for the training of the multi-layer perceptron (MLP) classifier was developed. The method adopted here showed good results for the clustering and classification of AE signals from different sources for the specimens studied in this research. But, clustering does not always lead to a unique solution and some failure mode characteristics were more easily identifiable than others. Thus further study for enriching of the training dataset is warranted. The high performance efficiency achieved by the developed neural network model for damage identification in full scale specimens further confirms the potential of the developed methodology in being feasible for damage identification in full-scale structures