Damage detection in reinforced concrete square slabs using modal analysis and artifical neural network

Tools

Ahmed, MS, 2016. Damage detection in reinforced concrete square slabs using modal analysis and artifical neural network. PhD, Nottingham Trent University.

[thumbnail of 1- PhD Thesis_Mezgeen Ahmed (2016).pdf]

Preview

Text
1- PhD Thesis_Mezgeen Ahmed (2016).pdf - Published version
Download (9MB) | Preview

Abstract

Reinforced concrete (RC) structures are usually subjected to various types of loadings, such as permanent, sustained and transient during their lifetime. Reinforced concrete slabs are one of the most fundamental structural elements in buildings and bridges, which might be exposed to unfavourable conditions such as, impaired quality control, lack of maintenance, adverse environmental effects, and inadequate initial design. Therefore, the resistant capacity of the affected elements would dramatically be reduced which most likely leads to the partial or whole collapse of the structure. Non-destructive testing (NDT) techniques can be used to inspect for defects without further damaging the tested component. Significant research and development have been conducted on the performance of vibration characteristics to identify damage in different types of structures. The vibrations based damage detection methods, particularly modal based methods, are found to be promising in evaluating the health condition of a structure in terms of detection, localisation, classification and quantification of the potential damage in the structure. Damage in composites and the non-homogeneous material is tricky to assess from a surface inspection alone. Although the development of NDTs, especially experimental modal analysis (EMA), has been pushed forward by the aerospace industry where composites materials are employed in many safety critical applications, EMA is not widely employed to diagnose all types of RC structural members. Damage detection in reinforced concrete square slabs is the primary aim of this study. This is achieved experimentally using experimental modal analysis (EMA) and numerically using finite element method (FEM). Artificial neural network (ANN) is also used in this study to classify the void sizes. A whole testing procedure of EMA on freely supported slab was established in this research. It is based on impact hammer technique, as a relevant excitation source for field measurements. After the quality of the measurements had been ensured, the experimental data was collected from four pairs laboratory-scale reinforced concrete slabs modelled with various ranges of parameters. After collecting data, Matlab software was employed to obtain modal parameters, such as natural frequencies, mode shapes and modal damping ratios from two RC square slabs. EMA and FEM studies were undertaken to assess and improve modelling technique for capturing the aim. FEM was used to model the RC slabs using commercial ANSYS software. To balance model simplicity of RC slabs with the ability to reliably predict their dynamic response, both predicted and measured dynamic results were compared to ensure that the analytical model represents the experimental results with reasonable accuracy. ANSYS software was also employed to numerically extract the natural frequencies of the slab. Then, using Matlab software, the extracted natural frequencies were fed as the input to the ANN to classify the void sizes in the slab. The dynamic properties of the slab were investigated for each of four pairs to evaluate modal parameters (natural frequencies, damping ratio and mode shapes) sensitivity to slab's dimensions, degree of damage owing to incremental loading and induced void. The performance of EMA based on impact hammer technique was credibly tested and verified on measurements, which were collected from eight slabs with various parameters. EMA efficiency was conclusively proved on data from modal parameters sensitivity to slab's dimensions, incremental loading and induced void. The results indicated that using a bigger reinforced concrete slabs (1200 x 1200 mm2) could potentially have further reduced the discrepancy between theoretical (analytical and numerical) and experimental natural frequencies than smaller slabs (600 x 600 mm2). In general, for the specimens tested slabs, natural frequencies were more sensitive to the damage introduced than the damping ratio because the damping did not consistently increase or decrease as damage increased. The changes in mode shapes tended to increase with increasing damage level. Even small damage induced poised changes to the mode shapes, but it may not be obvious visually. Utilising sophisticated methods for damage identification, which are vital steps in higher level of damage detection in structures, is one of the major contributions to the knowledge. The proposed Modal Assurance Criterion (MAC) and Coordinate Modal Assurance Criterion (COMAC) techniques as advanced statistical classification model were employed in this study. From the vibration mode shapes induced void location can be identified via MAC and COMAC techniques when both intact and damaged data were compared. MAC provided a clear change in the mode shape while the COMAC provided the change in specific a location whereby the location of damage was identified. The outcomes of this two techniques can show the realistic location of the void. Beside the aforementioned contributions in this research, the feasibility of a Feed-Forward Back Propagation Neural Network (FFBPNN) was investigated using ten natural frequencies as input and the void sizes as output. Excellent results were obtained for damage identification of four void sizes, showing that the proposed method was successfully developed for damage detection of slabs. The results proved that the precision of the models was reduced when dealing with small size void. The large size void was detected more accurately than small size void as expected. This is because the natural frequencies of the small void of different location attributed together. Therefore, natural frequencies alone were not considerably good enough to make good identifications for small size void. Moreover, the natural frequencies set of three untrained void specifications were used as FFBPNN inputs to test the performance of the neural networks. The obtained results show that the proposed network can predict the void specifications of the unseen data with high accuracy. Overall, the methodology followed in this work for damage detection in reinforced concrete square slabs is novel when compared to the breadth and depth of all other previous works carried out in the field of reinforced concrete structures.

Item Type:	Thesis
Creators:	Ahmed, M.S.
Date:	March 2016
Divisions:	Schools > School of Architecture, Design and the Built Environment
Record created by:	Linda Sullivan
Date Added:	29 Jun 2016 09:14
Last Modified:	29 Jun 2016 09:14
URI:	https://irep.ntu.ac.uk/id/eprint/28035