Assessing the load carrying capacity of RC members through the use of artificial neural networks

The present study aims at developing a process for accurately assessing the load carrying capacity of reinforced concrete (RC) members through the use of appropriately trained artificial neural networks (ANNs). The development of the ANNs is based on a number of databases which include data obtained from a wide range of tests conducted on simple structural elements (e.g. beams, columns) describing certain aspects of the exhibited behaviour (i.e. load-bearing capacity, mode of failure) at the ultimate limit state (ULS) in relation to a number of design parameters. It is observed that the available test data often exhibit significant departures from their counterparts predicted by the relevant design codes. This raises concerns regarding the ability of the codes to provide design solutions capable of safeguarding the often stringent performance requirements usually associated with strength and ductility. The predictions of the ANNs are compared with their counterparts obtained from: (i) a wide range of tests, (ii) the current design codes (iii) nonlinear finite element analysis and (iv) the Compressive Force Path (CFP) method, which provides design solutions considerably different to those of the available design codes. ANNs are proven capable of directly assessing the available test data and objectively quantifying the effect of important design parameters on RC structural response. The comparative study reveals that for certain values of design parameters the predictions of the CFP method and the ANNs correlate closer to the relevant experimental data than their counterparts predicted by the RC design codes. This highlights an urgent need to re-assess the assumptions upon which the current RC design codes are based upon.