Adaptive neuro-fuzzy inference systems with improved optimisation for fibre-reinforced polymer-confined concrete strength prediction

Fibre-reinforced polymer (FRP)-confined concrete is often applied in engineering practice where structures are damaged. However, research on the behaviour of FRP-confined damaged concrete remains relatively limited. In this study, a machine learning approach was adopted to develop a unified model for the ultimate strength of damaged concrete, and a practical formula was derived for rapid strength estimation in engineering applications. To enhance the optimisation capability, the arithmetic optimisation algorithm (AOA) was first improved (IAOA) by incorporating the optimal neighbourhood learning strategy and a crisscross search mechanism. Then the algorithm is hybridised with adaptive neuro-fuzzy inference systems (ANFIS) for predicting the strength of FRP-confined damaged and undamaged concrete. The performance of the ANFIS-IAOA was compared with that of four other traditional ANFIS algorithms, namely ANFIS-particle swarm optimisation, ANFIS-grey wolf optimisation, ANFIS-genetic algorithm and ANFIS-AOA. 736 existing experimental data of FRP-confined damaged and undamaged concrete were collected to train and test the algorithms. Metrics including convergence speed and error evaluation factors were used to assess the performance of the algorithms. The results showed that the ANFIS-IAOA was applicable to predict the FRP-confined concrete strength and outperformed the other four hybrid ANFIS algorithms. Meanwhile, based on the principles of ANFIS, the formulaic output for predicting the strength of FRP-confined damaged and undamaged concrete was derived.