Hybrid Feed – Forward Back Propagation Neural Network Model: A Sensitivity Analysis of Corypha Utan Lam Fiber and Surkhi with Levenberg – Marquardt and Connection Weights Algorithm on a Fuzzy Inference System

Abstract

A current advancement in the concrete technology includes its ability to improve the mechanical properties significantly thru the utilization of fiber reinforced concrete. Due to the upsurge in the construction activities, the demand for the construction materials increase which can cause detrimental effects in the environment. Sustainability is one of the key issues that is needed to be addressed in the construction industry. Surkhi and Buntal Fiber was utilized as an alternative material in concrete production. Artificial Neural Network was utilized to obtain the best model for predicting compressive (f’c) and flexural strength (fb). Upon adopting Levenberg – Marquardt Algorithm and Hyperbolic tangent sigmoid as the training and transfer function, respectively, the final topology of the best model is 2-6-2 (Input Neuron-Hidden Neuron-Output Neuron). The weights produced from these model was utilized to determine the relative importance of Surkhi (S) and Buntal Fiber (BF) in compressive and flexural strength using Connection Weights (CW) algorithm. Based on the results of CW algorithm, the importance ranking for compressive strength is BF<S while for flexural strength is S<BF. Parametric analysis was also performed as part of the sensitivity analysis to observe the behavior of the f’c and fb upon employing varying amounts of surkhi and buntal fiber. The use of these artificial intelligence tools is in line with the transition to industry 4.0 and an essential tool for applying sustainability in the construction industry.