Enhancing Concrete Performance with Waste Rubber: An Artificial Neural Network Approach for Mix Ratio Optimization and Predictive Analysis
Due to its enhanced mechanical qualities and environmental sustainability, waste rubber in concrete materials has attracted more attention. In this investigation, a numerical model was suggested to clarify the effect of rubber tiering on strengths. Given the variety of raw materials available then, changing how concrete works to fit modern development is hard. The standard mixing ratio test method could be replaced with the artificial neural optimization model, and the concrete performance prediction could be realized accurately and quickly. This will allow for the more efficient use of a wide range of complex novel raw materials and improve the concrete mix ratio design method. Evaluating the workability, compressive, splitting, tensile, and bending strengths include replacing rubber waste particles with 10%, 20%, and 30% of aggregates by volume. Data was collected from the literature, and the results were analyzed using an Extreme Learning Machine (ELM) and Multiple Linear Regression (MLR) to accurately evaluate rubberized concrete beams' performance with micro-reinforcement. The results revealed that using rubber waste particles leads to a decline in the weight of the concrete sample and a level of Compressive Strength (CS) compatible with a load-bearing wall. The regression analysis revealed a high correlation between the independent and dependent variables, with R2 values ranging from 0.932 to 0.983 in both the training and testing phases. However, there was higher variability in the test phase, with RSD values ranging from 6.23% to 9.92%, compared to the training phase, with RSD values ranging from 5.12% to 6.77%. The study demonstrated the potential to use waste rubber in concrete composites and the importance of considering the training and test phase results for accurate predictions. Regarding accuracy, both models have relatively high R2 values (0.954 for ELM and 0.943 for MLR), indicating a strong correlation between the independent and dependent variables. However, ELM has a slightly lower RSD (6.88%) than MLR (5.45%). Regarding Mean Relative Error (MRE), both models have similar results (5.12% for ELM and 5.87% for MLR). In terms of time, the ELM model has a much faster running time of 2.45 seconds compared to the MLR model, which takes 28.011 seconds. ELM can make predictions faster than MLR, which could be significant in real-time applications.