Optimization of Rubber Compound Design Process Using Artificial Neural Network and Genetic Algorithm

In the rubber industry, the process of designing rubber compound is of great importance due to the impact on product specifications. The good performance of this process is a competitive advantage for manufacturers in this industry. The process of designing a rubber compound includes a set of activities related to selecting the best amount of raw materials to prepare a composition with the desired physical and mechanical properties. Currently, the most common method for designing a rubber compound is the experimental method based on trial and errors. This method is time consuming and expensive. In addition, the obtained combination is not necessarily the best combination. To improve the performance of the rubber compound we need to design the desired process, this research presented using a combination of artificial neural network and genetic algorithm, with an approach to reduce time and cost, while increasing accuracy. In this method, the behavior of the rubber compound was modeled with artificial neural network. Then, using Genetic Algorithm as a quick search technique.The optimal values of the four raw materials such as carbon, sulfur, oil and accelerator; in order to determine the specified value of the two characteristics .abrasion and rubber modulus at 300% elasticity at the lowest price. To evaluate the method, several samples of rubber compound designed with two method. The results showed that the artificial neural network model has the ability to predict the two characteristics of abrasion and modulus based on the four mentioned raw materials in the trained range with high accuracy. In addition, average results for genetic algorithm, is a price of 17% less and a design accuracy of 84.5% more than experimental method. The design speed with this method is 454 times higher than the experimental design speed. Based on the results, by designing the rubber compound with the integration of artificial intelligence and genetic algorithms has a better performance than the experimental method.