Estimation of pellet size and strength of limestone and manganese concentrate using soft computing techniques

This paper presents a soft computing approach to estimate wet or green pellet size and strength and signifies the importance of individual process parameter. A huge portion of available minerals and materials is in the form of fine powder that makes their management and utilization a tedious job. Pelletization, a size enlargement technique, is used to tackle aforementioned problems and provides benefits such as ease of handling and storage, convenient transportation and improved process efficiency. Besides other characteristics, pellet size and strength are of prime significance and their accurate estimation can enhance the product quality. Real-life constraints (e.g., time in conducting experiments, repetition of experiments considering measurement errors, availability of resources, etc.) pose difficulties in generating sufficient experimental data at the laboratory. Hence, the concept of random population generation of genetic algorithm is exploited to fulfill data requirement where fitness functions are formulated using multiple regression. Better visualisation of pellets attributes and operating conditions is done through self-organizing map which also helped in deducing some important facts. Sensitivity analysis and construction of learning curves is also performed in present study. Finally, multilayer perceptron, a popular supervised neural network model, is applied on synthetic data to approximate pellet properties. Agreement between estimated and corresponding experimental values ascertained that neural network based system can be employed as an effective and credible tool to anticipate characteristics of wet pellets.