Producer gas composition prediction using artificial neural network algorithm / Mohd Mahadzir Mohammud ... [et al.]

Nowadays, methods to increase efficiency in producer gas have become major issues in biomass gasification research. Producer gas is a renewable energy source that does not take as much time to obtain as fossil fuels. It is typically a mixture of combustible gases like carbon monoxide, hydrogen and methane, and non-combustible gases like carbon dioxide and nitrogen. A high percentage volume of combustible composition in the producer gas output will have a high calorific value or heat of combustion. These combustible gases are determined by the design of the gasifier. In today's era of Industrial Revolution 4.0 and Society 5.0, the use of simulation is highly prioritised in all aspects of engineering, especially in gasification applications. Simulation is a useful tool for learning about the governing principles and optimal operating points of the gasification process. Artificial intelligence (AI), is a major focus of Industry Revolution 4.0. In this project, the producer gas composition prediction is studied by computer simulation. The goals are to predict the output producer gas using an algorithm and to compare the trained prediction result with actual experiment data for rice husk gasification. This simulation was created with MATLAB software's artificial neural network (ANN). Three parameters (the height of the gasifier, the diameter of the gasifier, and the weight of the rice husk) are set as input data, and six types of the composition of producer gas (carbon dioxide, carbon monoxide, methane, oxygen, hydrogen, and nitrogen) are set as output data. The algorithm is trained, tested, and verified with the experiment data. It is then used to predict the output gas composition from the parameters of a gasification experiment that has been used before in UiTM’s laboratory. The simulation results of producer gas composition between prediction and actual values revealed a relative error of 1.159 %, 0.370 %, and 0.330 %. These results were less than 9% and were found to give a very good fit to the neural network algorithm.