Data-driven fuzzy modelling of a rotary dryer

It was examined how different fuzzy modelling approaches, such as neuro-fuzzy, fuzzy clustering and linguistic equation methods, apply to the modelling of a rotary dryer. Because rotary drying, one of the oldest process in industry, is a highly nonlinear, strongly interactive multivariable process, its modelling is a demanding task. Its mathematical model, consisting of partial differential equations with several experimental parameters, is very complex and cumbersome. Therefore, the data-driven model is attractive, especially because many experimental observations and operating experience exist. The paper describes the fuzzy modelling approaches applied to the modelling of a rotary dryer. The applicability of different approaches has been evaluated by simulations, with the data collected from a pilot plant rotary dryer. The performance was estimated by an error index root means squared method and by comparing the modelling results with the results achieved by a linear regression model and a neural network model. The results show that neuro-fuzzy, fuzzy clustering and linguistic equation methods apply well, and no big differences can be detected between the methods.