grinding mill system identification and control system design in comminution using artificial neural networks

Loading...
Thumbnail Image
Date
2004-04
Authors
Rama Putra, Teuku Andika
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Grinding processes are generally complex systems exhibiting strong nonlinearities and time delays, and hence are very demanding in control requirements. For economical and final product quality reasons, the size reduction operating conditions must be controlled in order to obtain a stabilized particle size distribution profile. This research reports on the control of grinding mill process using artificial neural networks. There are typically two steps involved when using neural networks for control i.e. the development of neural networks grinding mill plant model(neural networks based system identification) and the design of neurocontroller (neural networks based control - system design). System identification with multilayer perceptron and backpropagation using Levenberg-Marquardt method has been chosen to build the neural networks plant model as well as the use of neural networks as a generic model structure for the identification of nonlinear dynamic systems. The neural networks plant model uses process variable inputs and previous plant output to predict future values of the plant output. This model uses data collected from the operation of the actual grinding plant and the data collected from the simulator. The neural network is trained with backpropagation using Levenberg-Marquardt method. Some of the nonlinear system identification functions were used to identify the neural networks model. It was found that the size distribution obtained from the simulation of neural networks grinding mill model using the new input data (control input) from neurocontroller simulation approaches the desired size distribution.
Description
Keywords
System identification , System design in comminution , Artificial neural networks
Citation