A Hybrid Approach of Partial Least Squared Analysis and Artificial Neural Networks for Predictive Control of a Ceramic Process

Abstract

Artificial neural networks (ANNs) have been widely used in modelling and control of many practical industrial nonlinear processes. However, most of them have focused on a general approach for input selection. In this paper, a hierarchical hybrid approach of partial least squared (PLS) analysis and ANNs has been applied for predictive control of a real drying process in ceramic tile manufacturing. This approach is employed in order to promote the reliability of neural network model via reduction of the input set dimension. First, PLS analysis is done to arrive at the significant factors that influence the spray drying quality the most. Also, the significant factors are used to construct the predictive neural network model so called the Focus-NN model. Next, the reliability of the Focus-NN model is compared with the ANN model. In order to develop a predictive-control strategy using a more reliable model, i.e. the Focus-NN model, several scenarios as an accurate, fast running and inexpensive method are deployed to identify the optimal process settings considering the desired output.

Keywords:

• Ceramic industry
• Spray drying
• Artificial neural networks
• Predictive control
• Partial least squared analysis