Neural modelling in adsorption column of cholesterol-removal efficiency from milk

- The removal of cholesterol from milk by batch experiments is proposed.
- Artificial neural networks are applied for prediction of cholesterol removal.
- The ANN model is developed using a three layer feedforward backpropagation network.
- The operational parameters were studied on cholesterol-removal efficiency from milk.
- The model is able to predict the behaviour of process under different conditions.

In this study, the process of cholesterol removal from milk in an adsorption column with a continuous flow was modelled with artificial neural networks (ANN) models. The input operational parameters used for training the neural network include the bed height (1-3 cm), contact time (0-6 h) and flow-rate (3-9 mL/min). The cholesterol-removal efficiency (%) was defined as the output of the neural network. The neural network structure has been optimised by testing various training algorithms, different number of neurons and activation functions in a hidden layer. A high correlation coefficient (R2 average ANN = 0.98), a minimum mean-squared error (MSE) and the minimum root mean squared error (RMSE) showed that the neural model obtained was able to predict the cholesterol-removal efficiency in milk. Comparison of the model results and experimental data showed that the ANN model can estimate the behaviour of the cholesterol-removal process through adsorption under different conditions.