Determination of wall slip using rheological method and artificial intelligence techniques / Chin Ren Jie

Chin , Ren Jie (2019) Determination of wall slip using rheological method and artificial intelligence techniques / Chin Ren Jie. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (202Kb)
	PDF (Thesis PhD) Restricted to Repository staff only until 31 December 2021. Download (3516Kb)

Abstract

Suspension is very complex in nature. It exhibits several characteristics such as shear-induced migration, pattern formation and wall slip. Among the interesting characteristics, wall slip is the main focus of this study. Wall slip can be defined as a phenomenon in the flow of suspensions due to the movement of particles away from the wall boundary, leaving a thin liquid rich layer adjacent to the wall. It should be taken into consideration in material designing, manufacturing and transportation as it may cause inaccurate rheological measurements such as shear rate and viscosity. For example, the apparent (measured) shear rate is higher than that of the actual shear rate. Therefore, it is important to have a study on wall slip as the suspension has a wide application, especially for the industrial purpose. This study aims to investigate the relationship between the parameters of suspension (concentration, particle size and temperature) and wall slip. The rheological tests were conducted under different conditions, such as for particle sizes of 18 μm, 75.3 μm and 195.5 μm; volumetric concentration of 40%, 45%, 48%, 50%, 52% and 55%; temperature of 15°C, 25°C, 35°C, 45°C and 55°C. The result shows that the wall slip velocity increases with shear stress under the conditions when (i) concentration decreases, (ii) particle size increases and (iii) temperature increases. Current method for the actual shear rate prediction is a challenging task and quite time-consuming as several experimental works are required. It is also ineffective from the perspective of cost consumption if the material is costly. Therefore, the development of a mathematical model with an acceptable level of accuracy is required. Multi-Layer Perceptron Neural Network (MLP-NN), Adaptive Neuron-Fuzzy Inference System (ANFIS), Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) were employed to develop the prediction model by applying shear stress, volumetric concentration, particle size and temperature of suspension as input while the actual shear rate was kept as the output variable. For each approach, several models were developed and their performances were evaluated using statistical analyses. Among the examined models, the CNN model exhibits the best performance in terms of correlation coefficient (0.999), coefficient of determination (0.999), mean absolute error (0.0008), mean squared error (0.000002), root mean squared error (0.0015), Akaike Information Criterion (-18350), Bayesian Information Criterion (-18339) and percentage error (9.7%). So, the CNN model has the ability to predict the actual shear rate with a better accuracy. Overall, the novelty of this research study are determination of the relationship between wall slip and the influencing parameters (i.e. particle size, concentration and temperature) and the application of artificial intelligence techniques to mimic rheological actual shear rate under wall slip condition.

Actions (For repository staff only : Login required)