Artificial Intelligent Systems and Machine Learning

Nirjhar Bar ¹, Tapan Kumar Ghosh ¹, Manindra Nath Biswas ², Sudip Kumar Das ¹

Affiliations
1 Department of Chemical Engineering, University of Calcutta, 92, A. P. C. Road, Kolkata-700009, IN
2 Government College of Engineering & Leather Technology, LB Block, Sector III, Salt Lake City, Kolkata-700098, IN

Abstract

Experimental investigations have been carried out to determine the flow regime for air-water two-phase flow through 3 mm and 4 mm transparent tubes by visual observation. The feasibility of Artificial Neural Network (ANN) based techniques for the classifications of flow regimes are presented. Total 218 experimental data points are used in the ANN prediction. Five different well known ANN models have been tried to predict the flow regime. The ANN model based on MLP with Levenberg-Marquardt algorithm gives slightly better predictability over the other networks used.

Keywords

Flow Regime, Multilayer Perceptron, Radial Basis Function, Support Vector Machine, Principal Component Analysis.

Full Text

     

Nirjhar Bar ¹, Asit Baran Biswas ¹, Manindra Nath Biswas ², Sudip Kumar Das ¹

Affiliations
1 Department of Chemical Engineering, University of Calcutta, 92, A. P. C. Road, Kolkata-700009, IN
2 Government College of Engineering & Leather Technology, LB Block, Sector III, Salt Lake City, Kolkata-700098, IN

Abstract

Helical coils are extensively used in different process industries. The two-phase flow through the coils is more complex than that of straight pipe due to presence of centrifugal forces, the flow through coils is always developing in nature. The knowledge of frictional pressure drop is an important hydrodynamic parameter used for the designing the coil. The applicability of the Artificial Neural Networks (ANN) methodology was investigated using experimental data obtained from our earlier studies on the frictional pressure drop for gas-non-Newtonian liquid flow through helical coils in horizontal orientation. Multilayer Perceptron (MLP) trained with back propagation algorithm using four different transfer functions in a hidden layer is used in ANN. Statistical analysis is used for the comparison to identify the best network. The ANN’s capability to predict the two-phase frictional pressure drop across the coils is one of the best estimation methods with high accuracy.

Keywords

Artificial Neural Network, Multilayer Perceptron, Frictional Pressure Drop, Gas-Non-Newtonian Liquid Flow.

Full Text

     

Nirjhar Bar ¹, Manindra Nath Biswas ², Sudip Kumar Das ¹

Affiliations
1 Department of Chemical Engineering, University of Calcutta, 92, A. P. C. Road, Kolkata-700009, IN
2 Government College of Engineering & Leather Technology, LB Block, Sector III, Salt Lake City, Kolkata-700098, IN

Abstract

Applicability of Artificial Neural Networks (ANN) methodology was investigated using experimental data obtained from our earlier experimental studies on the frictional pressure drop per unit length for gas-non-Newtonian liquid flow through 45° bend. This approach proved its worth when rigorous fluid mechanics treatment based on the solution of first principle equations is not tractable. The proposed approach towards the prediction is done using a Multilayer Perceptron (MLP), which is trained with backpropagation algorithm with the help of four different transfer functions in a hidden layer. Statistical analysis confirms that the transfer function 1 with 20 processing elements in the hidden layer gives the best prediction.

Keywords

Bend, Artificial Neural Network (ANN), Multilayer Perceptron (MLP), Backpropagation (BP).

Full Text