Sushmit Patra

Bio: Sushmit Patra is an academic researcher from KIIT University. The author has contributed to research in topics: Noise & Thermal conduction. The author has co-authored 2 publications.

Generation of temperature profile by cascade-forward type artificial neural network in Non-Newtonian fluid flow problem under noisy data

Abstract: Cascade-forward type artificial neural network (CFANN) is explored to generate temperature profile for a non-Newtonian third grade fluid flowing through two parallel plates. Uniform and constant heat fluxes are supplied to both the plates. A semi analytical approach (Least Square Method LSM) is used to solve the governing equations under required boundary conditions. The velocity and the temperature profile obtained from the LSM, are perturbed by different levels of noise to mimic error in measurement. Thus, the perturbed velocity and temperature profiles are fed into CFANN for training. In CFANN, Scaled Conjugate Gradient (SCG) algorithm is used for training the neurons. Once training of CFANN is completed, a velocity profile (not part of the training data) is fed as input, and the temperature profile is obtained as output. The temperature profile obtained from CFANN found to be in very good agreement with the LSM results. This approach is suitable to solve the present types problem with small alterations, and removing the need to solve such problems by LSM or any other time consuming methods. This leads to time savings, and is useful for industries involved in non-Newtonian fluid like polymer, paints, blood, grease etc.

Scaled Conjugate Gradient Algorithm Trained ANN for Parameter Retrieval in Combined Heat Transfer in Porous Ceramic Matrix

Abstract: AbstractA combined mode conduction radiation problem is considered in 2 D rectangular porous ceramic matrix (PCM). The governing equations of the problem are solved by finite volume method (FVM) to compute temperature profiles for solid and gas phases. This solution is then used to train artificial neural network (ANN). Very popular scaled conjugate gradient algorithm is employed to train the neurons in ANN. The trained ANN model is analyzed for its robustness with the help of performance curves, histogram and regression analysis. The trained ANN model is fed with an unknown gas and solid temperature profile, and the ANN model is able to give the corresponding heat transfer coefficient (HTC), with good accuracy of 5.4%. The refression coefficient of 0.998 is obtained for the ANN model.KeywordsPorous ceramic matrixConductive-radiative transferParameter retrievalInverse analysisArtificial neural networkScaled conjugate gradient algorithm

Retrieval of Parameter in Combined Mode Conduction–Radiation Problem in Porous Ceramic Matrix by Artificial Neural Network

Abstract: AbstractArtificial neural network (ANN) is used to retrieve one parameter in conduction–radiation heat transfer problem in porous ceramic matrix. Air flows through a 2D rectangular porous ceramic matrix (PCM) with uniform velocity. The PCM is assumed to be conducting and radiating, also a localized heat generation zone is situated at center. All the governing equations together with appropriate boundary conditions are solved by using finite volume method (FVM), to compute the temperature profiles of the gas and the solid phase. Both the temperature profiles are generated for different values of heat transfer coefficient (HTC). The ANN is trained by using the solid and gas temperature profile, along with the corresponding HTC. Neurons in the ANN are trained by using Levenberg–Marquardt (LM). Once the ANN model is trained, it is analyzed and explored to determine one parameter in the problem. The trained ANN model is fed with an unknown solid and gas temperature profiles as input, the ANN gives back the corresponding HTC as output. The retrieval of HTC by LM algorithm is found to be very accurate.KeywordsPorous ceramic matrixConductive–radiative transferParameter retrievalInverse analysisArtificial neural networkLevenberg–Marquardt algorithm

Hybrid genetic algorithm-self organizing map network for decision support system: An application in combined mode conduction-radiation heat transfer in porous medium

Abstract: Hybrid Genetic Algorithm (GA)-Self organizing map (SOM) network is explored for the first time in determination of regime of operation of heat transfer in porous medium. Combined mode conduction-radiation heat transfer in porous medium is dependent on various key features of porous medium like scattering albedo ω, down-stream porosity ϕ2, convective coupling P2, and extinction coefficient β. Depending on the values of these key features, heat transfer in porous medium has 16 different regime of operation. Correct determination of the regime of operation of porous medium, helps in utilization of porous medium to its maximum potential. Different architecture of SOM is employed by GA, and the best is given as output. The best SOM is then used for the determination of regime of operation of porous medium. Two commonly used topologies in SOM: hexagonal and rectangular grids, are explored for the determination of the regime of operation of porous medium. Present approach, aims to simplify the design optimization of porous medium based devices like burners etc.