1. Ph.D Theses

Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/1/11

Browse

Now showing 1 - 3 of 3

ANN Modeling and Optimization of Power output from Horizontal Axis Wind Turbine
(National Institute of Technology Karnataka, Surathkal, 2019) Rashmi; A, Sathyabhama.; P, Srinivasa Pai.
Integration of wind energy with energy with existing power sources has been restricted due to its intermittent and stochastic nature. Hence, there is a great need to develop an accurate and reliable site-specific prediction model. Forecasting of wind speed which is an important parameter affecting turbine power output, will help the wind energy industry in proper planning, scheduling and controlling. Artificial Neural Network (ANN) has proved its capability in mapping such complex non-linear inputoutput relations. The main objective of the wind energy industry, is to reduce the cost and increase the power generation by optimizing the controllable parameters affecting the turbine power output. The metaheuristic optimization algorithms, which are robust to dynamic changes are proved to be successful in solving such complex real-world problems. This research work has been carried out in three different phases namely wind power prediction, wind power optimization and wind speed forecasting (WSF).The data for this research work has been collected from the Supervisory Control and Data Acquisition System (SCADA) of 1.5 MW, pitch regulated, three bladed, horizontal axis wind turbine, located in a large wind farm present in central dry zone of Karnataka, India. In the present study, different conventional and ANN models have been used to predict the power output of a turbine. ANN models have been developed based on batch learning and Online Sequential Extreme Learning Machine (OSELM) algorithms, by considering carefully selected variables affecting power output, namely wind speed, wind direction, blade pitch angle, density and rotor speed. Maximizing the power output of the wind turbine by optimizing the only controllable parameter namely blade pitch angle has been achieved using three different metaheuristic optimization algorithms. A vihybrid ANN multistep WSF model, which is a combination of OSELM, Cuckoo Search (CS) and Optimized Variational Mode Decomposition (OVMD) method, hence named OVMD-CS-OSELM has been proposed in the present study. The performance of this hybrid model has been then compared with the benchmark models. From this study it has been found that, the models based on Extreme Learning Machine (ELM) converge extremely faster with better generalization performance and generate a compact network structure compared to Backpropagation learning. Out of the fifteen models based on batch learning, the fully optimized RBF model with ELM learning resulted in good performance with Root Mean Square Error (RMSE) value of 1.73%. The detailed study of OSELM algorithm showed a RMSE value of 1.96%, which is slightly higher than the fully optimized RBF model. However,for the present application due to the online nature of the wind data, OSELM algorithm is highly preferable. CS optimization algorithm is found to be suitable in optimizing the blade pitch angle of the turbine and accordingly the optimization of the power output, due to its fast convergence and a highest Mean relative PG value of 17.329%. In comparison with benchmark models, the proposed WSF model showed clear benefits of OSELM over ELM, OVMD over Emperical Mode Decomposition and CS over Partial Autocorrelation function for modeling, data pre-processing and input feature selection, with percentage improvements in Mean Absolute Percentage Error (MAPE) of 3.35%, 48.19% and 12.05% respectively for 1-step ahead forecasting. The proposed model has been validated using a standard database, which is from a meteorological station located in Portugal, thereby establishing its use in WSF. This research work thus proposes efficient models based on ANN for wind power prediction, optimization and WSF, which is useful in proper planning, integration and scheduling in the wind energy industry, thereby making it more competitive and a promising renewable energy source.
Effect of Boundary Layer Trip and Tubercles on Aerodynamic Performance of E216 Airfoil
(National Institute of Technology Karnataka, Surathkal, 2019) Sreejith, B. K.; A, Sathyabhama.
The thesis presents experimental and numerical investigation on the aerodynamic performance of airfoil E216 in prestall region at Reynolds number (Re) of 1 × 105. The effect of boundary layer trip (BLT) of different shapes and leading edge tubercles are studied with an aim to improve the aerodynamic performance and to eliminate the laminar separation bubble (LSB) formation on the airfoil. Finally the effect of modification are addressed in terms of performance improvement in the power output of a conceptual turbine which is capable of generating 100W at wind speed of 6.5 m/s. In the study Rectangular (RT), Right angled triangular (RA) and Isosceles triangular (IT) shaped boundary layer trips of different height located at two different positions are investigated. Wind tunnel experiment is conducted with rectangular boundary layer trip and is used for validation of numerical methodology. The amplitude of the tubercles are varied from 2 mm to 8 mm and wavelength from 15.5 mm to 62 mm. The different combination of the amplitudes and wavelengths resulted in total of nine models. Out of these, wind tunnel experiment are conducted on three models and are used to validate the numerical results. Langtry-Menter 4-equation Transitional SST Model or γ − Reθ - SST model is used in the study to model laminar separation bubble and effect of trip and tubercles on it. The investigation revealed that the the airfoil stalls at 120 with lift coefficient (Cl) of 1.37 and drag coefficient (Cd) of 0.063. Maximum value of lift to drag ratio of 42.46 is obtained at AOA of 40. Surface pressure distribution over the airfoil shows the presence of laminar separation bubble. The laminar separation bubble (LSB) formed at a vdistance of x/c = 0.22 from leading edge at AOA of 60. The location of laminar separation bubble moved upstream with increase in AOA. Based on the location of laminar separation bubble at AOA of 60, boundary layer trips are positioned at 0.17c and 0.10c from leading edge of the airfoil. Result showed that boundary layer trip eliminates LSB partially or completely and improves the aerodynamic performance of the airfoil. Highest improvements of 16.7% in Cd and 34.6% in Cl=Cd are obtained for location-2 with the rectangular trip having lowest trip height of 0.3 mm at AOA of 80. In all the cases, improvement in performance is observed only up to trip height of 0.5 mm. There is no observable advantage for isosceles and right-angled triangular trips over rectangular trips. Improvement in Cl is observed for most of the tubercled models and is significant at high angles of attack. But simultaneous increase in drag coefficient resulted in little improvement in Cl=Cd for most of the cases. But tubercled model with amplitude 2 mm and wavelength of 62 mm (A2W62) produced a peak value of 46.91 at AOA 60 which is higher than the baseline by 7.37%. Compared to baseline, there is high suction peak pressure along the trough and lower along the peak. The low amplitude and low wavelength tubercled model exhibited smooth Cp distribution without any sign of strong LSB. The LSB moves upstream with increasing amplitude and wavelength. LSB along the trough is formed ahead of that at peak inducing three-dimensional wavy shaped LSB unlike the straight LSB for the baseline. The tubercles considerably reduced the size of LSB compared to baseline. Two pairs of counter rotating vortices are formed on the airfoil surface between the adjacent peaks at two different chord-wise locations which strongly alter the flow pattern over it. The effect of trips and tubercles is demonstrated using a wind turbine performance analysis using BEM theory and it is seen that average improvement in power coefficient by 1.65% is obtained with the boundary layer trip and by 0.64% is achieved with the tubercles.
Investigation of Pool Boiling Heat Transfer from Rough Surface and Microchannel Geometry Under Variable Heat Supply
(National Institute of Technology Karnataka, Surathkal, 2019) Ashok, Walunj Avdhoot.; A, Sathyabhama.
Enormous amount of heat is generated in the Economic Simplified Boiling Water Reactor (ESBWR) due to exponential heat generation from the fuel rod. A core melt accident occurs when the heat generated in the nuclear reactor exceeds the heat removed by the coolant to the point where at least one nuclear fuel element exceeds its melting point temperature. Critical Heat Flux (CHF) is the phase of boiling after which heat transfer coefficient drops resulting in the rapid increase in temperature of core. Hence, understanding the mechanism of CHF is important to control loss of coolant accident (LOCA). CHF enhancement may retard the LOCA in ESBWR. Passive enhancement techniques are the most suitable for the nuclear reactor application. In view of the facts discussed above, the CHF enhancement by two passive techniques namely, rough surfaces and microchannel geometries is investigated. The transient CHF enhancement is compared with the steady-state CHF upto 10 bar pressure. The experimental setup is designed to study the pool boiling of water at 1 bar, 5 bar and 10 bar pressures. The pool boiling experiments are conducted on the thick copper sample of 20 mm diameter at saturated condition of distilled water. The unidirectional scratches are made on the sample to obtain wide range of surface roughness varying from Ra=0.106 μm to Ra=4.03 μm. It is found that steady-state CHF increases with increase in the Ra. Improved wettability and increased nucleation site density resulted in the CHF enhancement by rough surface. The microchannel geometries namely, square (SM-1.0), parabolic (PM-1.6) and stepped (SM-1.6) were fabricated by VMC machining. The improved liquid supply through the channel space and significant bubble growth resulted in the CHF enhancement by the microchannel geometry. The CHF enhancement by SM-1.6 is highest among all the microchannel geometries. The experimental setup is commissioned with programmable power supply to compare the CHF of water during pool boiling on rough surface and microchannel geometry under steady-state and exponential heat supply. The time constant (γ) of exponential heat supply is varied from 1 to 6. It is found that both, rough surfaces and microchannel geometries enhance the transient CHF. However, transient CHF gradually decreasedwith increase in γ due to liquid-vapor instability during exponential heat supply. CHF increased with increase in the pressure at both the condition viz. steady and transient. Steady-state CHF for Ra=4.03 μm and SM-1.6 at P=10 bar is found to be 71.43% and 47.37% higher compared to the CHF at P=1 bar, respectively. The correlation for heat transfer coefficient is developed for prediction of transient boiling performance which includes the non-dimensional time constant γ. Present correlation predicts the experimental values of transient HTC with MAE of 14.91%. CHF model for rough surface, based on force balance approach, is developed incorporating the effect of time constant, bubble angle and roughness parameter viz. Ra, Sm to predict the boiling crisis during pool boiling. It predicts the experimental transient CHF with MAE of 11.89%. Boiling videos are recorded at 1000 fps using high speed camera during the experiments to study the bubble dynamics during pool boiling on rough surface and microchannel geometries upto 10 bar pressure. Bubble dynamics during pool boiling of saturated water is significantly affected by the surface characteristics i.e. surface roughness and microchannel. Prolonged nucleated boiling regime is noticed for rough surface at high pressure due to the capillary wicking in the unidirectional scratches which retards the horizontal coalescence. Forces acting vertically on the growing bubble are considered to predict the bubble departure diameter. The MAE between measured and predicted bubble departure diameter for the rough surface and microchannel geometries at all pressure is 17.09% and 13.30%, respectively.