Over the past decades, engineers have focused on a common goal - to reduce emissions by making industrial processes more efficient and by utilizing renewable energy sources. One possibility to reach this goal can be achieved by employing processes with non-ideal fluids, such as fluids at supercritical conditions in refrigeration, heat pump cycles and power cycles.
Most of the flows in industry are turbulent and hence the need to study turbulence in non-ideal fluids arose. Turbulence in non-ideal fluids is extremely challenging since there are many complex effects at play. One such effect is caused by variations in properties, which also occurs in compressible flows, flows with high concentration gradients, or flows in heat exchangers.

This thesis presents a review of the existing theory on semi-local scaling for variable property flows with the aim to take it one step further and apply it to turbulent heat flux modeling. Two types of variable property cases are analysed in this thesis; (1) low-Mach number flows with uniform pseudo-heating sources, (2) high-Mach number flows with non-uniform viscous heating. For the former, a Direct Numerical Simulation (DNS) data-base, already available at TU Delft, has been post-processed with the main goal to investigate if the semi-local theory can also be applied to thermal turbulence and its modeling. For the latter, additional DNS simulations of high-Mach number channel flows have been performed to investigate how the viscous heating and its correlations can be accounted for in the semi-local scaling framework.

Using a 2-equation heat flux model, we find that modeling thermal turbulence in semi-local scales considerably improves the results for low-Mach number flows. However, for high-Mach number flows, additional unknown (closure) terms arise due to fluctuations in the viscous heating source. A model for the source term in the enthalpy variance equation is successfully proposed. In addition, the DNS study of two high-Mach number flows with constant semi-local Reynolds number profiles shed light on the importance of a newly defined parameter (modified Eckert number) and also unveils one of the most important conditions in which semi-local theory can be compromised, e.g. extreme density gradients.

The global warming and increase in energy use demands the industry to come up with ideas to lower the CO₂ production and save valuable energy. Organic Rankine cycles have a great potential as low grade heat can be converted into electrical energy. In the last ten years, research in this particular field has increased together with the amount of installed systems. Triogen is a company which does a lot of research on how to increase the power output of their Organic Rankine Cycle systems. Their product involves a radial inflow cantilever turbine connected to a single shaft to the generator and the main pump which makes the system quite unique. Because of the fluctuating heat input the system scarcely runs on its design condition so research has to be done to ensure the turbine efficiency is as high as possible to extract as much power out of the heat source as possible. This study investigates whether the pump strategy can be modified in order to increase the system power output for part load situation. By means of elaborate turbine computational fluid dynamics simulations, an optimal rotational speed op n = 475 hz is found for all turbine inlet pressures. A simple cycle model is made in Excel in order to compute the increased power output, given the optimal rotational speed. The most cost effective and straight forward way to implement the fixed rotational speed in the system, a throttle valve is chosen. The amount of extra power gained using the optimal rotational speed is approximately 7 kW, which is an increase of 8 %. For evaporator heat inputs higher than 900 kW_th, the operation with throttle valve is no longer beneficial and the operation on design condition is favoured.

The total electrical energy consumption by all the operational data centers (located all over the world) is enormous (approx. 1% of the global electricity demand 20000TWh). This electrical energy is required 24x7 to operate and cool all the IT equipment present in the data center. The electrical energy required to cool all the servers present in the white space can range from as low as 10% to as high as 40% of the total data center electrical consumption. The server's inlet temperature has to be within the ASHRAE recommended range (18o C – 27o C), so that they can function correctly.   A simplified design of a raised floor white space with hot aisle / cold aisle configuration is considered. The tile flow rate through the floor tiles influences the server's inlet temperature. To control the tile flow rate, 11 design variables of the data center white space are identified. These are the position of the 4 perforated plates, the amount of perforations of each perforated plates, the floor tiles perforation, the raised floor height, and the CRAH distance to the cabinet. 600 design samples of the white space are generated by applying the Latin Hypercube Sampling (LHS) technique on these 11 design variables. A well-validated CFD software 6SigmaRoom by Future Facilities is used to generate the CFD results. The standard k-ε model is used to model the turbulence in the CFD simulations, in a steady-state condition. A database of 600 samples is generated by recording the cabinet's inlet temperature, flow rate of floor tiles from the CFD simulations, and the corresponding changeable design parameters generated by the LHS technique. The error due to CFD simulation is estimated at less than 4% for the tile flow rate and 1.7o C for the server inlet temperature.  Four Artificial Neural Networks (ANN) are trained on the data from the database to predict the floor tile's tile flow rate and the cabinet's inlet temperature, respectively. The average R2 prediction (testing) accuracy is 0.97 for the tile flow rate predictions and 0.92 for the cabinet's inlet temperature predictions. Their average prediction error is less than 5% for the tile flow rate and less than 20 C for the cabinet's inlet temperature. The Non-dominated Sorting Genetic Algorithm-II (NSGA-II), a variant of the genetic algorithm, is used to find the optimum values of the 11 design parameters of the white space. These optimum values are going to ensure the server's inlet temperature to be within the ASHRAE recommended range. The genetic algorithm optimizes the design variables based on the predictions made by the neural network predicting the tile flow rate. The values of the optimized design parameters are verified using the 6SigmaRoom software by comparing the server's (mean) inlet temperature of the optimized case with the non-optimized case. More number of servers have their (mean) inlet temperature below 270 C in the optimized case as compared to the non-optimized case. The electrical power required by the CRAHs to cool the white space is reduced by 10% in the optimized case as compared to the power which is required by the CRAHs in the non-optimized case. In this study, a CFD simulation of the white space took 40 minutes. The neural networks took less than a minute to make the predictions, and the NSGA-II algorithm took less than 10 minutes to find the optimized design parameters of the white space.  Thus, in this thesis, it is shown that using an artificial neural network and a genetic algorithm, in combination with computational fluid dynamics gives satisfying results in optimizing the white space design, required to keep the server's inlet temperature within the ASHRAE recommended range. The computational time required to find the optimum white space design is also reduced by using a neural network and a genetic algorithm. The prediction by the neural network and the optimization performed by the genetic algorithm can be improved further with the availability of more training data and in-depth knowledge of applying these techniques (the neural network and the genetic algorithm) in predicting and optimizing the solutions respectively.  

High-fidelity optimisation studies are a useful asset in the design of critical components for large gas turbines. These studies require the computation of numerous computationally expensive CFD simulations and result in predominantly optimisation graphs of design objectives (e.g. Pareto-front figure). The quantity of generated data is intractable and therefore in practice not stored. All the physical insight into the flow-fields is lost and no data can be accessed nor any adjustments can be made. This work researches the combination of CFD simulations and Machine Learning (ML) algorithms to simultaneously: 1) reduce computational cost and time, 2) retain physical insight by focusing on the prediction of flow-fields and 3) keep the ability to access information or to make adjustments. This research starts with the identification of 19 potential holding ML-algorithms. A suitable algorithm is chosen with the help of 4 criteria, that is referred to as the Proper-Orthogonal Decomposition (POD) - Interpolation algorithm. In studied literature it reportedly provided very high speed-up times, good accuracy, additional learning benefit and is relative simple. The POD-Interpolation algorithm is implemented on two test cases: the Von Karman Vortex Street and a Shear Mixing Layer. The results are summarised for both as unphysical and highly inaccurate. Two root-causes are identified: the first is an invalid assumption in the methodology and the second dominant root-cause is a fundamental flaw. Interpolation can never accurately predict dominant frequencies in a flow-field. With the inability of the POD-Interpolation algorithm to predict physical accurate flow-fields, a new proposed algorithm is implemented. This is referred to as the Sparse-Dynamic-Mode-Decomposition (sDMD) - Scaling method. This method relies on accurate scaling relations to predict flow-fields. However, almost no application cases have known scaling relations. The sDMD-Scaling algorithm is capable of discovering and using these newly discovered scaling relations to predict flow-fields. This is proven for the Von Karman Vortex Street case with self-made scaling relations: the Mean Residual Error of the predicted flow-fields rarely exceeds 7%. An error reduction factor of ≈ 10 is achieved when compared to the POD-Interpolation approach. The design space from the test case is computed ≈ 5.6 times faster with the CFD and sDMD-Scaling combination than with the conventional CFD approach. Furthermore, any new design point can be computed ≈ 309 times faster with the trained sDMD-Scaling algorithm than with a CFD simulation. The sDMD-Scaling algorithm proves that it is possible to combine CFD and ML to achieve the 3 aforementioned goals for a canonical fluid dynamic test case. Several recommendations are presented to improve the sDMD-Scaling algorithm for more complex cases. However, given the difficulties and efforts that went into the relative straightforward test cases, it is difficult to foresee any large-scale implementation of any CFD and ML combination for predicting flow-fields.

This study examines Gas Path Analysis (GPA) as a method for the predictive maintenance of the Honeywell 331-500 Auxiliary Power Unit (APU). The 331-500 is used on the Boeing 777 and is susceptible to a sudden failure during operation due to the release of a first stage turbine blade. Investigations found the failure mechanisms of the blade release to be hot corrosion and thermal fatigue. Prolonged operation and the ingestion of environmental contaminants cause corrosion and a loss of material of the first stage Nozzle Guide Vane (NGV) which leads to a reduction in turbine performance. To counter this loss in performance, the APU controller increases the fuel flow, hence Turbine Inlet Temperature (TIT), which leads to under-platform crack growth of the first stage rotor and the eventual release of a blade.

A thermodynamic model of the 331-500 is created and validated using the Gas Turbine Simulation Program. Test cell data is used to characterise the APU at its design point before historical failure data comprising on-wing APU performance is analysed to assess the condition of the turbine. GPA, in the form of adaptive modelling, is used to adjust the turbine efficiency to match the model with the performance data thereby determining the deterioration of the machine.

Analysis highlights a relationship between the gradient of turbine efficiency, the ratio of the controller- to model-calculated TIT, and the condition of the first stage NGV. A larger gradient and TIT ratio indicate significant material loss in the NGV and that the turbine is heavily deteriorated. From the analysis, a rule functioning as a threshold on both the turbine efficiency gradient and the TIT ratio is formulated and is capable of predicting the historical first stage turbine blade failures up to 600-250 cycles before failure. This rule may then be used for the monitoring of current and future 331-500s in operation with the aim of preventing failure.

The main objective of the thesis is to analyze the potential for improving the aerodynamic efficiency of Siemens SGT5-2000E turbine section by optimizing the blade shapes. An adjoint-based shape optimization is implemented at the mid-span of the axial turbine. The optimization is performed, for the stator and the rotor individually, to reduce the entropy generation (objective function) with mass flow rate as the constraint.

The design optimization methodology is implemented using SU2, an open-source computational fluid dynamics (CFD) tool coupled with the adjoint-based optimization technique. The SU2 optimizer algorithm begins by computing the objective function of the existing design by using the flow solver. The flow simulation is performed by solving the RANS equations and SST turbulence model. The discrete adjoint solver utilizes the objective function and constraints to evaluate the gradients of the objective function with respect to the design variables. Each of the design variables is altered to improve the shape and the gradients are used to find an optimal search direction. The algorithm is structured to iterate until an optimal shape is determined.


The optimization methodology is implemented for the existing stator to reduce the entropy generation and an optimal shape is determined. Then, the rotor is optimized with the outlet conditions of the optimized stator as the inlet conditions. The optimized stator and rotor resulted in a significant decrease in entropy generation of about 16% and 24% respectively. Finally, with the optimized blades the stage simulation is performed which resulted in 1.4% increase in the total-to-total efficiency compared to the baseline stage.