Abstract A numerical study based on steady RANS with the SST turbulence model was performed to investigate the flow and heat transfer in a ribbed U-duct with a rectangular cross section that converges linearly in the radially outward direction and diverges linearly in the radially inward direction under rotating and non-rotating conditions. Parameters examined include rotation number (Roi = 0, 0.0219, 0.0336, 0.0731), Reynolds number (Rei = 46,000, 100,000, and 154,000), and the duct's taper angle (α = 0° and 1.41°) under conditions relevant to gas turbines used for electric power generation. Results obtained show that increasing the taper angle from 0° to 1.41°, which appears negligibly small, significantly increases both the friction coefficient and the Nusselt number whether there is rotation or not. With rotation at Roi = 0.0336 and Rei = 100,000, the maximum increase in the average friction coefficient and Nusselt number was found to be 41.7% and 36.6% respectively. Without rotation at Rei = 46,000, those are 11.5% and 14.7% respectively.

Blades in gas turbines have cross sections at the root that are larger than those at the tip so that internal cooling passages in blades are tapered. In this study, a reduced-order model (ROM) based on the integral continuity, momentum, and energy equations for a thermally and calorically perfect gas was developed to enable rapid assessments of radially outward flow in a tapered duct subjected to constant heat flux with and without rotation. The following parameters were investigated by using the ROM developed: taper angle (alpha = 0 degrees, 1.5 degrees, 3.0 degrees), ratio of mean radius to hydraulic diameter (R-m/D-h = 45, 150), rotation number (Ro = 0, 0.025, 0.25), Reynolds number (Re = 37,000, 154,000), and thermal loading (q" = 5x10(4), 10(5) W/m(2)). Results obtained show the density and pressure variation along the duct to be most affected by rotation number; velocity to be more affected by duct's taper angle; and temperature by rotation number and taper angle. Results obtained also show the temperature along the duct could decrease if the taper is sufficiently large even with high heat flux into the duct because taper increases velocity, which converts thermal energy to mechanical energy. Within the range of parameters studied, the mass flow of the cooling flow could be reduced by as much as 44% for a tapered duct to achieve bulk temperature variation similar to that of a duct without a taper. The ROM developed was validated by comparing its predictions with grid-converged CFD results obtained by steady RANS with the SST turbulence model. The maximum relative errors for density, velocity, temperature, and pressure distributions along the duct were found to be 0.6%, 3.3%, 0.4%, 0.3% for the smooth section of the duct and 3.2%, 5.6%, 0.9%, 3.0% for the ribbed section of the duct. Thus, the ROM developed performs nearly as well as CFD based on RANS but orders of magnitude more efficient computationally.

Abstract This work presents a novel study for identifying alterations in the control states of a desuperheater system based on real closed-loop data from a coal-fired power plant operating under various loads using linear and nonlinear system identification techniques. Specifically, Transfer Functions (TFs) and Gaussian Processes within a Nonlinear AutoRegressive eXogenous model (GP-NARX) are utilized. The desuperheater system comprises two units, north and south, each modeled as a single-input single-output (SISO) system based on spray valve positions and outlet temperatures. To identify changes in the control states using TFs, deviations in the coefficients of three poles and two zeros transfer functions are analyzed. Significant shifts in the control states of the north desuperheater are observed when transitioning from nominal to half and low loads, with deviations of up to four orders of magnitude. Substantial changes in control states are also observed for the south desuperheater when moving from nominal to low load, with a deviation in the coefficients of up to five orders of magnitude, whereas the transition from nominal to half load shows a smaller deviation of up to three orders of magnitude. In the GP-NARX approach, model uncertainties are used to indicate the changes in the control states. The south desuperheater showed a significant uncertainty of up to 8°F from the nominal to the low load, evidencing a change in the control states. Regarding the north desuperheater, increased uncertainty, up to 6°F, is also observed but in shorter time intervals when compared to the south desuperheater. Ultimately, this work shows that both approaches can be used as a basis for system identification, employing real closed-loop power plant data.

Large-eddy simulations (LES) were performed to study the turbulent flow in a channel of height H with a staggered array of pin fins with diameter D = H/2 as a function of heating loads that are relevant to the cooling of turbine blades and vanes. The following three heating loads were investigated—wall-to-coolant temperatures of Tw/Tc = 1.01, 2.0, and 4.0—where the Reynolds number at the channel inlet was 10,000 and the back pressure at the channel outlet was 1 bar. For the LES, two different subgrid-scale models—the dynamic kinetic energy model (DKEM) and the wall-adapting local eddy-viscosity model (WALE)—were examined and compared. This study was validated by comparing with data from direct numerical simulation and experimental measurements. The results obtained show high heating loads to create wall jets next to all heated surfaces that significantly alter the structure of the turbulent flow. Results generated on effects of heat loads on the mean and fluctuating components of velocity and temperature, turbulent kinetic energy, the anisotropy of the Reynolds stresses, and velocity-temperature correlations can be used to improve existing RANS models.

Turbine inlet temperatures in advanced gas turbines could be as high as 2000 °C. To prevent ingress of this hot gas into the wheelspace between the stator and rotor disks, whose metals can only handle temperatures up to 850 °C, rim seals and sealing flows are used. This study examines the abilities of large eddy simulation (LES) based on the WALE subgrid model and Reynolds-averaged Navier–Stokes (RANS) based on the SST model in predicting ingress in a rotor–stator configuration with vanes but no blades, a configuration with experimental data for validation. Results were obtained for an operating condition, where the ratio of the external Reynolds number to the rotational Reynolds number is 0.538. At this operating condition, both LES and RANS were found to correctly predict the coefficient of pressure, Cp, located downstream of the vanes and upstream of the seal, but only LES was able to correctly predict the sealing effectiveness. This shows Cp by itself is inadequate in quantifying externally induced ingress. RANS was unable to predict the sealing effectiveness because it significantly under predicted the pressure drop in the hot gas path along the axial direction, especially about the seal region. This affected the pressure difference across the seal in the radial direction, which ultimately drives ingress.

View Video Presentation: https://doi.org/10.2514/6.2023-2683.vid The efficient global optimization (EGO) algorithm (also called Bayesian optimization) has been successfully applied to simulation-based engineering design optimization problems which are severely limited by time. The EGO algorithm is nearly always used in combination with Gaussian process regression (GPR) (also called Kriging) since it provides a prediction model as well as an uncertainty model. The computational cost of GPR, however, grows quickly with the number of samples, which can limit it to design problems with a few number of variables. EGO is not restricted to GPR and the only requirement is a surrogate with an uncertainty model. Deep neural networks (DNNs) have been shown to be capable of handling a large number of samples. In this paper, a new EGO algorithm using DNN-based prediction and uncertainty (EGONN) is proposed. Given an initial set of samples, EGONN iteratively constructs two DNNs, one for the function prediction and the other for the prediction uncertainty, and utilizes them to find the next sample within the design space with the expected improvement infill criterion and, subsequently, update the DNN function prediction model. In this work, the DNN-based prediction uncertainty uses a separate, static training data set. The proposed EGONN algorithm is demonstrated on two analytical test problems, a bounded one-dimensional minimization problem and a constrained two-dimensional minimization problem, and compared with the EGO algorithm. The results show that EGONN yields comparable results as EGO with same number of initial and infill sample budget.

AbstractEngineering design research has largely focused on normative models of decision analysis based on small world causal frames where uncertainty can be resolved as probabilities or probability distributions. However, today we need to design solutions for our built environment that are sustainable, just, and able to adapt. Because of the scale and complexity of our world, designs that address sustainability, justice, and adaptability are dominated by unresolvable uncertainty. This requires large world frames and new engineering design frameworks and tools that provide a much broader and nuanced understanding of the impact of our engineering decisions. In this paper we propose that these tools will need to link quantitative and qualitative data and engineering judgment using narrative decision-making processes. To support this, we provide two examples where engineering decision-making is based in part on narrative processes. We then identify five research areas that require additional research to support large-world frames including (1) how can we create microcosms that enable transition between large- and small-world frames and (2) how engineers develop conviction to act using the narratives they create.

In gas turbines, the hot gas exiting the combustor can have temperatures as high as 2000 °C, and some of this hot gas enter into the space between the stator and rotor disks (wheelspace). Since the entering hot gas could damage the disks, its ingestion must be minimized. This is carried out by rim seals and by introducing a cooler flow from the compressor (sealing flow) into the wheelspace. Ingress and egress into rim seals are driven by the stator vanes, the rotor and its rotation, and the rotor blades. This study focuses on the ingress and egress driven by the rotor and its rotation. This is carried out by performing wall-resolved large eddy simulation (LES) around an axial seal in a rotor–stator configuration without vanes and blades. Results obtained show the mechanisms by which the rotor and its rotation induce ingress, egress, and flow trajectories. Kelvin–Helmholtz instability was found to create a wavy shear layer and displacement thickness that produces alternating regions of high and low pressures around the rotor side of the seal. Vortex shedding on the backward-facing side of the seal and its impingement on the rotor side of the seal also produces alternating regions of high and low pressures. The locations of the alternating regions of high and low pressures were found to be statistically stationary and to cause ingress to start on the rotor side of the seal. Vortex shedding and recirculating flow in the seal clearance also cause ingress by entrainment. With the effects of the rotor and its rotation on ingress and egress isolated, this study enables the effects of stator vanes and rotor blades to be assessed.

This work proposes a novel adaptive global surrogate modeling algorithm which uses two neural networks, one for prediction and the other for the model uncertainty. Specifically, the algorithm proceeds in cycles and adaptively enhances the neural network-based surrogate model by selecting the next sampling points guided by an auxiliary neural network approximation of the spatial error. The proposed algorithm is tested numerically on the one-dimensional Forrester function and the two-dimensional Branin function. The results demonstrate that global surrogate modeling using neural network-based function prediction can be guided efficiently and adaptively using a neural network approximation of the model uncertainty.

This paper presents data from a preliminary study designed to test the hypothesis that narrative is the primary decisionmaking tool employed in engineering design. We also look for evidence that analysis and optimization provide anchor points that drive the narrative forward, even in the presence of unresolvable uncertainties. The data in this paper comes from student engineers who are engaged in an 8-month, NASA-sponsored design competition. We code 46 statements from our interviews, spanning 10 different narrative elements as defined by Fenton-O’Creevy and Tuckett [1]. We find that engineers use multiple narrative elements to develop conviction when reaching action readiness, and that of the 46 coded statements, only 14 corresponded to quantitative analysis in the form of models and calculations. We find that in engineering design the decision-making journey involves a combination of quantitative analysis and qualitative engineering judgment that is evaluated to develop conviction and reach a state of action.

Gas turbine systems are widely used in the power industry because they provide continuous and reliable power to the electrical grid. One of the main concerns for implementing gas turbine systems is the maintenance costs. Therefore, predictive maintenance methods driven by Deep Learning (DL) models present an opportunity to extract important information and knowledge from the process data to minimize maintenance costs and reduce equipment failure rates. A previous study aimed to benchmark various state-of-the-art DL models for predicting compressor air leak with multivariate time-series data from a modified recuperated gas turbine system. However, the brute-force approach used to select the hyper-parameters of the DL models could be improved. This paper aims to address the hyper-parameter optimization process of the best performers for predicting the next future time-step: GRU-LSTM, Sequential CNNLSTM, and BI-LSTM. In addition, a BI-GRU model was implemented and a common grid search algorithm was combined with proposed algorithms for automating the selection of hyper-parameter values to build the DL models. The datasets were provided from experiments conducted at the U.S. Department of Energy's National Energy Technology Laboratory (NETL) Hybrid Performance (Hyper) Facility. Results suggest better performance can be obtained from the already good performing benchmarked models; however, reproducing the best results for some models may take more training cycles. The BI-GRU model exhibited the most reproducible results across all tests.