Comparison of CNNs and ViTs Based Hybrid Models Using Gradient Profile Loss for Classification of Oil Spills in SAR Images

Oil spillage over a sea or ocean’s surface is a threat to marine and coastal ecosystems. Spaceborne synthetic aperture radar (SAR) data has been used efficiently for the detection of oil spills due to its operational capability in all-day all-weather conditions. The problem is often modeled as a semantic segmentation task. The images need to be segmented into multiple regions of interest such as sea surface, oil spill, look-alikes, ships and land. Training of a classifier for this task is particularly challenging since there is an inherent class imbalance. In this work, we train a convolutional neural network (CNN) with multiple feature extractors for pixel-wise classification; and introduce to use a new loss function, namely ‘gradient profile’ (GP) loss, which is in fact the constituent of the more generic Spatial Profile loss proposed for image translation problems. For the purpose of training, testing and performance evaluation, we use a publicly available dataset with selected oil spill events verified by the European Maritime Safety Agency (EMSA). The results obtained show that the proposed CNN trained with a combination of GP, Jaccard and focal loss functions can detect oil spills with an intersection over union (IoU) value of 63.95%. The IoU value for sea surface, look-alikes, ships and land class is 96.00%, 60.87%, 74.61% and 96.80%, respectively. The mean intersection over union (mIoU) value for all the classes is 78.45%, which accounts for a 13% improvement over the state of the art for this dataset. Moreover, we provide extensive ablation on different Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) based hybrid models to demonstrate the effectiveness of adding GP loss as an additional loss function for training. Results show that GP loss significantly improves the mIoU and F1 scores for CNNs as well as ViTs based hybrid models. GP loss turns out to be a promising loss function in the context of deep learning with SAR images.

Surface Roughness Impact on Boundary Layer Transition and Loss Mechanisms over a Flat-Plate under a Low-Pressure Turbine Pressure Gradient

An experimental study has been conducted to investigate the effects of surface roughness on the profile loss of a flat-plate with a contoured wall. All of the measurements have been conducted for the suction side pressure gradient of a high-lift low pressure turbine airfoil at the fixed Reynolds number (Rec) and freestream turbulence intensity (Tu) of 1.2 · 105 and 3.2%, respectively, representing a cruise condition. The time-resolved streamwise and wall-normal velocity fields for three different surface roughness values of Ra/C · 105 = 0.065, 4.417 and 7.428 have been measured with a 2D hot-wire probe. For the smooth surface, a laminar separation bubble forms from about 60% of the chord; and laminar-to-turbulent transition occurs during reattachment. Since the portion of turbulent flow over the flat-plate is relatively small, the overall profile loss is mainly determined by the momentum deficit generated during transition. Increased roughness decreases the maximum height and length of the separation bubble but does not affect the separation bubble onset location. The beneficial effects of increased surface roughness on the profile loss appear in the separated shear layer and reattachment. Increased surface roughness increases turbulent mixing in the separated shear layer. Thus, the shear layer thickness and momentum deficit are reduced. In addition, increased surface roughness reduces the length scale and turbulence intensity of the shed vortices. Consequently, turbulent mixing and momentum deficit during reattachment of boundary layers are decreased, resulting in a lower profile loss.

Aerodynamic Mitigation of Mechanical Constraints in Small Compressor Blade Profiles

The drive for ever higher aircraft efficiency inevitably leads to a reduction in core compressor size; over the past 30 years the height of the rear stages of a typical aeroengine has reduced by 40%. This trend will continue; a further reduction of almost 50% is expected by 2050. It is shown in this paper that the two greatest sources of increased profile loss as core size is reduced comes from the mechanical and manufacturing constraints placed on the maximum and trailing edge thickness. Furthermore, mitigation strategies are presented which successfully minimized or eliminated these loss sources. An experimental study showed that local modification to the trailing edge shape from a semi-circle to a more elliptical shape almost entirely eliminated the increase in trailing edge loss associated with a reduction in compressor size. Elliptical trailing edges improved loss by both narrowing the wake and increasing the base pressure. Despite the local nature of the modification, the elliptical trailing edges had a powerful effect on flow turning. This was accounted for by redesigning each design to achieve the specified exit flow angle. A set of “as manufactured” small core blades was created by combining measured data from existing manufacture methods with airfoils scaled to ensure minimum thicknesses necessary for mechanical integrity along the blade were not breached. Core sizes down to 40% of current designs were run in CFD. It is often assumed that the loss in efficiency as core size is reduced is unavoidable. However, it was shown that approximately half of this is directly due to the increase in thickness to satisfy tolerancing. Part to part deviation has negligible effect at all but the smallest compressor size tested (at 40% of current values).

A Key Flow Parameter to the Profile Loss of Low-Pressure Turbine Blades

For an optimum performance design of low-pressure turbine (LPT) blades, it is crucial to understand the generation mechanism of profile loss properly. As the profile loss is usually taken to be the loss generated inside the blade boundary layer due to viscous effects, most of the efforts for the performance optimization have concentrated on the reduction in the boundary layer loss using the flow parameters that represent the loss generation in the boundary layers. Kodama and Funazaki [1] investigated the generation mechanism of profile loss from a view point of blade drag forces, friction drag force and pressure drag force, and suggested that the loss due to pressure drag is dominant in the profile loss of a typical LPT blade. The loss due to pressure drag is not a boundary layer loss that is generated in the boundary layers, but a mixing loss that is generated downstream of the trailing edge. It is necessary to clarify a key flow parameter to the loss due to pressure drag for an effective performance optimization. This paper aims at investigating the flow parameter that is a measure of the profile loss. In the investigation, the profile loss is broken down into the loss components which are expressed by the boundary layer integral parameters at the trailing edge. Then the loss components are categorized into the loss due to friction drag or the loss due to pressure drag. The loss level of each component is evaluated by using the results of steady Reynolds Averaged Navier-Stokes (RANS) simulations to assess the contribution to the total profile loss. The evaluations are conducted for two kinds of blade profiles at three different Reynolds numbers. It is found that the largest contributor to the loss due to pressure drag, consequently to the total profile loss, is the loss associated with a mixing of accelerated free stream flow by the flow blockage at the trailing edge plane. The loss level is simply determined by the flow blockage. This suggests that the flow blockage at the trailing edge plane is the most important flow parameter for an optimum performance design of LPT blades.

A New Loss Generation Body Force Model for Fan/Compressor Blade Rows: An Artificial-Neural-Network Based Methodology

Body force models of fans and compressors are widely employed for predicting performance due to the reduction in computational cost associated with their use, particularly in nonuniform inflows. Such models are generally divided into a portion responsible for flow turning and another for loss generation. Recently, accurate, uncalibrated turning force models have been developed, but accurate loss generation models have typically required calibration against higher fidelity computations (especially when flow separation occurs). In this paper, a blade profile loss model is introduced which requires the trailing edge boundary layer momentum thicknesses. To estimate the momentum thickness for a given blade section, an artificial neural network is trained using over 400,000 combinations of blade section shape and flow conditions. A blade-to-blade flow field solver is used to generate the training data. The model obtained depends only on blade geometry information and the local flow conditions, making its implementation in a typical computational fluid dynamics framework straightforward. We show good agreement in the prediction of profile loss for 2D cascades both on and off design in the defined ranges for the neural network training.

The Effect of Reaction on Compressor Performance

Reaction is the fundamental parameter by which the asymmetry of the velocity triangle of a stage is set. Little is understood about the effect that reaction has on either the efficiency or the operating range of a compressor. A particular difficulty in understanding the effect of reaction is that the rotor and stator have a natural asymmetry caused by the centrifugal effects in the rotor boundary layer being much larger than that in the stator boundary layer. In this paper a novel approach has been taken: McKenzie's ‘linear repeating stage’ concept is used to remove the centrifugal effects. The centrifugal effects are then reintroduced as a body force. This allows the velocity triangle effect and centrifugal force effect to be decoupled. The paper shows the surprising result that, depending on how the solidity is set, a 50% reaction stage can either result in the maximum, or the minimum, profile loss. When the centrifugal effects are removed, 50% reaction is shown to minimise endwall loss, maximise stage efficiency and maximise operating range. When the centrifugal effects are reintroduced, the compressor with the maximum design efficiency is found to rise in reaction by 5% (from 50% reaction to 55% reaction) and the compressor with the maximum operating range is found to rise in reaction by 15% (from 50% reaction to 65% reaction).

Influence of Scanner Precision and Analysis Software in Quantifying Three-Dimensional Intraoral Changes: Two-Factor Factorial Experimental Design

Background Three-dimensional scans are increasingly used to quantify biological topographical changes and clinical health outcomes. Traditionally, the use of 3D scans has been limited to specialized centers owing to the high cost of the scanning equipment and the necessity for complex analysis software. Technological advances have made cheaper, more accessible methods of data capture and analysis available in the field of dentistry, potentially facilitating a primary care system to quantify disease progression. However, this system has yet to be compared with previous high-precision methods in university hospital settings. Objective The aim of this study was to compare a dental primary care method of data capture (intraoral scanner) with a precision hospital-based method (laser profilometer) in addition to comparing open source and commercial software available for data analysis. Methods Longitudinal dental wear data from 30 patients were analyzed using a two-factor factorial experimental design. Bimaxillary intraoral digital scans (TrueDefinition, 3M, UK) and conventional silicone impressions, poured in type-4 dental stone, were made at both baseline and follow-up appointments (mean 36 months, SD 10.9). Stone models were scanned using precision laser profilometry (Taicaan, Southampton, UK). Three-dimensional changes in both forms of digital scans of the first molars (n=76) were quantitatively analyzed using the engineering software Geomagic Control (3D Systems, Germany) and freeware WearCompare (Leeds Digital Dentistry, UK). Volume change (mm3) was the primary measurement outcome. The maximum point loss (μm) and the average profile loss (μm) were also recorded. Data were paired and skewed, and were therefore compared using Wilcoxon signed-rank tests with Bonferroni correction. Results The median (IQR) volume change for Geomagic using profilometry and using the intraoral scan was –0.37 mm3 (–3.75-2.30) and +0.51 mm3 (–2.17-4.26), respectively (P<.001). Using WearCompare, the median (IQR) volume change for profilometry and intraoral scanning was –1.21 mm3 (–3.48-0.56) and –0.39 mm3 (–3.96-2.76), respectively (P=.04). WearCompare detected significantly greater volume loss than Geomagic regardless of scanner type. No differences were observed between groups with respect to the maximum point loss or average profile loss. Conclusions As expected, the method of data capture, software used, and measurement metric all significantly influenced the measurement outcome. However, when appropriate analysis was used, the primary care system was able to quantify the degree of change and can be recommended depending on the accuracy needed to diagnose a condition. Lower-resolution scanners may underestimate complex changes when measuring at the micron level.

The Effect of Reaction on Compressor Performance

Reaction is the fundamental parameter by which the asymmetry of the velocity triangle of a stage is set. Little is understood about the effect that reaction has on either the efficiency or the operating range of a compressor. A particular difficulty in understanding the effect of reaction is that the rotor and stator have a natural asymmetry caused by the centrifugal effects in the rotor boundary layer being much larger than that in the stator boundary layer. In this paper a novel approach has been taken: McKenzie’s ‘linear repeating stage’ concept is used to remove the centrifugal effects. The centrifugal effects are then reintroduced as a body force. This allows the velocity triangle effect and centrifugal force effect to be decoupled. The paper shows the surprising result that, depending on how the solidity is set, a 50% reaction stage can either result in the maximum, or the minimum, profile loss. When the centrifugal effects are removed, 50% reaction is shown to minimise endwall loss, maximise stage efficiency and maximise operating range. When the centrifugal effects are reintroduced, the compressor with the maximum design efficiency is found to rise in reaction by 5% (from 50% reaction to 55% reaction) and the compressor with the maximum operating range is found to rise in reaction by 15% (from 50% reaction to 65% reaction).

Effects of the Large Eddy Simulation Calculation Parameters on Prediction of Profile Loss in an Axial Cascade at Off-Design Incidence

Large Eddy Simulation (LES) has already been widely applied for unsteady flow simulations (e.g., axial cascade simulation), showing its superiority in the case of turbulent flow fields. However, there are several obstacles in the practical use of LES as a cascade design tool for gas turbine engines. Several calculation parameters such as the inlet turbulence, spanwise extents, and many other boundary conditions, affect calculation results for LES simulation and can cause reliability problems. This study concentrates on investigating the effects of LES calculation parameters on the prediction of profile loss at high air inlet angle off-design condition. The LES flow field results were found to be sensitive to four parameters (i.e., inlet turbulence, spanwise domain extent, time step size, and outlet domain). The inlet turbulence affects the size of the laminar separation bubble, while the spanwise domain extent affects the velocity fluctuations of the wake in correspondence of the large-scale vortices shedding. The time step size affects the velocity fluctuations of the wake because of the number of times that the low-pass filter is applied. Finally, the outlet domain extent affects the whole calculation domain (from the leading edge to the wake). Overall, the results show that all four calculation parameters can change the flow characteristics, affecting the prediction of profile loss at off-design condition when using LES.