Umeå University's logo

Wheel loaders in mines and construction sites repeatedly load soil from a pile to load receivers. Automating this task presents a challenging planning problem since each loading’s performance depends on the pile state, which depends on previous loadings. We investigate an end-to-end optimization approach considering future loading outcomes and transportation costs between the pile and load receivers. To predict the evolution of the pile state and the loading performance, we use world models that leverage deep neural networks trained on numerous simulated loading cycles. A look-ahead tree search optimizes the sequence of loading actions by evaluating the performance of thousands of action candidates, which expand into subsequent action candidates under the predicted pile states recursively. Test results demonstrate that, over a horizon of 15 sequential loadings, the look-ahead tree search is 6% more efficient than a greedy strategy, which always selects the action that maximizes the current single loading performance, and 14% more efficient than using a fixed loading controller optimized for the nominal case.

Structured regularization allows machine learning models to consider spatial relationships among parameters, leading to results that generalize better and are more interpretable compared to norm penalties. In this study, we evaluated a novel structured regularization method that incorporates approximate morphology operators defined using harmonic mean-based fW-filters. We extended this method to multiclass classification and conducted experiments aimed at classifying magnetic resonance images (MRI) of subjects into four stages of Alzheimer's disease progression. The experimental results demonstrate that the novel structured regularization method not only performs better than standard sparse and structured regularization methods in terms of prediction accuracy (ACC), F1 scores, and the area under the receiver operating characteristic curve (AUC), but also produces interpretable coefficient maps.

We propose a novel way to incorporate morphology operators through structured regularization of machine learning models. Specifically, we introduce a feature map in the models that performs structured variable selection. The feature map is automatically processed by approximate morphology operators and is learned together with the model coefficients. Experiments were conducted with linear regression on both synthetic data, demonstrating that the proposed methods are effective in selecting groups of parameters with much less noise than baseline models, and on three-dimensional T1-weighted brain magnetic resonance images (MRI) for age prediction, demonstrating that the proposed methods enforce sparsity and select homogeneous regions of non-zero and relevant regression coefficients. The proposed methods improve interpretability in pattern analysis. The minimum size of features in the structured variable selection can be controlled by adjusting the structuring element in the approximate morphology operator, tailored to the specific study of interest. With these added benefits, the proposed methods still perform on par with commonly used variable selection and structured variable selection methods in terms of the coefficient of determination and the Pearson correlation coefficient.

Near-field and radiation coupling between nearby radiating elements is unavoidable, and it is considered a limiting factor for applications in wireless communications and active sensing. This article proposes a density-based topology optimization approach to design decoupling networks for such systems. The decoupling network is designed by formulating an optimization problem that considers both energy transmission and reflection at the network ports. We replace the radiating elements by their time-domain impulse response for efficient computations and to enable the solution of the design problem using gradient-based optimization methods. We use the adjoint-field method to compute the gradients of the optimization objectives. Additionally, nonlinear filters are applied during the optimization procedure to impose minimum-size control on the optimized designs. We demonstrate the concept by designing the decoupling network for a two-element planar antenna array; the antenna is designed in a separate optimization problem. The optimized decoupling networks provide a signal path that destructively interferes with the coupling between the radiating elements while preserving their individual matching to the feeding ports. Compact decoupling networks capable of suppressing the mutual coupling by more than 10 dB between two closely separated planar antennas operating around 2.45 GHz are presented and validated experimentally.

We use a density-based topology optimization approach to design dualband planar metallic antennas. The design problem is formulated based on the time-domain Maxwell's equations, solved using the finite-difference timedomain (FDTD) method. The antenna design is formulated as an optimization problem where the received and reflected energy by the antenna in two frequency bands, centered around 2.5 GHz and 5.5 GHz, are optimized. Two design examples that exhibit outstanding performance are presented. In one design case, we employ a nonlinear filtering scheme to impose size control on the optimized design and ensure manufacturability.

It is well understood that spanwise arrays of roughness elements can be used to generate steady streaks in boundary layers. This modulation of the boundary layer has the potential to attenuate the growth of Tollmien–Schlichting (TS) waves which can lead to the transition to turbulence in low turbulence intensity environments, such as those experienced by an aircraft's fuselage in atmospheric flight. This article applies density based topology optimization in order to design roughness elements capable of exploiting the aforementioned stabilizing effect as a means of passive flow control. The geometry of the roughness elements are represented using a Brinkman penalization when conducting Direct Numerical Simulations (DNS) to simulate the streaky boundary layer flow. Similarly, the unsteady linearized Navier–Stokes equations are evolved to assess the spatial growth of the TS waves across the flat plate. The optimization procedure aims to minimize the TS wave amplitude at a given downstream position while a novel constraint is used promoting a stable baseflow. The optimization problem is solved with gradient descent algorithms where the adjoint-variable method is used to compute gradients. This method has been applied to three initial material distributions yielding three distinct and novel designs capable of damping the downstream growth of the TS wave significantly more than a reference Minature Vortex Generator (MVG) of comparable size. The optimized designs and streaky baseflows they induce are then studied using an energy budget analysis and local stability analysis.

Sonic black holes (SBHs) are waveguides intended to slow down the wave propagation speed and focus the energy towards the end of the device. However, the extent to which these effects occur, as well as the degree of wave dispersion introduced, has not been systematically quantified. This article investigates these aspects through transient finite-element computations, analyzing the properties of a novel, numerically optimized SBH with enhanced wave-focusing capabilities. The investigation utilizes the lossless acoustic wave equation as well as a linearized compressible flow formulation to account for viscothermal losses. We analyze the wave focusing and filtering properties of the SBH by monitoring the pressure amplitude and the transmission and reflection coefficients. Moreover, we examine the effective wave propagation speed along the centerline of SBH and assess the similarity of pressure wave packets using cross-correlations. Our results reveal that the optimized SBH not only enhances wave focusing but also on average effectively slows down wave propagation, demonstrating the device's potential as a true sonic black hole. By investigating two crucial aspects - wave-slowing effect and signal dispersion - that were not previously explored, we provide a deeper understanding of the device's functionality and operational mechanisms, including how its design influences wave-focusing performance and local wave speed.

This article proposes a boundary strip indicator for density-based topology optimization that can be used to estimate the design’s surface area (perimeter in 2D) or identify a coating layer. We investigate the theoretical properties of the proposed boundary strip indicator and propose a differentiable approximation that preserves key properties, such as non-negativity. Finally, we use the boundary strip indicator in a heat conduction design optimization problem for a coated structure. The resulting designs show a strong dependence on the properties of the coating.

We explore sim-to-real transfer of deep reinforcement learning controllers for a heavy vehicle with active suspensions designed for traversing rough terrain. While related research primarily focuses on lightweight robots with electric motors and fast actuation, this study uses a forestry vehicle with a complex hydraulic driveline and slow actuation. We simulate the vehicle using multibody dynamics and apply system identification to find an appropriate set of simulation parameters. We then train policies in simulation using various techniques to mitigate the sim-to-real gap, including domain randomization, action delays, and a reward penalty to encourage smooth control. In reality, the policies trained with action delays and a penalty for erratic actions perform nearly at the same level as in simulation. In experiments on level ground, the motion trajectories closely overlap when turning to either side, as well as in a route tracking scenario. When faced with a ramp that requires active use of the suspensions, the simulated and real motions are in close alignment. This shows that the actuator model together with system identification yields a sufficiently accurate model of the actuators. We observe that policies trained without the additional action penalty exhibit fast switching or bang–bang control. These present smooth motions and high performance in simulation but transfer poorly to reality. We find that policies make marginal use of the local height map for perception, showing no indications of predictive planning. However, the strong transfer capabilities entail that further development concerning perception and performance can be largely confined to simulation.

The waveguide acoustic black hole (WAB) effect is a promising approach for controlling wave propagation in various applications, especially for attenuating sound waves. While the wave-focusing effect of structural acoustic black holes has found widespread applications, the classical ribbed design of waveguide acoustic black holes (WABs) acts more as a resonance absorber than a true wave-focusing device. In this study, we employ a computational design optimization approach to achieve a conceptual design of a WAB with enhanced wave-focusing properties. We investigate the influence of viscothermal boundary losses on the optimization process by formulating two distinct cases: one neglecting viscothermal losses and the other incorporating these losses using a recently developed material distribution topology optimization technique. We compare the performance of optimized designs in these two cases with that of the classical ribbed design. Simulations using linearized compressible Navier–Stokes equations are conducted to evaluate the wave-focusing performance of these different designs. The results reveal that considering viscothermal losses in the design optimization process leads to superior wave-focusing capabilities, highlighting the significance of incorporating these losses in the design approach. This study contributes to the advancement of WAB design and opens up new possibilities for its applications in various fields.