Recherches Scientifiques Luxembourg

The role of geothermal energy in the energy transition is rapidly gaining recognition. Its storage technologies provide flexible solutions to address the intermittent nature of other renewable resources. Abandoned mines and other underground caverns have recently attracted particular attention as valuable geothermal reservoirs.In this context, a pilot project was recently initiated in the flooded abandoned underground slate exploitation area of Martelange (Belgium & Luxembourg). This area, with its long industrial history, represents a significant opportunity for future geothermal applications. The large volume of water stored within these underground caverns, in combination with heat-pumps, is sufficient to provide heating and cooling for a quite large number of buildings.Where natural regeneration, depending on groundwater flows and the thermal conductivity of the surrounding soil, proves insufficient and where water temperature changes exceed 5 °C, active regeneration may be implemented. This involves the installation of (un)glazed solar thermal absorbers to collect low-cost solar heat during summer and cold during winter, transforming the system into a form of solar district heating. In such a scenario, the flooded slate caverns would function as a seasonal energy storage body.The objective of the project is to develop a modern, highly energy-efficient urbanized area with a commercial zone, based on a low-temperature urban heat and cold network fed by heat-pumps. Achieving this requires a detailed assessment of the local geological and hydrogeological conditions, especially the structure and connections of the slate caverns, to ensure optimal and sustainable use of the available water resources. The first results from this study will be presented.

In safety-critical systems (e.g., autonomous vehicles and robots), Deep Neural Networks (DNNs) are becoming a key component for computer vision tasks, particularly semantic segmentation. Further, since DNN behavior cannot be assessed through code inspection and analysis, test automation has become an essential activity to gain confidence in the reliability of DNNs. Unfortunately, state-of-the-art automated testing solutions largely rely on simulators, whose fidelity is always imperfect, thus affecting the validity of test results. To address such limitations, we propose to combine meta-heuristic search, used to explore the input space using simulators, with Generative Adversarial Networks (GANs), to transform the data generated by simulators into realistic input images. Such images can be used both to assess the DNN accuracy and to retrain the DNN more effectively. We applied our approach to a state-of-the-art DNN performing semantic segmentation, in two different case studies, and demonstrated that it outperforms a state-of-the-art GAN-based testing solution and several other baselines. Specifically, it leads to the largest number of diverse images leading to the worst DNN accuracy. Further, the images generated with our approach, lead to the highest improvement in DNN accuracy when used for retraining. In conclusion, we suggest to always integrate a trained GAN to transform test inputs when performing search-driven, simulator-based testing.

The development of effective socio-technical information systems requires designers to balance ethical goals with functional capabilities and business objectives. To address this challenge, we propose EDGE, a novel ethical design method that focuses on two key activities in the systems development process—ethical grounding and evaluation of ethical impacts. We provide an empirical example of the EDGE method by applying it to the redesign of a sustainable application in the energy industry. This empirical example reveals the need for ethical clarity to ground and evaluate the app to achieve improved ethical performance and social impacts. We conclude with the implications of the new ethical design method and future research directions for its application and enhancement.

The presented work deals with the improvement of the evaporation model of the ATHLET (Analysis of Thermal and Hydraulics of Leaks and Transients) system code to be applied to a passive containment cooling system of a nuclear power plant. For the model validation, INTRAVIT (Investigation of Passive Heat Transfer in a Variably Inclined Tube) test facility setup at the University of Luxembourg was used. The first part of the paper presents a review of the existing literature on evaporation models that revealed that those models significantly simplify the physical processes that occur. Next, a modified evaporation model is proposed that offers a realistic description of various evaporation processes and the start of bubble formation using a nucleation model, and a surface density calculation model is introduced that is necessary for evaporation simulation. The final part of this work explored five different system configurations to test the evaporation model: three condenser tube inclinations (5 deg, 60 deg, and 90 deg), two riser lengths (1 m and 2.5 m), and different thermal loads. They made it possible to simulate several experiments for stable and unstable natural circulation and to verify the proposed model.

Accurate cropland classification is important for sustainable land management and agricultural planning. Remote sensing data, such as the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI), have proven instrumental in vegetation monitoring. Traditional ma-chine learning and deep learning have shown remarkable results in land classification using remote sensing data. However, when facing an immense volume of data or high-dimensional features, deep learning requires large models, extensive parameters, and substantial training resources. To address the above issues, this paper proposes Cropland Quantum Learning (CQL), a quantum-classical hybrid method that utilizes quantum machine learning to extract features from geospatial information, and integrates it with a single-layer fully connected classifier to locate cultivation regions of a given target crop. We conduct comprehensive experiments to demonstrate the effectiveness of our proposed method on NDVI and EVI datasets. The results show that the proposed CQL can achieve performance comparable to traditional deep learning while significantly reducing the number of model parameters. In some datasets, even achieved better results in multiple metrics. This study provides a new solution for quantum-enhanced approaches in geospatial analysis.

Machine learning has achieved remarkable success across diverse domains. Nevertheless, concerns about interpretability in black-box models, especially within Deep Neural Networks (DNNs), have become pronounced in safety-critical fields like healthcare and finance. Classical machine learning (ML) classifiers, known for their higher interpretability, are preferred in these domains. Similar to DNNs, classical ML classifiers can exhibit bugs that could lead to severe consequences in practice. Test input prioritization has emerged as a promising approach to ensure the quality of an ML system, which prioritizes potentially misclassified tests so that such tests can be identified earlier with limited manual labeling costs. However, when applying to classical ML classifiers, existing DNN test prioritization methods are constrained from three perspectives: 1) Coverage-based methods are inefficient and time-consuming; 2) Mutation-based methods cannot be adapted to classical ML models due to mismatched model mutation rules; 3) Confidence-based methods are restricted to a single dimension when applying to binary ML classifiers, solely depending on the model’s prediction probability for one class. To overcome the challenges, we propose MLPrior, a test prioritization approach specifically tailored for classical ML models. MLPrior leverages the characteristics of classical ML classifiers (i.e., interpretable models and carefully engineered attribute features) to prioritize test inputs. The foundational principles are: 1) tests more sensitive to mutations are more likely to be misclassified, and 2) tests closer to the model’s decision boundary are more likely to be misclassified. Building on the first concept, we design mutation rules to generate two types of mutation features (i.e., <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">model mutation features</b> and <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">input mutation features</b> ) for each test. Drawing from the second notion, MLPrior generates <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">attribute features</b> of each test based on its attribute values, which can indirectly reveal the proximity between the test and the decision boundary. For each test, MLPrior combines all three types of features of it into a final vector. Subsequently, MLPrior employs a pre-trained ranking model to predict the misclassification probability of each test based on its final vector and ranks tests accordingly. We conducted an extensive study to evaluate MLPrior based on 185 subjects, encompassing natural datasets, mixed noisy datasets, and fairness datasets. The results demonstrate that MLPrior outperforms all the compared test prioritization approaches, with an average improvement of 14.74%∼66.93% on natural datasets, 18.55%∼67.73% on mixed noisy datasets, and 15.34%∼62.72% on fairness datasets.

This paper presents findings on machine learning agent behavior prediction in a board game application developed by a group of students. The goal of this research is to create a model facilitating collaboration between a user and an AI to play together in the board game using a Human-in-the-Loop architecture. By injecting explainability, the aim is to enhance communication and understanding between the user and the AI agent. Featuring a competitive Artificial Intelligence (AI) based on the Proximal Policy Optimization model, this research explores methods to make AI decisions transparent for enhanced player understanding. Two predictive models, a Decision Tree (DT) and a Deep Learning (DL) classifier, were developed and compared. The results show that the DT model is effective for short-term predictions but limited in broader applications, while the DL classifier shows potential for long-term prediction without requiring direct access to the game's AI. This study contributes to understanding human-AI interaction in gaming and offers insights into AI decision-making processes.

Orthogonal frequency-division multiplexing (OFDM) is widely used to mitigate inter-symbol interference (ISI) from multipath fading. However, the open nature of wireless OFDM systems makes them vulnerable to jamming attacks. In this context, pilot jamming is critical as it focuses on corrupting the symbols used for channel estimation and equalization, degrading the system performance. Although neural networks (NNs) can improve channel estimation and mitigate pilot jamming penalty, they are also themselves susceptible to malicious perturbations known as adversarial examples. If the jamming attack is crafted in order to fool the NN, it represents an adversarial example that impairs the proper behavior of OFDM systems. In this work, we explore two machine learning (ML)-based jamming strategies that are especially intended to degrade the performance of ML-based channel estimators, in addition to a traditional Additive White Gaussian Noise (AWGN) jamming attack. These ML-based attacks create noise patterns designed to reduce the precision of the channel estimation process, thereby compromising the reliability and robustness of the communication system. We highlight the vulnerabilities of wireless communication systems to ML-based pilot jamming attacks that corrupts symbols used for channel estimation, leading to system performance degradation. To mitigate these threats, this paper proposes an adversarial training defense mechanism desined to counter jamming attacks. The effectiveness of this defense is validated through simulation results, demonstrating improved channel estimation performance in the presence of jamming attacks. The proposed defense methods aim to enhance the resilience of OFDM systems against pilot jamming attacks, ensuring more robust communication in wireless environments.