Science informatique et génie électrique - Publications // Electrical Engineering and Computer Science - Publications

Permanent URI for this collectionhttps://hdl.handle.net/10393/12853

Browse

Now showing 1 - 20 of 106

Long-Term Photovoltaic System Performance in Cold, Snowy Climates
(2025) Tonita, Erin M.; Jordan, Dirk C.; Ovaitt, Silvana; Toal, Henry; Hinzer, Karin; Pike, Christopher; Deline, Chris
As countries around the world transition towards renewable energy, there is increasing interest in using photovoltaic (PV) technologies to help decarbonize northern and alpine communities due to their scalability and affordability. However, a barrier to large-scale adoption of PV in cold climates is long-term performance uncertainty under snowfall, freeze–thaw cycles, low temperatures, and high winds. In this work, we provide a comprehensive review of published silicon degradation rates in cold Köppen–Geiger climate classifications of Dfb (humid continental), Dfc (subarctic), and ET (tundra). We first analyze the system degradation rates of three subarctic ground-mounted photovoltaic sites in North America using the RdTools year-on-year method: an Al-BSF double-axis tracking site in Fairbanks, Alaska (65° N); a PERC and silicon heterojunction bifacial vertical and south-tilted site in Fairbanks, Alaska; and a PERC south-facing fixed-tilt site in Fort Simpson, Northwest Territories (62° N). Degradation rates of these newly analyzed sites vary between −0.4%/year and −1.5%/year. Combining these data with previously reported cold climate degradation rates, we show that the distribution of cold climate degradation peaks at −0.1%/year to −0.2%/year but has a large tail with rates above −0.5%/year. The average reported cold climate degradation rate is −0.45%/year, whereas the median value is −0.33%/year. These results suggest that despite frequent freeze–thaw cycles and potential exposure to high wind and snow loads, PV systems in cold climates tend to degrade slower than PV systems in warmer climates. The limited sample size of reported degradation rates in cold climates (27) motivates the need for further data acquisition and monitoring efforts as new technologies are deployed.
Underactuated MIMO Airship Control Based on Online Data-Driven Reinforcement Learning
(IEEE, 2023-10-01) Boase, Derek; Gueaieb, Wail; Miah, Md Suruz
In this work, a novel online model-free controller for an underactuated dirigible is developed based on reinforcement learning and optimal control theory. A reinforcement learning structure is used while overcoming the dependence of the value function on future values by introducing a neural network that is adapted using input-output data. The suboptimal critic neural network is structured such that optimality is guaranteed over the interval from which the data is valid. The system performance is validated using a highly realistic physics engine, Gazebo, with the robot operating system (ROS) interface and the results are compared to the performance of a model-based controller specifically designed to control the airship model. It is emphasized that the proposed formulation does not leverage any knowledge of vehicle dynamics and thus is considered a vehicle agnostic control strategy.
A state-of-the-art review on topology and differential geometry-based robotic path planning—part I: Planning under static constraints
(2024-03-20) Radhakrishnan, Sindhu; Gueaieb, Wail
Autonomous robotics has permeated several industrial, research and consumer robotic applications, of which path planning is an important component. The path planning algorithm of choice is influenced by the application at hand and the history of algorithms used for such applications. The latter is dependent on an extensive conglomeration and classification of path planning literature, which is what this work focuses on. Specifically, we accomplish the following: typical classifications of path planning algorithms are provided. Such classifications rely on differences in knowledge of the environment (known/unknown), robot (model-specific/generic), and constraints (static/dynamic). This classification however, is not comprehensive. Thus, as a resolution, we propose a detailed taxonomy based on a fundamental parameter of the space, i.e. its ability to be characterized as a set of disjoint or connected points. We show that this taxonomy encompasses important attributes of path planning problems, such as connectivity and partitioning of spaces. Consequently, path planning spaces in robotics may be viewed as simply a set of points, or as manifolds. The former can further be divided into unpartitioned and partitioned spaces, of which the former uses variants of sampling algorithms, optimization algorithms, model predictive controls, and evolutionary algorithms, while the latter uses cell decomposition and graph traversal, and sampling-based optimization techniques.This article achieves the following two goals: The first is the introduction of an all-encompassing taxonomy of robotic path planning. The second is to streamline the migration of path planning work from disciplines such as mathematics and computer vision to robotics, into one comprehensive survey. Thus, the main contribution of this work is the review of works for static constraints that fall under the proposed taxonomy, i.e., specifically under topology and manifold-based methods. Additionally, further taxonomy is introduced for manifold-based path planning, based on incremental construction or one-step explicit parametrization of the space.
Constraint-Free Discretized Manifold-Based Path Planner
(2023-10-14) Radhakrishnan, Sindhu; Gueaieb, Wail
Autonomous robotic path planning in partially known environments, such as warehouse robotics, deals with static and dynamic constraints. Static constraints include stationary obstacles, robotic and environmental limitations. Dynamic constraints include humans, robots and dis/appearance of anticipated dangers, such as spills. Path planning consists of two steps: First, a path between the source and target is generated. Second, path segments are evaluated for constraint violation. Sampling algorithms trade memory for maximal map representation. Optimization algorithms stagnate at non-optimal solutions. Alternatively, detailed grid-maps view terrain/structure as expensive memory costs. The open problem is thus to represent only constraint-free, navigable regions and generating anticipatory/reactive paths to combat new constraints. To solve this problem, a Constraint-Free Discretized Manifolds-based Path Planner (CFDMPP) is proposed in this paper. The algorithm’s first step focuses on maximizing map knowledge using manifolds. The second uses homology and homotopy classes to compute paths. The former constructs a representation of the navigable space as a manifold, which is free of apriori known constraints. Paths on this manifold are constraint-free and do not have to be explicitly evaluated for constraint violation. The latter handles new constraint knowledge that invalidate the original path. Using homology and homotopy, path classes can be recognized and avoided by tuning a design parameter, resulting in an alternative constraint-free path. Path classes on the discretized constraint-free manifold characterize numerical uniqueness of paths around constraints. This designation is what allows path class characterization, avoidance, and querying of a new path class (multiple classes with tuning), even when constraints are simply anticipatory.
A state-of-the-art review on topology and differential geometry-based robotic path planning—part II: Planning under dynamic constraints
(2024-03-25) Radhakrishnan, Sindhu; Gueaieb, Wail
Path planning is an intrinsic component of autonomous robotics, be it industrial, research or consumer robotics. Such avenues experience constraints around which paths must be planned. While the choice of an appropriate algorithm is application-dependent, the starting point of an ideal path planning algorithm is the review of past work. Historically, algorithms were classified based on the three tenets of autonomous robotics which are the ability to avoid different constraints (static/dynamic), knowledge of the environment (known/unknown) and knowledge of the robot (general/model specific). This division in literature however, is not comprehensive, especially with respect to dynamics constraints. Therefore, to remedy this issue, we propose a new taxonomy, based on the fundamental tenet of characterizing space, i.e., as a set of distinct, unrelated points or as a set of points that share a relationship. We show that this taxonomy is effective in addressing important parameters of path planning such as connectivity and partitioning of spaces. Therefore, path planning spaces may now be viewed either as a set of points or, as a space with structure. The former relies heavily on robot models, since the mathematical structure of the environment is not considered. Thus, the approaches used are variants of optimization algorithms and specific variants of model-based methods that are tailored to counteract effects of dynamic constraints. The latter depicts spaces as points with inter-connecting relationships, such as surfaces or manifolds. These structures allow for unique characterizations of paths using homotopy-based methods. The goals of this work, viewed specifically in light with dynamic constraints, are therefore as follows: First, we propose an all-encompassing taxonomy for robotic path planning literature that considers an underlying structure of the space. Second, we provide a detailed accumulation of works that do focus on the characterization of paths in spaces formulated to show underlying structure. This work accomplishes the goals by doing the following: It highlights existing classifications of path planning literature, identifies gaps in common classifications, proposes a new taxonomy based on the mathematical nature of the path planning space (topological properties), and provides an extensive conglomeration of literature that is encompassed by this new proposed taxonomy.
Modeling Human-AI Cognitive Alignment on Protected Data
(2025-02-14) Darveau, Vivianne; Darveau, Peter
This study explores the quantifying of cognitive alignment between human expert reasoning and Large Language Model (LLM) generated solutions, in protected sensitive data environments, through Research Data Management (RDM) practices that are crucial to trustworthy AI systems. Using economic risk assessments as our data domain, we propose a novel approach that leverages oneAPI's unified computing capabilities to process and synthesize sensitive data, while maintaining privacy, to establish a performance baseline for human-centered Artificial Intelligence (AI). Our preliminary study analyzes 10 economic cases, first by modeling the topics with Latent Dirichlet Allocation (LDA) and human analysis, and then by comparing patterns with the LLM generated insights using accelerated topic modeling. The methodology introduces a four-tier privacy preservation metric that quantifies information exposure rates, entity detection, and topic-level abstraction. Initial results demonstrate a 82% topic alignment between human-AI reasoning patterns, while maintaining a privacy preservation of 84% on our proposed scale. The oneAPI implementation shows promising results in handling unified computer-intensive privacy-preserving transformations. This research contributes to the field of privacy-aware AI-human collaboration in sensitive data domains, where reasoning alignment and data protection are crucial.
The use of AI and robotics in armed conflicts
(2024-12-01) Meslin, Alexis; Ten Thij, Esger; Novitzky, Peter; Intahchomphoo, Channarong
This systematic literature review (SLR) explores existing and newly emergent ethical and legal challenges associated with the use of AI and robotics in armed conflicts. We conducted an extensive review of relevant scholarly publications associated with (lethal) autonomous weapons systems (LAWS). Besides the ethical and legal principles, we also explore emergent technical applications associated with these technologies in armed conflict(s). Our particular focus is to compare literature from the last 12 years with publications since the outbreaks of recent armed conflicts from the perspective of LAWS. We engage in exploring and identifying the shifts in ethical arguments and discourse, as well as shifts in policy subject themes, and standards setting around the use of emergent technology in relation with AI and robotics. Our contribution analyses emergent socio-technical themes and arguments relevant for engineers, policy-makers, and other interdisciplinary scholars across a variety of disciplines.
Planarizing Spalled GaAs(100) Surfaces by MOVP
(2024-11-13) Forcade, Gavin P.; McMahon, William E.; Yoo, Nicholas; Neumann, Anica N.; Young, Michelle; Goldsmith, John; Collins, Sarah; Hinzer, Karin; Packard, Corinne E.; Steiner, Myles A.
III-V photovoltaic devices have demonstrated exceptional performance across various applications, with controlled crystal fracturing, known as controlled spalling, emerging as a promising method to reduce costs by enabling substrate reuse. Spalling GaAs(100) substrates, a commonly used substrate in III-V photovoltaics, results in faceted ridges that must be planarized to grow high-quality photovoltaic devices. Here we demonstrate that a GaAs(100) wafer offcut towards [01̅1] and spalled towards [011] can be efficiently planarized by growing C:GaAs by metal-organic vapor phase epitaxy (MOVPE) on the surface, with up to 95% of the nominally deposited material used to fill the valleys between ridges. We find that reducing the offcut to 2° enhances the planarizing capability of C:GaAs. A surface morphology model indicates that the density of surface dangling bonds significantly influences the growth evolution of undoped GaAs surfaces. In contrast, the model suggests that the effectiveness of C:GaAs as a smoothing layer stems from modifying the atomic surface structure and, consequently, the associated sticking coefficients of the facets, which can alter the evolution of surface morphology. Our findings provide guidelines for the epitaxial planarization of semiconductor surfaces and improve the understanding of MOVPE growth on non-planar surfaces.
AI Systems Adoption of Unified Research Data Management on Accelerator Computing
(2024-09-30) Darveau, Peter
Research data is expected to grow exponentially with the adoption of artificial intelligence (AI) and machine learning (ML). Robust data management practices are crucial for ensuring data integrity, provenance tracking, and adherence to ethical and regulatory standards, which is essential for building trustworthy AI systems. This paper explores the adoption of oneAPI, an open standards-based programming model, for streamlining research data management across diverse AI systems. It also explores containerization to ensure consistent execution across heterogeneous Cloud-based environments while providing security over sensitive data-based systems. By leveraging oneAPI's cross-architecture capabilities, including Data Parallel C++ (DPC++) and the other AI toolkits based on oneAPI, researchers can develop secure and performant AI solutions that seamlessly process and analyze sensitive data across heterogeneous computing environments. This unified approach proposes a framework for consistent data handling and reproducibility of research computing results where data confidentiality, security and integrity are concerns notably in the Cloud. Through a case study example, this paper discusses the benefits of adopting oneAPI for AI research data management (RDM), highlighting its potential to accelerate scientific discoveries while maintaining robust security and privacy standards.
Using Process Mining with Pre- and Post-Intervention Analysis to Improve Digital Service Delivery: A Governmental Case Study
(2024-08-30) Trottier, Jacques; Van Woensel, William; Wang, Xiaoyang; Mallur, Kavya; El-Gharib, Najah; Amyot, Daniel
We present a case study of Process Mining (PM) for per- sonnel security screening in the Canadian government. We consider cus- tomer (process time) and organizational (cost) perspectives. Further- more, in contrast to most published case studies, we assess the full pro- cess improvement lifecycle: pre-intervention analyses pointed out initial bottlenecks, and post-intervention analyses identified the intervention impact and remaining areas for improvement. Using PM techniques, we identified frequent exceptional scenarios (e.g., applications requir- ing amendment), time-intensive loops (e.g., employees forgetting tasks), and resource allocation issues (e.g., involvement of non-security person- nel). Subsequent process improvement interventions, implemented using a flexible low-code digital platform, reduced security briefing times from around 7 days to 46 hours, and overall process time from around 31 days to 26 days, on average. From a cost perspective, the involvement of hiring managers and security screening officers was significantly reduced. These results demonstrate how PM can become part of a broader digital trans- formation framework to improve public service delivery. The success of these interventions motivated subsequent government PM projects, and inspired a PM methodology, currently under development, for use in large organizational contexts such as governments.
A Model-Free Kullback-Leibler Divergence Filter for Anomaly Detection in Noisy Data Series
(2022-11-17) Zhou, Ruikun; Gueaieb, Wail; Spinello, Davide
We propose a Kullback-Leibler Divergence (KLD) filter to extract anomalies within data series generated by a broad class of proximity sensors, along with the anomaly locations and their relative sizes. The technique applies to devices commonly used in engineering practice, such as those mounted on mobile robots for non-destructive inspection of hazardous or other environments that may not be directly accessible to humans. The raw data generated by this class of sensors can be challenging to analyze due to the prevalence of noise over the signal content. The proposed filter is built to detect the difference of information content between data series collected by the sensor and baseline data series. It is applicable in a model-based or model-free context. The performance of the KLD filter is validated in an industrial-norm setup and benchmarked against a peer industrially-adopted algorithm.
Support Vector Machines: Modeling The Dual Cognitive Processes of an SVM
(2023) Darveau, Peter
Can machines think fast and logical like us? In this study, we explore whether support vector machines (SVMs) - the workhorses of AI (Artificial Intelligence) - exhibit human-like heuristic judgment alongside mathematical optimization. Our experiments reveal that nonlinear SVMs can act as cognitive mimics, making surprisingly "intuitive" shortcuts reminiscent of Kahneman and Tversky's dual process theory. Yet SVMs avoid our irrational biases by combining heuristics with optimal statistical learning. These cognitive cousins both leverage the power of mental shortcuts, but only humans trip up. Our multidisciplinary results illuminate the psychology behind AI's decisions, with profound implications. We glimpse mind-like heuristics emerging from rigid math, suggesting new directions for human-aligned AI. But mysteries remain on whether SVMs' heuristic gambles are features or flaws. Do their information-savvy shortcuts point towards the essence of intuition? We discuss implications for interpreting modern AI through cognitive psychology lenses while identifying key differences. This multidisciplinary work aims to provide novel empirical insights on the interplay between heuristic and optimal practices in an important class of machine learning algorithms. The results shed light on developing human-aligned classifiers that balance the strengths of both heuristic and logical thinking. This paper takes a step towards unravelling the inner workings of one of the most used artificial intelligence models, Support Vector Machines.
Decision Trees: Modeling with fast intuition and slow, deliberate analysis
(2023) Darveau, Peter
The Dual Nature of Decision Trees Decision trees demonstrate a fascinating duality between human intuition and mathematical optimization. Psychologists like Kahneman and Tversky revealed how people rely on mental shortcuts and biased, heuristic-based thinking. This mirrors how decision trees use simple, hierarchical branching based on key features - just like our minds categorize objects using decisive traits. Yet decision trees are also rigorously constructed by calculating metrics like information gain that maximize analytical power. This parallels the structured analysis of rational thinking, optimizing the tree mathematically. Supported by various works by D. Kahneman, Busemeyer et al., and researchers at the university of Ottawa, this duality gives decision trees their interpretability and versatility. The visual tree structure appeals to intuitive pattern recognition, while optimized construction exploits powerful analytical techniques. Understanding this fusion between intuitive shortcuts and calculated reasoning is key to advancing decision tree capabilities and addressing their ethical and regulated use in AI applications.
Persuasive system design for runners using smart sport training technologies
(2022) Gauthier-Kwan, Maxime
The goal of this paper is to understand the persuasive technologies used for smart sport training by runners. The Persuasive Systems Design (PSD) by Oinas-Kukkonen & Harjumaa (2009) was used as the framework, and its components were identified. This takes in consideration the analysis of the persuasion context (intent, event, strategy), design of system qualities, and intended behaviour changes. This work explores and consolidates findings from 35 works. It provides different perspectives from this multidisciplinary field such as psychology, technology, social sciences, sport sciences, etc. The hope is that this framework will help advance the understanding of persuasive technologies design for smart sport training for runners.
A Policy Iteration Approach for Flock Motion Control
(2021) Qu, Shuzheng; Abouheaf, Mohammed; Gueaieb, Wail; Spinello, Davide
The flocking motion control is concerned with managing the possible conflicts between local and team objectives of multi-agent systems. The overall control process guides the agents while monitoring the flock-cohesiveness and localization. The underlying mechanisms may degrade due to overlooking the unmodeled uncertainties associated with the flock dynamics and formation. On another side, the efficiencies of the various control designs rely on how quickly they can adapt to different dynamic situations in real-time. An online model-free policy iteration mechanism is developed here to guide a flock of agents to follow an independent command generator over a time-varying graph topology. The strength of connectivity between any two agents or the graph edge weight is decided using a position adjacency dependent function. An online recursive least squares approach is adopted to tune the guidance strategies without knowing the dynamics of the agents or those of the command generator. It is compared with another reinforcement learning approach from the literature which is based on a value iteration technique. The simulation results of the policy iteration mechanism revealed fast learning and convergence behaviors with less computational effort.
A Data-Driven Model-Reference Adaptive Control Approach Based on Reinforcement Learning
(2021) Abouheaf, Mohammed; Gueaieb, Wail; Spinello, Davide; Al-Sharhan, Salah
Model-reference adaptive systems refer to a consortium of techniques that guide plants to track desired reference trajectories. Approaches based on theories like Lyapunov, sliding surfaces, and backstepping are typically employed to advise adaptive control strategies. The resulting solutions are often challenged by the complexity of the reference model and those of the derived control strategies. Additionally, the explicit dependence of the control strategies on the process dynamics and reference dynamical models may contribute in degrading their efficiency in the face of uncertain or unknown dynamics. A model-reference adaptive solution is developed here for autonomous systems where it solves the Hamilton-Jacobi-Bellman equation of an error-based structure. The proposed approach describes the process with an integral temporal difference equation and solves it using an integral reinforcement learning mechanism. This is done in real-time without knowing or employing the dynamics of either the process or reference model in the control strategies. A class of aircraft is adopted to validate the proposed technique.
An Adaptive Fuzzy Reinforcement Learning Cooperative Approach for the Autonomous Control of Flock Systems
(2021) Qu, Shuzheng; Abouheaf, Mohammed; Gueaieb, Wail; Spinello, Davide
The flock-guidance problem enjoys a challenging structure where multiple optimization objectives are solved simultaneously. This usually necessitates different control approaches to tackle various objectives, such as guidance, collision avoidance, and cohesion. The guidance schemes, in particular, have long suffered from complex tracking-error dynamics. Furthermore, techniques that are based on linear feedback strategies obtained at equilibrium conditions either may not hold or degrade when applied to uncertain dynamic environments. Pre-tuned fuzzy inference architectures lack robustness under such unmodeled conditions. This work introduces an adaptive distributed technique for the autonomous control of flock systems. Its relatively flexible structure is based on online fuzzy reinforcement learning schemes which simultaneously target a number of objectives; namely, following a leader, avoiding collision, and reaching a flock velocity consensus. In addition to its resilience in the face of dynamic disturbances, the algorithm does not require more than the agent position as a feedback signal. The effectiveness of the proposed method is validated with two simulation scenarios and benchmarked against a similar technique from the literature.
Real-Time Measurement-Driven Reinforcement Learning Control Approach for Uncertain Nonlinear Systems
(2023) Abouheaf, Mohamed; Boase, Derek; Gueaieb, Wail; Spinello, Davide; Al-Sharhan, Salah
The paper introduces an interactive machine learning mechanism to process the measurements of an uncertain, nonlinear dynamic process and hence advise an actuation strategy in real-time. For concept demonstration, a trajectory-following optimization problem of a Kinova robotic arm is solved using an integral reinforcement learning approach with guaranteed stability for slowly varying dynamics. The solution is implemented using a model-free value iteration process to solve the integral temporal difference equations of the problem. The performance of the proposed technique is benchmarked against that of another model-free high-order approach and is validated for dynamic payload and disturbances. Unlike its benchmark, the proposed adaptive strategy is capable of handling extreme process variations. This is experimentally demonstrated by introducing static and time-varying payloads close to the rated maximum payload capacity of the manipulator arm. The comparison algorithm exhibited up to a seven-fold percent overshoot compared to the proposed integral reinforcement learning solution. The robustness of the algorithm is further validated by disturbing the real-time adapted strategy gains with a white noise of a standard deviation as high as 5%.
Probabilistic description of short-term cloud dynamics from rapid sampling of the solar spectral irradiance
(2022) Anderson, Nick; Tatsiankou, Viktar; Hinzer, Karin; Beal, Richard M.; Schriemer, Henry P.
Solar irradiance variability due to stochastic cloud dynamics can cause unwanted fluctuations in the output voltage of photovoltaic (PV) modules. These dynamics must in particular be understood at very-short and short time scales if grid interconnection and generation/load balance requirements are to be maintained for PV distributed across the grid edge. Using a recently-created database for Ottawa, Canada, a 6-month longitudinal study was conducted with a specific focus on cloud dynamics. A spectral pyranometer was used to derive full-range spectral and broadband global horizontal irradiance under all sky conditions every 250 ms. Exploiting the infrared (IR) measurement channel of this software augmented multi-filter radiometer allowed the cloud dynamics to be probed across time scales ranging from the subsecond to ~30 minutes. Seven distinct sky conditions were self-consistently determined without sky imaging. Probability distributions, established via kernel density estimates (KDE), allowed the statistical dependence of these conditions on the spectral clear-sky index to be found. The stochastic nature of the spectral irradiance variability was probed using spectral clear-sky index increments, over time steps that were found to span three distinct variability regimes.
Height distributions of uncapped InAs/InGaAsP/InP quantum dashes and their effect on emission wavelengths
(2022) Obhi, Ras-Jeevan K.; Schaefer, Sebastian W.; Valdivia, Christopher E.; Poole, Philip J.; Liu, Jiaren; Lu, Zhenguo; Hinzer, Karin
The emission wavelength of self-assembled quantum dashes can be controlled by their height. Uncapped InAs/InGaAsP/InP quantum dashes are found to have two distinct heights, which we have measured with atomic force microscopy and denoted the plateau and the peak heights. These heights range from 0.50 nm to 2.35 nm. Under the same growth conditions, for increasing uncapped quantum dash heights we observe an increase in the photoluminescence peak emission wavelength from approximately 1535 to 1543 nm for the capped layers. A growth temperature of 520°C is determined to achieve uniform height distribution for 1550 nm emission using chemical beam epitaxy.

Browse

Recent Submissions