Distracted driving is a major factor in road accidents and fatalities, particularly due to the widespread use of mobile phones. This study aims to detect visual driving distractions using AI models, focusing on feature extraction and machine learning techniques to create an accurate detection method. Twenty-six participants took part in a simulated driving experiment using Euro Truck Simulator to create realistic driving scenarios. Participants completed two distinct scenarios: one involving a standard GPS-guided route and another while reading from a handheld smartphone to simulate distraction. Data was collected from both the simulator and an eye-tracking device, capturing critical metrics such as vehicle speed, steering angle, pedal usage, gaze direction, and head movement. The collected data was preprocessed by removing irrelevant timestamps, scaling values, and synchronizing the timestamps to a frequency of 20 Hz to ensure consistency. From this processed data, key features such as mean, standard deviation, minimum, maximum, and median values were extracted at 1 Hz intervals to provide a robust statistical summary of each signal. Several machine learning models were applied, with the Gaussian Process Classifier demonstrating the best performance, achieving an accuracy of 0.976 and an AUC of 0.996. These results highlight the potential of AI-driven systems to significantly enhance driving safety by accurately identifying distracted driving behaviors, paving the way for real-time driver monitoring solutions.
The systems engineering and robotics communities evolve in parallel, although they share common concepts and related issues. This article proposes a state-of-the-art of publications tackling both Systems Engineering (SE) and Robotics, to identify added value of Systems Engineering for Robotics, and lessons learnt for Systems Engineering from robotics use cases. In the framework of the Systems Engineering community, this synthesis of works from academics, industrial and government agencies should ease the building of SE-based seamless integrated framework for architecture, design, verification and validation of heterogeneous swarms of collaborating unmanned ground vehicles and unmanned aerial vehicles.
The attempt to backdoor SSH via an infiltration operation in one of its dependencies, namely the XZ library, illustrated the importance of robust software supply chain security. One of the tools to achieve this objective is software bill of materials (SBOM). In this paper, we proposed enriching data from our internal SBOM tooling with maintenance-related metadata from GitHub and Node Package Manager (npm) repositories to get a better understanding of our supply chain, regarding its susceptibility to XZ-type infiltration. We defined several features that we believed contributed to this risk and calculated a risk metric to identify the most extreme cases to test the validity of our approach. In doing so, we did not identify any case of supply chain attacks but identified libraries that could be susceptible to it. In addition, during the data analysis we encountered several cases where developers were making seemingly irrational tradeoffs between productivity gains and technical debt, including cyber security risks.
Bangladesh, grappling with high population density and severe traffic congestion, has witnessed a surge in air travel demand due to inadequate transportation infrastructure. This trend has propelled the growth of the domestic aviation industry, with several private companies and a government-owned entity operating domestic flights. However, monopolistic tendencies and barriers to entry pose challenges for new market entrants. In response, this paper presents a comprehensive analysis of Bangladesh's domestic aviation sector, focusing on comprehensive analysis of growth opportunities. Over a four-month period, our study constructs a comprehensive dataset from diverse sources, including flight data, operational expenditures, passenger surveys, and interviews, to evaluate the scope and growth potential within the domestic aviation market. Leveraging machine learning algorithms and novel uncertainty quantification techniques like prediction intervals, we investigate the profitability of flights on different routes based on operational costs and ticket fares. Prediction intervals perform better than point predictions in this case, as they offer a range of likely outcomes, providing more valuable insights into potential profit margins, offering valuable guidance for new entrants seeking opportunities in the market. The dataset's trustworthiness is ensured through rigorous validation, with a trustworthiness score of 83.5 %. Additionally, the dataset is available upon request to facilitate further research and collaboration.
A network created with Low Earth Orbit (LEO) satellites has advantages, such as lower latencies and network equity to remote Internet regions. However, the inherent mobility of a LEO satellite network sets it apart from its terrestrial counterpart. Studying how LEO networks are structured and interconnected is essential for understanding the performance and efficiency of LEO network topologies. This paper explores the realm of LEO satellite networks created using Inter Satellite Links (ISL). We simulated three topologies: Sparse, +Grid, and Extended +Grid. We evaluated their performance on a major LEO satellite constellation and recorded key network characteristics. We assess the simplicity, hop count, latency, and stability (the count of total path changes over a 10-minute interval) of these connection strategies in simulated LEO networks. Our evaluations highlight that although the Sparse topology is simpler than the other two, it has a higher latency than the +Grid topology. Our results indicate that using an Extended +Grid topology over a +Grid topology can reduce hop count down to 50% in a LEO mega-constellation while using the same amount of ISL hardware. Although Sparse Topology shows similar hop count and stability results, its additional latency makes it less desirable to +Grid and Extended +Grid topologies. The trade-offs between these attributes and performance and reliability across diverse topologies in LEO satellite networks need further research.
Condition-based maintenance has progressed substantially through the integration of IoT and advanced sensing technologies. Prognostics and Health Management (PHM) has emerged as a pivotal strategy for enhancing reliability and minimizing downtime by facilitating early fault detection and optimizing maintenance planning. Predicting the Remaining Useful Life (RUL) is a vital component of prognostics and health management for machinery. It effectively prevents catastrophic failures and reduces the costs associated with unplanned maintenance. Numerous Deep Learning (DL) methodologies have been applied for predicting RUL owing to the versatility of their architectures. This paper investigates the use of Graph Neural Networks (GNNs) for RUL estimation. GNNs are adept at modeling complex relationships within equipment data by representing components and their connections as nodes and edges in a graph, thereby capturing intricate dependencies and operational patterns. Building on recent advancements, this study conducts a comparative analysis of various methods for constructing adjacency matrices used in GNNs, crucial for accurate RUL estimation. The study leverages the Graph Attention Model (GAT) to assess the effectiveness of different graph structures in predicting RUL, especially under conditions of machinery degradation. Our findings emphasize the importance of edge quality and quantity in determining the performance of GAT models. This paper concludes with an in-depth discussion on the impact of various graph construction methods on RUL estimation, providing insights for future research and practical applications in industrial settings.
This paper shares experience in teaching Systems Engineering (SE) and Model-Based Systems Engineering (MBSE) at ISAE-SUPAERO, college of aerospace engineering at the University of Toulouse, France. Fundamental to the development of complex systems, SE has its place in different educational programs, at different levels, and with different approaches. Starting from the mission statement of a development, passing via requirements engineering, operational analysis, functional analysis, logical and physical architecture design, verification and validation, the production phase, the use phase, and finally the retirement phase, SE is explained to the students. At each step, the authors of the paper give insight in the results of the teaching of these topics, the challenges and the lessons learned. Having transitioned from document-centric approaches to model-based approaches, teaching MBSE has become important. After all, students that graduate from top universities are expected to know how MBSE works, how to use some tools so that they can easily adopt other tools in the work life, and to use the tools to the fullest advantage including formal verification of designs. The authors of the paper share their experience in teaching MBSE in the aerospace engineering programs at Master's level. MBSE is viewed in terms of languages, tools and methods. The paper focuses discussion on SysML, its support by a free tool that offers simulation and formal verification capabilities, and a method that applies to a broad variety of systems. Teaching MBSE cannot be dissociated from teaching SE. Experience in teaching MBSE is discussed in the paper and links with classes on formal methods are also identified as far as formal verification of SysML models is concerned.
Research on the design of autonomous robot swarms has explored many directions over the past ten years (multi-robot systems, coordination protocols, biomimicry, centralized or distributed architecture) but has struggled to combine these different axes in a holistic design approach. The junction between autonomous robot swarm engineering and systems of systems engineering must be pushed further. The first step is to evaluate the contribution of systems of systems architecture to the design of autonomous robot swarms. We propose two initial tests of the relevance of this approach: do the methods and languages of systems of systems architecture allow us to describe the dynamics of these autonomous robot swarms? Do the concepts of system of systems architecture allow us to establish a taxonomy of autonomous robot swarms? This approach should also allow us to test the current limits of the standards of system of systems architecture: taxonomies too focused on the governance of systems of systems, and coexistence without unification of numerous static architecture methods (Model-Based Systems Engineering, Architecture Frameworks, Hypergraphs, etc.) and dynamic architecture methods (Dynamical Systems, System Dynamics, State Machine Diagrams, Event-B modeling, Agent-Based Modeling, BPMN, Activity Diagrams, etc.)
Platform level details are critical for product development in the AI space, with solutions coming in from multiple technology vendors. It is necessary to have specialized debug features to extract key platform information. This paper studies the current capabilities and challenges in AI solutions NPI (New Product Introduction) development and fleet debug. For NPI, the paper talks about gaps in PCIe debugging, RAS tool usage and vendor diagnostic tools. Debug capability gaps for in-band and out of band are highlighted from a fleet debug perspective. To address these gaps the paper highlights using OCP Compliance tools and OCP compliant debug hardware for short term while pushing for long term solutions to address debug standardization for the AI products
The optimal assignment of Information Technology (IT) tickets is a critical solution in our ever-expanding digital age. To maximize efficiency and minimize the burden on IT staff, this article proposes a novel approach: the Stacked Role Assignment (GRAABD-WS). GRAABD-WS leverages the principles of E-CARGO, an object-oriented framework designed to solve complex systems. The core concept involves mapping the relationship between tickets and technicians, described as roles and agents, respectively. We evaluate the suitability of technicians for specific tickets by assessing their individual workloads and skill sets, resulting in an “Agent Fitness” value. This value varies based on the roles required by each ticket. By ranking technicians based on their abilities and current workloads, GRAABD-WS ensures that the most suitable technicians are assigned to each task. The effectiveness of this algorithm is demonstrated by its ability to optimally assign 11 agents to 40 tickets within 3.71 seconds. The results highlight the potential of the Stacked Role Assignment algorithm to enhance IT ticket management, providing a robust solution for dynamic and complex IT environments.
Freezing of Gait (FoG) is a debilitating symptom affecting over 50% of Parkinson's disease (PD) patients, significantly increasing the risk of falls and impacting their quality of life. This paper presents a novel, scalable approach to FoG detection using machine learning models and wearable sensors, designed for user-independent operation on resource-constrained embedded platforms. We leverage the DAPHNet dataset, which includes accelerometer data from PD patients, to extract comprehensive time and frequency domain features. Various machine learning classifiers, including resource-efficient models like ProtoNN, are trained and evaluated for their performance in detecting FoG episodes. Our results demonstrate that ProtoNN achieves robust and consistent performance, particularly with larger window sizes and diverse patient data. The findings underscore the importance of model selection, feature extraction and training strategies in developing practical FoG detection systems. This research contributes to the advancement of real-time, portable monitoring solutions, paving the way for improved management of FoG in PD patients.
This paper introduces a fuzzy set-theoretic approach to environment modeling as an alternative to conventional Occupancy Grid Mapping (OGM). The proposed method generalizes the crisp definitions of map cells and sensor measurements by employing fuzzy sets. Consequently, it utilizes fuzzy inference with interpretable rules instead of binary Bayesian inference to determine occupancy states from sensory data. As a result, the fuzzy mapping technique addresses uncertainty while eliminating assumptions about data distribution or cell occupancy probabilities. A notable advantage of the proposed Fuzzy Occupancy Grid Mapping (FOGM) is its ability to evaluate the overall occupancy states of sub-regions of the environment in a single query, unlike traditional methods that require point-by-point assessments. While this advantage introduces increased computational complexity, the mapping process is designed to ensure real-time operation. Another prominent feature of the FOGM is that it creates continuous maps with smooth occupancy variations as compared to the discrete binary maps the traditional OGM generates. The performance of the proposed technique is validated through experiments involving a mobile robot localized via a motion capture system. The robot collects range data in real-time using a LiDAR sensor to be used for mapping. Results demonstrate that the fuzzy mapping approach effectively and accurately represents the environment while maintaining computational efficiency. Its capability in parallel execution on multiple processing cores further enhances real-time performance.
The Spike Neural Network (SNN) is a neural network that simulates the transmission of signals in brain cells. It utilizes the low-energy characteristics of neurons transmitting discrete signals, which have attracted widespread attention. This paper proposes an SNN with the threshold voltage drop to optimize the SNN. The model focuses on controlling the threshold voltage to reduce the effect of residual membrane voltage in the Mean Square Error (MSE). Reduced threshold voltage stimulates neurons to fire more spikes. The initialized membrane voltage SNN (I-SNN) is an optimized method that adding an extra membrane voltage in neurons. This paper optimizes the I-SNN to process negative data, introducing a variant called the signed initialized membrane voltage SNN (SI-SNN). The threshold voltage drop SNN and the SI-SNN in different numbers of neurons are compared for MSE in predicting the Lorenz System. The results indicate that the SNN with threshold voltage drop effectively reduces MSE, especially in the model with fewer neurons. After confirming the feasibility of this algorithm, this paper implements the threshold voltage drop method in FPGA. The MSEs of no optimized SNN and threshold voltage drop in FPGA are 0.0015 and 0.0013, respectively. The threshold voltage drop SNN demonstrates satisfactory performance.
Model-Based Systems Engineering leverages inter-connected models to improve Systems Engineering practice. Enabling technologies for MBSE often leave the details of engineering processes implicit. Digital threads are being codified to link authoritative sources of information across a digital modeling environment. When considered as workflows, digital threads make explicit the details of MBSE processes. As systems in their own right, digital threads could benefit from the same processes and analyses used in MBSE. In this paper, we refine (1) the notion of digital threads as workflows and (2) the environment necessary to execute them. Digital threads are built by composing atomic capabilities that perform specific authoring, verification, analysis, and reporting operations and represent the automatable part of the digital engineering workflow. We derive requirements for modeling digital threads describing engineering process workflows built via composition of atomic capabilities, as well as for modeling the execution environment. We illustrate how to model digital threads and atomic capabilities using SysMLv2 action definitions. Finally we describe our interpreter for executing Sys$M$ Lv2-based digital threads for AADL models. Combined, these contributions (1) allow systems engineer to build digital threads within their modeling environment, reducing the barrier towards defining tailored digital threads, and (2) move towards enabling modeling the process of modeling.
In today's rapidly evolving space environment, Space Domain Awareness (SDA) has become pivotal for ensuring the safety and security of space operations. With the increasing number of satellites and growing concerns over adversarial behaviors, timely detection of maneuvers is crucial for ensuring strategic advantage in space. This study assesses the feasibility of electro-optical (EO) ground sensors in detecting pre-maneuver photometric changes to use as leading indicators of satellite maneuvers. By leveraging real-world geosynchronous satellite data, we explore whether measurable changes in photometric characteristics could serve as early warnings for orbital maneu-vers. Our study sought to investigate Bayesian online change point detection as a statistical method for real-time anomaly detection. Our findings reveal that while EO sensors can detect abrupt photometric changes, there are a number of challenges that limit their predictive potential, including observation gaps, differences between propulsion systems, and confounding factors. The increasing economic and strategic importance of Earth's most congested orbits highlights the need for advanced automated SDA techniques. By evaluating the potential for EO sensors to provide early indications of satellite maneuver, this study supports the development of faster and more accurate satellite maneuver detection systems.
The rapid evolution of Artificial Intelligence (AI) requires a rethinking of practices and better anticipation of impacts, as highlighted by several new regulations. Different communities, including engineers, auditors, business leaders, standardization experts, and ethicists, are developing AI evaluation frameworks with scoring systems to progressively operationalize good practices for AI. The development of AI evaluation frameworks should be closely tied to disciplines such as systems and software engineering, requirements engineering, quality management, risk analysis, verification and validation, and decision support. This paper formalizes the development of AI evaluation frameworks as six non-linear activities to better leverage these disciplines. It is intended to be of particular interest to communities engaged in AI evaluation but not familiar with systems engineering. It draws on the authors' experience in defining end-to-end methodologies for AI trustworthiness and designing evaluation frameworks. The activities presented are: (1) Managing the evaluation framework as a product in its own right; (2) Managing the evaluation framework as a generic specification for different objects; (3) Formalizing progressive good practices, possibly in the form of requirement levels; (4) Clustering these requirements into understandable dimensions; (5) Defining a global scoring scheme; and (6) Tailoring the framework to account for industry-specific considerations and risks. This formalization aims to enhance the effectiveness of AI evaluation frameworks.
Efficient planning for sustainable growth is an imperative requirement for every country of the world, especially the so-called developing countries. But the most formidable challenge involved in adopting the desired developmental process lies in grappling with the complexity of the socio-economic systems. Given its size, varied socio-cultural sub-systems, great disparity in the distribution of income, varying degrees of sophistication in agricultural and industrial practices, and such other factors, the complexity involved in developmental planning in the Indian context is of a very high order. However, Soft Systems Approach makes it possible to visualize a consensus-based mental model as a digraph, depicting the structural connections between relevant elements of the system in a contextual relationship. The present study has exploited this approach to structure the potent elements of the public investment system of India through Interpretive Structural Modelling (ISM) method, which is a computer-based technique that enables groups and individuals in visualizing the relationships between system elements and developing graphical representations of complex systems. Prior to the ISM, the system elements of the public investment system of India were arrived at through Nominal Group Technique that involved participation of 15 domain experts. The ISM exercise with the domain experts resulted in a systems model (digraph) of 37 elements, which was also subjected to MICMAC analysis. The major contribution of the study is that it gives a bird's-eye view of the public investment system of India to assist the policy makers in focusing on the required inputs at the desired nodes of the digraph, for optimizing the output of resources in achieving the desired developmental goals.
Industry 5.0, with its human-centric vision, emphasizes seamless interaction between humans and machines, paving the way for enhanced user experiences in various sectors, including e-commerce. This study investigates the impact of human-centered design (HCD) on fostering trust and satisfaction in e-commerce platforms, addressing a notable gap in the literature. The research identifies seven critical aspects of HCD: intuitive interfaces, user involvement, transparency, personalization, security, accessibility, and aesthetics. A mixed-method approach was employed, starting with content analysis to identify relevant factors and select two e-commerce platforms, one with high HCD quality and one with low HCD quality. A total of 100 participants performed shopping tasks on these platforms, followed by surveys measuring perceived trust, satisfaction, and purchase intentions. Statistical analysis revealed that the high-HCD platform significantly outperformed the low-HCD platform in terms of clarity, perceived security, transaction transparency, ease of use, overall satisfaction, and repurchase likelihood. The findings highlight the transformative potential of integrating HCD principles with advanced technologies in e-commerce to align with Industry 5.0's human-centric goals. By enhancing user trust and satisfaction, HCD contributes to improved user loyalty, better business outcomes, and sustainable growth in digital commerce. This study provides valuable insights for designing e-commerce platforms that resonate with the ethical, social, and technological principles of Industry 5.0.
Large Language Models (LLMs) often face challenges in generating accurate and reliable information, particularly in knowledge-intensive tasks. This limitation, referred to as hallucination, occurs when models produce content that is incorrect, irrelevant, or unsupported by evidence. Retrieval Augmented Generation (RAG) solutions provide a promising approach by integrating relevant external knowledge, enabling models to generate factually grounded responses. This study evaluates the performance of a base LLM model, a fine-tuned DistilBERT model, and two RAG architectures, Naive RAG and Graph RAG, to study their impact on reducing hallucinations and enhancing contextual understanding. Using subsets of HaluEval, Squad-V2, and TriviaQA benchmark datasets, the base model achieved accuracies of 10.18%, 12.67%, and 5.46% respectively; Naive RAG resulted in 44.56%, 19.04%, and 35.32% accuracies; while the fine-tuned LLM model's accuracies were 72.5%, 72.31%, and 88.7% respectively. Graph RAG resulted in 8.85% and 15.12% accuracies using Squad-V2 and TriviaQA, respectively. Our findings show that while fine-tuned LLMs outperform baseline models, incorporating RAG solutions did not result in significant performance improvements, suggesting that the incorporation of external knowledge may not always align with the needs of the task. Experiments demonstrate that Graph RAG handles complex queries by leveraging relationships within structured knowledge graphs. Data organized as a knowledge graph may enable Graph RAG solutions reach their full potential by utilizing their capacity to efficiently retrieve contextually relevant information. Although computing complexity remains a restriction, this study shows that RAG topologies might not consistently enhance LLM reliability in practical situations.
This paper presents the Argumentative Rule-based Explanatory Framework (AREF), a novel methodology in Explainable Artificial Intelligence (XAI) aimed at delivering comprehensible and transparent explanations for predictions and classifications made by machine learning models. AREF combines elements of argumentation theory with a rule-based system to elucidates the reasoning behind machine learning outcomes, offering a transparent view into the otherwise opaque processes of these models. The core of AREF is the innovative use of the apriori algorithm to extract rules from datasets, which serve as the basis for constructing arguments. Additionally, AREF introduces a distinctive approach to developing attack relationships among arguments, facilitating a comprehensive argumentation structure. To demonstrate the efficacy and adaptability of AREF, the framework is applied to seven different machine learning algorithms across three datasets: a basic boolean dataset for foundational demonstrations, and the well-known iris and breast cancer datasets for more complex scenarios. The results highlight the capability of AREF to provide clear, rule-based explanations across diverse machine learning models.
