IFIP Open Access Digital Library
Not a member yet
22614 research outputs found
Sort by
The Innovation Challenge Bootcamp Model
No full textPart 4: Collaborative Ecosystems: Skills for Resilient FuturesInternational audienceEducation has responded to industrial demands, to new technological developments and capacities required for the application of concepts to generate models that describe reality. Traditional educational models where the student acts as a recipient of knowledge have been surpassed by the demands of the new automated and connected industrial currents. These educational currents, despite taking advantage of information resources, active learning techniques and team collaboration, are limited by other aspects that hinder their dissemination. Thus, the role of the educator, spaces required, teamwork and available resources must be combined in order to follow the pace of technology adoption and thus cover the new educational paradigms. This article explores an educational model based on challenges that has been implemented with researchers, lecturers, and university students. This model has had a positive impact on the participants in project development and entrepreneurship skills
Towards a Reference Architecture for Collaborative Decision Support Systems for Natural Disasters Management
No full textPart 5: Collaborative Ecosystems: Technologies for Resilient FuturesInternational audienceNatural disasters (ND) have become increasingly frequent, causing impactful social, economic, and environmental consequences. Civil defenses and related actors are the ones in charge of handling ND throughout their life cycles regarding their different types and intensities. There are plenty of issues that should be considered when facing a ND, including very complex analyses and collaborative decision-making. Studies show an increasing use of decision support systems (DSS) to assist civil defenses in ND. However, it was observed that the numerous DSS for ND cover only isolated parts of the problem, making it difficult to design a complete, scalable, and globally coherent roadmap for civil defenses and software developers. This article proposes a reference architecture aiming at acting as a blueprint for derivation of particular ND DSS. A software prototype has been developed to show a derived DSS based on a civil defense organization. Conclusions and next steps of this work are presented at the end
Design of a Collaborative Network for Mapping Digital Skills for Industry 5.0
No full textPart 4: Collaborative Ecosystems: Skills for Resilient FuturesInternational audienceThis paper outlines a conceptual model of a Future Skills Collaborative Network (FSCN), a multi-stakeholder initiative designed to address the widening digital skill gaps in industry. By utilizing models to map required digital skills—from basic to specialized proficiencies—the FSCN aims to identify and potentiate training pathways that align with industrial needs. Utilizing open data for pilot analyses, this study demonstrates the practical application of the FSCN in identifying trends and relationships between digital skills, job requirements, and technological advancements. The findings underscore foreseen digital skills and the benefits of the FSCN’s potential to significantly address the challenges presented by the digital transformation in Industry 5.0
Utilizing Natural Language Processing for Enhancing Collaborative Value-Driven Design of Smart Product Service System: Smart E-Vehicle Application
No full textPart 7: Collaborative Networks as Driver of Innovation in Organizations 5.0: ModelsInternational audienceManufacturing companies are increasingly transitioning from a product-centric to a smart Product Service System (smart PSS) approach to enhance customer satisfaction, service offerings, and product competitiveness through a combination of usage scenarios and digital components. In the context of Industry 5.0 transformation such as developing the Smart Electric Vehicle (SEV), the automotive industry faces the challenge of understanding customers’ descriptions of usage scenarios and translating the qualitative aspects of these scenarios into quantitatively assessed product features for collaborative value co-creation in smart PSS design. This paper addresses this challenge through utilizing Natural Language Processing (NLP) joint with Value-Driven Design (VDD) method for successfully supported a collaborative value exploration of in the smart PSS design stage. A case study was collaborated with a global automotive Original Equipment Manufacturer (OEM), Volkswagen, through proposing a NLP BERT model for VDD of Smart Electric Vehicle (SEV) design. Validation activities were performed by deploying the developed BERT model to the case company based on the scenario design of new car models
Multivariate Time-Series Methods with Uncertainty Estimation for Correcting Physics-Based Model: Comparisons and Generalization for Industrial Drilling Process
No full textPart 1: Deep LearningInternational audienceThis study employs a hybrid methodology, integrating data-driven and physics-based models to refine the latter’s predictions. Referred to as Hybrid Analysis and Modeling (HAM), this approach combines a physics-based model, solving multi-phase flow equations for cuttings transport, with advanced machine learning models to enhance predictive accuracy in hole cleaning operations. Previous research demonstrated two HAM methodologies (an intrusive and a non-intrusive approach) for uni-variate time-series data, improving predictions of the physics-based model. In this study, we develop multi-variate approaches with uncertainty estimation, comparing four machine learning models, ranging from simple linear methods to the advanced non-linear methods, i.e.: ARIMAX, XGBoost, Transformer, and Long-Short Term Memory (LSTM) models. Uncertainty estimates are also plotted to elucidate each model’s capacity to refine the physics-based model, accounting for epistemic uncertainty arising from knowledge gaps in the machine learning model and/or aleatoric uncertainty inherent in the data. Applied to an industrial drilling process, the hybrid approach facilitates accurate prediction of a key variable, equivalent circulating density, essential for process monitoring. In the current study, the LSTM model outperforms others by avoiding overfitting on unseen test datasets. This work illustrates the potential of the presented hybrid methodology to generalize and enhance predictions across all depths and time-steps during drilling operations, contingent upon the availability of more measurement datasets for training/testing. Thus, HAM methodology holds promise for refining physics-based models in various process industry operations for correcting physics-based models in other process industry operations
Pollutant Concentration Prediction by Random Forest to Estimate a Contaminant Source Position
No full textInternational audienceThe goal of Source Term Estimation (STE) is to accurately identify the parameters that describe the source of a release, namely the position and strength . This requires a reliable dispersion model. Recent advancements have integrated machine learning with the Gaussian dispersion model, enhancing the prediction of pollutant concentrations while mitigating the impact of complex terrains. However, pollutant dispersion varies significantly across different atmospheric stability classes. Addressing this, we introduce a novel strategy termed the Multiple Learning Model (MLM), which segments predictions based on stability classes: Neutral (), Unstable (), and Stable (). This approach, by building models to specific atmospheric conditions, promises more precise source estimations. In comparative study, MLM not only refined prediction accuracy from 0.04 to 0.06 but also improved source location estimates, narrowing the discrepancy to 7 m–25 m from the actual source, a marked improvement over traditional random forest models. This methodological advancement underscores the potential of stability-class-specific models in enhancing the accuracy and reliability of pollutant source estimations
Benign Paroxysmal Positional Vertigo Disorders Classification Using Eye Tracking Data
No full textPart 2: Machine LearningInternational audienceNystagmus is a neurological condition characterized by involuntary and rhythmic eye movements. These abnormal eye movements can be indicative of various underlying neurological and vestibular disorders, impacting visual stability and affecting an individual’s perception of their surroundings. Benign Paroxysmal Positional Vertigo (BPPV) is a special case of nystagmus where brief episodes of dizziness are triggered by specific head movements. However, the accurate diagnosis of BPPV still heavily relies on the precise interpretation of nystagmus induced by positional tests, which often require specialized expertise. In this paper, we developed an AI framework that detects and tracks the movement of pupil central and dilation overtime. These time-series data are then classified into different types of nystagmus. In contrast to classical image processing approaches, we employ convolutional neural networks as a baseline and use the averaging of numerous models without incurring extra inference or memory costs. The results from experiments demonstrate that when given the patient’s eye video data, our framework is capable of classifying the specific BPPV disorder out of six possible types with an average accuracy of 91% on the publically available challenging and unbalanced dataset
Detecting Illicit Data Leaks on Android Smartphones Using an Artificial Intelligence Models
No full textPart 2: Machine LearningInternational audienceIn today’s digital landscape, hackers and espionage agents are increasingly targeting Android, the world’s most prevalent mobile operating system. We introduce DeepDetector - a system based on artificial intelligence to recognize data thefts in Android. This model is based upon a large dataset comprising of clean and tainted network traffic trained using a Random Forest Classifier. DeepDetector scores high in two main areas as it achieves 82.9% accuracy for connection anomaly detection and 89.9% recall in connection anomaly detection whereas it gets 78.9% accuracy and 81.6 recall in terms of detection of under the system mounted with Raspberry Pi, automatic data collection, preparing of a dataset, training and testing of the model, as well as leak detection are ensured. In this regard, DeepDetector offers a viable way of enhancing Android user security
Online Reinforcement Learning for Designing Automotive Hybrid Assembly Sequence: A Task Clustering-Guided Approach
No full textPart 1: Reinforcement/Natural LanguageInternational audienceThis paper proposes a novel approach that integrates hierarchical clustering (HC) into reinforcement learning algorithms to address the simultaneous resolution of hybrid assembly line balancing (ALB-1) and assembly sequence planning (ASP). The proposed approach attempts to capture implicit constraints, derived from accumulated experiences and industry-specific knowledge, enhancing the adaptability of solutions. The inclusion of the clustering algorithm enhances the decision-making process of the reinforcement learning agent through the introduction of a problem-specific similarity reward. To evaluate the effectiveness of the approach, three adapted methods are implemented and tested: QL-HC, SARSA-HC, and SARSA without HC. The experimental results demonstrate the superior performance of our novel approach compared to traditional techniques, with SARSA-HC exhibiting particularly impressive results with 82% of similarity to the expert’s sequence. This approach offers increased flexibility, adaptability, and the capacity to incorporate expert knowledge and lessons learned from prior assembly line design experiences and proves to be a valuable tool for enhancing efficiency and reducing time-to-market in manufacturing settings
AI-Driven Sentiment Trend Analysis: Enhancing Topic Modeling Interpretation with ChatGPT
No full textPart 1: Reinforcement/Natural LanguageInternational audienceUnderstanding the sentiment trends of large and unstructured text corpora is essential for various applications. Despite extensive application of sentiment analysis and topic modeling, extracting meaningful insights from the vast amount of textual data generated on social media platforms presents unique challenges due to the short and noisy nature of the text. In this study, we propose a methodology for analyzing sentiment trends in social media, including data collection, data preprocessing, sentiment analysis, social network graph construction, and topic modeling interpretation using ChatGPT. By integrating ChatGPT with topic modeling techniques such as LDA and BERTopic, we aim to enhance the interpretability of sentiment-related topics and gain deeper insights into sentiment trends in social media conversations. Through a case study focusing on parental hesitancy toward child vaccination, we illustrate the applicability and utility of our proposed methodology in real-world social media analysis scenarios, demonstrating its effectiveness in topic modeling interpretation and enhancing understanding of social media discourse. The integration of ChatGPT and BERTopic yielded improved topic interpretation for the short text of large corpus based on the coherence score of the original posts and generated description of the topic, ultimately reducing the cost and time required for topic interpretation by humans