Search

Nowadays data is acquiring an indisputable importance in every field including engineering. In the past, experimental data was used to calibrate state-of-the art models. Once the model was optimally calibrated, numerical simulations were run. However, data can offer much more, playing a more important role than calibration or statistical analysis in the modeling/simulation process. Indeed, today data is gathered and used to train models able to replace complex engineering systems. The more and better the training data, the more accurate the model is. However, in engineering experimental data use to be the best data but also the most expensive in time and computing effort. Therefore, numerical simulations, cheaper and faster, are used instead but, even if they are closed to reality, they always present an error related to the ignorance of the engineer over the complex real system. It seems thus coherent to take advantage of each approach. This leads to the “hybrid twin” rationale. On the one hand, numerical simulations are computed as primary data source, assuming their inherent error. On the other hand, some experimental data is gathered to train a machine learning correction model which fills the prediction-measurement gap. However, learning this ignorance gap becomes difficult in some fields such as fluids dynamics, where a regression over the localized solutions can lead to non physical interpolated solutions. Therefore, the “hybrid twin” methodology proposed in this article relies on Optimal Transport theory, which provides a mathematical framework to measure distances between general objects and a completely different interpolation approach between functions.

Background. Marker-less systems based on digital video cameras and deep learning for gait analysis could have a deep impact in clinical routine. A recently developed system has shown promising results in terms of joint center position but has not been yet evaluated in terms of gait outcomes. Research question. How does this novel marker-less system compare to a marker-based reference system in terms of clinically relevant gait parameters? Methods. The deep learning method behind the developed marker-less system was trained on a dedicated dataset consisting of forty-one asymptomatic and pathological subjects each performing ten walking trials. The system could estimate the three-dimensional position of seventeen joint centers or keypoints (e.g., neck, shoulders, hip, knee, and ankles). We evaluated the marker-less system against a marker-based system in terms of differences in joint position (Euclidean distance), detection of gait events (e.g., heel strike and toe-off), spatiotemporal parameters (e.g., step length, time), kinematic parameters (e.g., hip and knee extension-flexion), and inter-trial reliability for kinematic parameters. Results. The marker-less system was able to estimate the three-dimensional position of joint centers with a mean difference of 13.1 mm (SD = 10.2 mm). 99% of the estimated gait events were estimated within 10 milliseconds of the corresponding reference values. Estimated spatiotemporal parameters showed zero bias. The mean and standard deviation of the differences of the estimated kinematic parameters varied by parameter (for example, the mean and standard deviation for knee extension flexion angle were -3.0° and 2.7°). Inter-trial reliability of the measured parameters was similar to that of the marker-based references. Significance. The developed marker-less system can measure the spatiotemporal parameters within the range of the minimum detectable changes obtained using the marker-based reference system. Moreover, except for hip extension flexion, the system showed promising results in terms of several kinematic parameters. Chaire BiomecAM

We build an original synthetic dataset of 2D mechanical designs alongside their mechanical and geometric constraints, GMCAD. Such a dataset allows training Deep Learning (DL) models for Design for Additive Manufacturing (DfAM) to incorporate and control Computer-Aided-Design (CAD) features with mechanical performance. Geometric AM constraints are often complex to describe, depending on applications, processes, materials. They often lack explicit mathematical descriptions, belong exclusively to the CAD world, and hardly can be integrated into mechanical design, hampering AM design freedom. DL models have recently emerged as a potential to reconcile both CAD and Computer Aided-Engineering (CAE) worlds. They derive data-driven geometric rules over mechanical designs, allowing fine-grained control over the geometry during the design phase, contrary to the conventional CAD-to-CAE sequential approach. DL models, however, need high-quality labeled data, and merging CAD features to CAE aspects is challenging as they rely on different formats, rules, and tools. GMCAD dataset solves this issue following these building steps. (i) Building a DL-mechanical conditions predictor from a dataset generated by a density-gradient-based Topology Optimization method (TO); an AM-synergetic design generation tool. (ii) Creating a CAD dataset inspired by the TO-based designs. (iii) Predicting the mechanical conditions of the CADs using the DL predictor of mechanical conditions. Last, we evaluate the mechanical performance of GMCAD’s designs and derive statistics over CAD and CAE features. Designs of GMCAD show the significant influence of minor geometric changes, explaining the intricate design task of conforming both with functionality and geometric constraints. Consequently, having GMCAD is advantageous to train DL models to generate designs accounting for all these constraints simultaneously, without the need for time-consuming trial and error techniques. Such models could enhance DfAM and go beyond AM; they can also enhance other challenging fields as CAD automatic reconstruction, reverse engineering, isogeometric design and paves the way to multi-objective controllable design generation. Expleo France

Despite the freedom Additive Manufacturing (AM) offers when manufacturing organic shapes, it still requires some geometrical criteria to avoid a part's collapse during printing. The most synergetic design approach to AM is Topology Optimization (TO), which finds an optimal free-form given mechanical constraints. However, it is hard for TO to integrate these layout geometry-related constraints and it seldom proposes printable shapes. Therefore, this work leverages the Deep Learning (DL) capability to handle spatial correlations within the mechanical design process by integrating the layout and mechanical constraints at the conceptual level. It proposes a DL-layout-driven solution (DL-TO) trained via a triple-discriminator Generative Adversarial Network (GAN) framework. The DL-TO's performance is demonstrated by generating mechanically valid 2D designs conforming with layout constraints in a fraction of a second. DL-TO's creativity is illustrated by its capability to generate designs with unseen input constraints (passive/active elements) and to propose several shapes for the same input mechanical constraints, a task that is hard for a traditional TO to achieve. Expleo France and the French National Association of Research and Technology (ANRT)

The growth of Information Technologies in industrial contexts have resulted in data proliferation. These data often underlines useful information which can be of great benefit when it comes to decision-making. Key Performance Indicators (KPIs) act simultaneously as triggers and drivers for decision-making. When they deviate from their targets, decisions must be rapidly and well made. Therefore, experts need to understand the underlying relationships between KPIs deviations and operating conditions. However, they often interpret deviations empirically, or by following methods that may be time consuming, or not exhaustive. This article proposes a generic neural networkbased approach for improving decision-making, by ensuring that decisions are consistent and made as early as possible. On the one hand, the proposal relies on improving KPIs deviations prediction, which is made possible thanks to the automatic construction of neural networks using neuro-evolution. On the other hand, the decision-making consistency is improved by identifying, among the operating conditions, contextual variables that most influence a given KPI of interest. This identification, which guide the decision-making process, is based on the analysis of the final weights of the neural network used for the KPI deviation prediction, given the contextual variables. bourse doctorale

Publication déjà téléversée dans HAL sous la référéence hal-03954342 ATONTE (ATlantis methodology for ONtology development from Texts and Experts) is a methodology for the manual development of low-level seed ontologies. The modelling process is based on a combination of knowledge from non-fiction text corpora such as manuals, information guides or sets of instructions, and the knowledge of domain experts. This article presents the five key steps of the ATONTE process. Seed ontologies created with ATONTE can be used to develop and populate knowledge graphs for use in specific applications within given technical domains.

International audience; Understanding how the brain responds to sensory inputs from non-invasive brain recordings like magnetoencephalography (MEG) can be particularly challenging: (i) the high-dimensional dynamics of mass neuronal activity are notoriously difficult to model, (ii) signals can greatly vary across subjects and trials and (iii) the relationship between these brain responses and the stimulus features is non-trivial. These challenges have led the community to develop a variety of preprocessing and analytical (almost exclusively linear) methods, each designed to tackle one of these issues. Instead, we propose to address these challenges through a specific end-to-end deep learning architecture, trained to predict the MEG responses of multiple subjects at once. We successfully test this approach on a large cohort of MEG recordings acquired during a one-hour reading task. Our Deep Recurrent Encoder (DRE) reliably predicts MEG responses to words with a three-fold improvement over classic linear methods. We further describe a simple variable importance analysis to investigate the MEG representations learnt by our model and recover the expected evoked responses to word length and word frequency. Last, we show that, contrary to linear encoders, our model captures modulations of the brain response in relation to baseline fluctuations in the alpha frequency band. The quantitative improvement of the present deep learning approach paves the way to a better characterization of the complex dynamics of brain activity from large MEG datasets.

International audience; Background: The deep learning-based human pose estimation methods, which can estimate joint centers position, have achieved promising results on the publicly available human pose datasets (e.g., Human3.6 M). However, these datasets may be less efficient for gait study, particularly for clinical applications, because of the limited number of subjects, their homogeneity (all asymptomatic adults), and the errors introduced by marker placement on subjects’ regular clothing. Research question: How a new human pose dataset, adapted for gait study, could contribute to the advancement and evaluation of marker-less motion capture systems? Methods: A marker-less system, based on deep learning-based pose estimation methods, was proposed. A new dataset (ENSAM dataset) was collected. Twenty-two asymptomatic adults, one adult with scoliosis, one adult with spondylolisthesis, and seven children with bone disease performed ten walking trials, while being recorded both by the proposed marker-less system and a reference system – combining a marker-based motion capture system and a medical imaging system (EOS). The dataset was split into training and test sets. The pose estimation method, already trained on the Human3.6 M dataset, was evaluated on the ENSAM test set, then reevaluated after further training on the ENSAM training set. The joints coordinates were evaluated, using Bland-Altman bias and 95 % confidence interval, and joint position error (the Euclidean distance between the estimated joint centers and the corresponding reference values). Results: The Bland-Altman 95 % confidence intervals were substantially improved after finetuning the pose estimation method on the ENSAM training set (e.g., from 106.9 mm to 17.4 mm for the hip joint). With the new dataset and approach, the mean joint position error varied from 6.2 mm for ankles to 21.1 mm for shoulders. Significance: The proposed marker-less system achieved promising results in terms of joint position errors. Future studies are necessary to assess the system in terms of gait parameters.

This paper investigates the potential of Deep Learning (DL) for data-driven topology optimization (TO). Unlike the rest of the literature that mainly applies DL to TO from a mechanical perspective, we developed an original approach to integrate mechanical and geometrical constraints simultaneously. Our approach takes as input the mechanical constraints (Boundary conditions, loads configuration, volume fraction) alongside the geometrical ones (total number of elements, minimum overhang, maximum length, minimum thickness) and generates a 2D design complying with these constraints. Thus, it combines the best of both mechanical (CAE) and geometrical design worlds. Conversely, geometrical design constraints are complex, not yet formalized, and contradictory between Additive Manufacturing (AM) processes, applications, and materials. Some are even descriptive, lacking a well-defined mathematical description, or are well-defined but proprietary and inaccessible. Hence, despite the synergy between AM and TO, integrating AM constraints into the TO formulation is still a hurdle. Furthermore, even when their integration is possible, TO’s convergence to a solution is compromised. On the other hand, DL has proven robust in capturing geometrical and spatial correlations. Consequently, our approach solves the previously listed setbacks by aligning DL to serve Design for AM (DfAM); there is no need to identify an analytical formula for a geometrical constraint but simply a sufficient number of examples describing it, and convergence is no longer a blockade when the DL model is trained on converged designs. Our approach tailors the design’s geometrical aspects with great flexibility and creativity. It reconciles design and manufacturing and accelerates the design life cycle of a part. Moreover, it can be easily updated to include additional constraints and can be implemented in the future into CAD software as a lighter and faster generative design module.

merci d'ajouter aux méta données Collaboration/Projet : EcoSD Network Le développement d’approches d’Intelligence Artificielle (IA) dans de nombreux domaines et pour de nombreuses applications interroge, au-delà des enjeux de performances techniques et économiques, sur ses conséquences environnementales. Intégré dans la croissance accélérée du numérique, cette étude cherche à analyser les principaux éléments à retenir pour évaluer les impacts environnementaux de l’IA, en se concentrant, pour commencer, sur sa consommation énergétique. This research study synthesis for the Ademe, the French energy mastering agency, aims at analysing the environmental impacts and footprint of Artificial Intelligence by focusing on energy consumption issues. ADEME et Réseau Nat. EcoSD (écoconception de Systèmes Durables)

The deployment of Deep Learning (DL) strategies is particularly advantageous in Structural Health Monitoring (SHM) based of lamb Wave (LW) propagation due to the high quantity of data collected by the network of piezoelectric transducers (PZT) during all the life cycle of the composite structure. However, such strategies rely on large training databases, difficult to collect experimentally. The use of numerical simulations faces that issue, but the modelsnever fit perfectly to the real structures, leading to error of diagnostic. We propose here to use Transfer Learning (TL) approaches to reduce the predictions errors of a Convolutional Neural Network (CNN) trained with numerical data. The network predicts the size and the position of a damage on a composite plate equipped with PZT. It is trained on a large source database composed of different damage scenarios on a composite plate. A second smaller target database is generated with small variations on the mechanic properties and PZT positions to simulate manufacturing uncertainties. Those uncertainties lead to prediction errors of the CNN. A Domain Adaptation (DA) based on Optimal Transport (OT) is used to project the target data on the source domain and therefore reduces the predictions error of the CNN. These TL approach should allow us to close the gap between experimental and numerical data. FUI MONARQUE

These are a series of online platforms that allow users to rate and comment on VR virtual reality applications. In this paper, we develop a topic model, namely the general and sparse topic model, that automatically identifies a set of features of VR applications from user reviews. In our context, we overcome two severe challenges (i.e., internal noise and limited features mentioned in each review) to successfully learn the features of VR applications. Specifically, we introduce a general topic and a “spike and slab” prior. In addition, we design a collapsed Gibbs sampling algorithm for model inference. We apply this topic model to a dataset from Oculus (namely VREVIEW), and show that our model can identify some distinct, economically meaningful features for VR applications, e.g., “entertainment and fun,” “challenge,” “immersive,” and “sickness.” Our research provides implications for VR consumer behavior analysis, optimizing user experience in virtual environments, and VR application recommendation. This work is supported by the National Natural Science Foundation of China (72101072, 91846201, 72171071, 71722010), the Postdoctoral Research Foundation of China (2021M690852), the Fundamental Research Funds for the Central Universities (JZ2021HGQB0272) and the National Engineering Laboratory for Big Data Distribution and Exchange Technologies.

Recent significant progress in Virtual Reality (VR) applications and environments raised several challenges. They proved to have side effects on specific users, thus reducing the usability of the VR technology in some critical domains, such as flight and car simulators. One of the common side effects is cybersickness. Some significant commonly reported symptoms are nausea, oculomotor discomfort, and disorientation. To mitigate these symptoms and consequently improve the usability of VR systems, it is necessary to predict the incidence of cybersickness. This paper proposes a machine learning approach to VR’s cybersickness prediction based on physiological and subjective data. We investigated combinations of topological data analysis with a range of classifier algorithms and assessed classification performance. The highest performance of Topological Data Analysis (TDA) based methods was achieved in combination with SVMs with Gaussian RBF kernel, indicating that Gaussian RBF kernels provide embeddings of physiological time series data into spaces that are rich enough to capture the essential geometric features of this type of data. Comparing several combinations with feature descriptors for physiological time series, the performance of the TDA + SVM combination is in the top group, statistically being on par or outperforming more complex and less interpretable methods. Our results show that heart rate does not seem to correlate with cybersickness. This work was supported in part by a grant from the French-German University (UFA-DFH) No. CDFA 03–19

Ensuring continued quality is challenging, especially when customer satisfaction is the provided service. It seems to become easier with new technologies like Artificial Intelligence. However, field data are necessary to design an intelligent assistant but are not always available. Synthetic data are used mainly to replace real data. Made with a Generative Adversarial Network or a rendering engine, they aim to be as efficient as real ones in training a Neural Network. When synthetic data generation meets the challenge of object detection, its capacity to deal with the defect detection challenge is unknown. Here we demonstrate how to generate these synthetic data to detect defects. Through iterations, we apply different methods from literature to generate synthetic data for object detection, from how to extract a defect from the few data we have to how to organize the scene before data synthesis. Our study suggests that defect detection may be performed by training an object detector neural network with synthetic data and gives a protocol to do so even if at this point, no field experiments have been conducted to verify our detector performances under real conditions. This experiment is the starting point for developing a mobile and automatic defect detector that might be adapted to ensure new product quality.

Virtual reality and augmented reality (V/AR) are powerful techniques for training, assistance, or supporting activities in the industry. However, the deployment of V/AR techniques in the industry is held back by various factors. The difficulty in adapting an original CAD model to a 3D model for V/AR is one of these factors. Currently, the adaptation process is very time consuming, and its result is unsatisfactory. The adaptation result is often of poor quality or it is poorly adapted to objective of the V/AR activity. A formalization of the specific needs of each V/AR activity will allow improving the level of satisfaction. To ensure a result of the adaptation that meets these specifications, it is necessary to know the most appropriate adaptation process. We, therefore, propose an approach using classification methods to predict the best adaptation process from cases of CAD model preparation for specific V/AR activities. Our approach allows obtaining a 3D model for V/AR without compromise between the quality of the adaptation and its duration.The developer of V/AR application could produce 3D content faster and better. This should facilitate the deployment of V/AR techniques in the industry. The proposed approach is illustrated and validated for two different objectives of industrial V/AR activities.

This paper is to propose a control scheme, a combination of the classical field-oriented control (FOC) technique and artificial intelligence (AI), to obtain constant torques in multiphase non-sinusoidal permanent magnet synchronous machine (PMSM) drives. Higher torque density, easier fabrication, and lower costs are several advantages of non-sinusoidal back electromotive force (back-EMF) machines over sinusoidal ones. However, multi-harmonics existing in back-EMFs possibly generate torque ripples, reducing torque quality of the drive. Therefore, in this paper, an adaptive linear neuron (ADALINE), a simple type of AI, is combined with the classical FOC technique to eliminate these torque ripples. The proposed control scheme is validated by numerical results with a seven-phase PMSM. In addition, these results are compared with an existing strategy to prove its effectiveness. This work has been achieved within the framework of CE2I project. CE2I is co-financed by European Union with the financial support of European Regional Development Fund (ERDF), French State and the French Region of Hautsde- France.

This study is to deal with unwanted current harmonics in rotating (d-q) frames of a 7-phase non-sinusoidal permanent magnet synchronous machine (PMSM) in a wye-connected winding topology. The machine is supplied by a 7-leg voltage source inverter (VSI) fed by a DC-bus voltage. In control, current responses are expected to properly track their references. However, several unwanted harmonics of the non-sinusoidal back electromotive force (back-EMF) and the inverter nonlinearity generate unwanted harmonic components in d-q currents. These current harmonics cannot be nullified by controllers such as conventional proportional-integral (PI) controllers. Consequently, the current responses cannot track their references. In this study, a combination of conventional PI controllers and simple adaptive linear neurons (ADALINEs) is proposed to eliminate these current harmonics, improving current control quality of the drive. The effectiveness of the proposed control structure is verified by experimental results. Thai Nguyen University of Technology and CE2I project

In order to have a full control on their processes, companies need to ensure real time monitoring and supervision using Key Performance Indicators (KPI). KPIs serve as a powerful tool to inform about the process flow status and objectives’ achievement. Although, experts are consulted to analyze, interpret, and explain KPIs’ values in order to extensively identify all influencing factors; this does not seem completely guaranteed if they only rely on their experience. In this paper, the authors propose a generic causality learning approach for monitoring and supervision. A causality analysis of KPIs’ values is hence presented, in addition to a prioritization of their influencing factors in order to provide a decision support. A KPI prediction is also suggested so that actions can be anticipated.

International audience; Multiscale surface characterization is a powerful tool that is used for process monitoring, surface quality control, and manufacturing optimization by establishing a link between process variables and functional performances. The multiscale decomposition approach using continuous and discrete wavelets is widely applied to take into account scales dependency. However, the optimal choice of the analysis wavelet function and the number of decomposition level is still a big issue. In this article, the artificial neural network theory was combined with the wavelet concept and was optimized based on the genetic algorithm to identify the relevant wavelet function for multiscale characterization of abraded surface topographies. Then, an extensive wavelet function library was developed and the proposed algorithm was applied to topographic data obtained from various abrasive finishing processes. The results show the pertinence of this approach to select the relevant wavelet, and a universal relevant wavelet function for abraded surface characterization was determined.

Maps have long been seen as a single cartographic product for different uses, with the user having to adapt their interpretation to his or her own needs. On-demand mapping reverses this paradigm in that it is the map that adapts to the user’s needs and context of use. Still often manual and reserved for professionals, on-demand mapping is evolving toward an automation of its processes and a democratization of its use. An on-demand mapping service is a chain of several consecutive steps leading to a target map that precisely meets the needs and requirements of a user. This article addresses the issue of selecting relevant thematic layers with a specific context of use. We propose a knowledge-based recommendation approach that aims to guide a cartographer through the process of map-making. Our system is based on high- and low-level ontologies, the latter modeling the concepts specific to different types of maps targeted. By focusing on maritime maps, we address the representation of knowledge in this context of use, where recommendations rely on axiomatic and rule-based reasoning. For this purpose, we choose description logics as a formalism for knowledge representation in order to make cartographic knowledge machine readable.

National audience; Les référentiels de données géoréférencées sont de plus en plus utilisés pour permettre l'annotation spatiale de documents textuels et ainsi faciliter l'accès à leur contenu, voire son analyse spatiale. En revanche, peu de travaux se sont intéressés à l'extraction d'information géographique à partir de textes pour alimenter de tels référentiels. Le travail présenté dans cet article explore les potentialités de l'extraction d'information spatiale indirecte (noms de lieux, relations spatiales, etc.) dans les textes des Instructions Nautiques produites par le Service Hydrographique et Océanographique de la Marine (SHOM). La méthode proposée combine une approche lexicale et une approche à base de patrons linguistiques puis est comparée aux principales approches d'extraction d'information géographique en français. L'objectif est de construire et peupler une ontologie assurant le lien entre les connaissances textuelles sur l'environnement côtier et les référentiels cartographiques associés afin de développer des services d'aide à la navigation à base de raisonnement.

Authors of papers retaincopyright and release the workunder a Creative CommonsAttribution 4.0 InternationalLicense (CC-BY) Scientists have long quantified empirical observations by developing mathematical models that characterize the observations, have some measure of interpretability, and are capable of making predictions. Dynamical systems models in particular have been widely used to study, explain, and predict system behavior in a wide range of application areas, with examples ranging from Newton’s laws of classical mechanics to the Michaelis-Menten kinetics for modeling enzyme kinetics. While governing laws and equations were traditionally derived by hand, the current growth of available measurement data and resulting emphasis on data-driven modeling motivates algorithmic approaches for model discovery. A number of such approaches have been developed in recent years and have generated widespread interest, including Eureqa (Schmidt & Lipson, 2009), sure independence screening and sparsifying operator (Ouyang, Curtarolo, Ahmetcik, Scheffler, & Ghiringhelli, 2018), and the sparse identification of nonlinear dynamics (SINDy) (Brunton, Proctor, & Kutz, 2016). Maximizing the impact of these model discovery methods requires tools to make them widely accessible to scientists across domains and at various levels of mathematical expertise. This project is a fork of sparsereg( Quade, 2018). SLB acknowledges funding supportfrom the Air Force Office of Scientific Research (AFOSR FA9550-18-1-0200) and the ArmyResearch Office (ARO W911NF-19-1-0045). JNK acknowledges support from the Air ForceOffice of Scientific Research (AFOSR FA9550-17-1-0329). This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under GrantNumber DGE-1256082.

International audience; This paper first presents a state of the art of context modeling approaches in an information system from a general point of view. We then propose an approach based on a spatial-temporal ontology to automate the data selection step in the process of making personalized maps. The proposed approach of context modeling takes into consideration the user's profile and preferences, as well as events related to an environment. We then aim, by logical reasoning, to be able to deduce a selection of categories and objects that will contribute to the creation of a personalized map.; Cet article présente tout d'abord un état de l'art des approches permettant la modélisation du contexte dans un système d'information, d'un point de vue géné-ral. Nous proposons ensuite une approche fondée sur une ontologie spatio-temporelle afin d'automatiser l'étape de la sélection des données dans un contexte plus spécifique de confection de cartes personnalisées. L'approche proposée tient compte du profil de l'utilisateur et de ses préférences, ainsi que des événements afférents à un environ-nement. Nous souhaitons ensuite, par raisonnement logique, pouvoir en déduire une sélection de catégories et d'objets qui contribuera à la confection d'une carte personnalisée.

International audience; X-ray based quantitative analysis of spine parameters is required in routine diagnosis or treatment planning. Existing tools commonly require manual intervention. Attempts towards automation of the whole procedure have mainly focused on vertebral bodies, whereas other regions such as the posterior arch also bear considerable amount of useful information. In this study, we combine a specific design of contextual visual features with a multi-class Random Forest classifier to perform pixel-wise segmentation and identification of all cervical spine spinous processes, on sagittal radiographs. Segmentations were evaluated on 62 radiographs, comparing to manual tracing. Correct identification was obtained for all subjects, and segmentation returned mean  SD values of: Dice coefficient =88  8%; Hausdorff distance =2.1  1.4 mm and; mean surface distance =0.6  0.4 mm. The derived geometric parameters can be used to reduce the amount of manual intervention needed for spine modeling or to measure clinical indices.

This paper proposes a new objective metric of visually induced motion sickness (VIMS) in the context of navigation in virtual environments (VEs). Similar to motion sickness in physical environments, VIMS can induce many physiological symptoms such as general discomfort, nausea, disorientation, vomiting, dizziness and fatigue. To improve user satisfaction with VR applications, it is of great significance to develop objective metrics for VIMS that can analyze and estimate the level of VR sickness when a user is exposed to VEs. One of the well-known objective metrics is the postural instability. In this paper, we trained a LSTM model for each participant using a normal-state postural signal captured before the exposure, and if the postural sway signal from post-exposure was sufficiently different from the pre-exposure signal, the model would fail at encoding and decoding the signal properly; the jump in the reconstruction error was called loss and was proposed as the proposed objective measure of simulator sickness. The effectiveness of the proposed metric was analyzed and compared with subjective assessment methods based on the simulator sickness questionnaire (SSQ) in a VR environment, achieving a Pearson correlation coefficient of .89. Finally, we showed that the proposed method had the potential to be deployed within a closed-loop system and get real-time performance to predict VR sickness, opening new insights to develop user-centered and customized VR applications based on physiological feedback. China Scholarship Council under grant No. 201708390014

International audience; Designing appealing products plays a key role in commercial success. Understanding the relationships between aesthetic properties and shape characteristics of a product can contribute to define user-friendly and interactive designing tools supporting the early design phases. This paper introduces a generic framework for mapping aesthetic properties to 3D free form shapes. The approach uses machine learning techniques to identify rules between the user-defined classifications of shapes and the geometric parameters of the underlying free form surfaces and to create an efficient classification model. The frameworkhas been set up and validated focusing on the flatness aesthetic property but is generic and can be applied to others. Several experiments have been conducted to understand if there is a consistency among people in the judgement of a specific aesthetic properties, if and to which extent the surrounding of the judged surface affects the perception consistency, and which are the surface geometric quantities influencing the perception. A graphic user interface has been designed to allow a fast classification of thousands of shapes automatically generated. The experiments have been conducted following a systematic methodology comparing two different approaches. The results confirm that the perception of flatness is commonly shared by the majority and the most relevant attributes have been identified. Additionally, it results that the surrounding information extension and context influence the perception of the flatness strengthening the classification consistency. The way those results can be used to design new interactive tools and to improve the product design process is discussed.

International audience; In this paper, we present a new deep learning-based human activity recognition technique. First, we track and extract human body from each frame of the video stream. Next, we abstract human silhouettes and use them to create binary space-time maps (BSTMs) which summarize human activity within a defined time interval. Finally, we use convolutional neural network (CNN) to extract features from BSTMs and classify the activities. To evaluate our approach, we carried out several tests using three public datasets: Weizmann, Keck Gesture and KTH Database. Experimental results show that our technique outperforms conventional state-of-the-art methods in term of recognition accuracy and provides comparable performance against recent deep learning techniques. It’s simple to implement, requires less computing power, and can be used for multi-subject activity recognition.

Manually annotating medical images with few landmarks to initialize 3D shape models is a common practice. For instance, when reconstructing the 3D spine from biplanar X-rays, the spinal midline, passing through vertebrae body centers (VBCs) and endplate midpoints, is required. This paper presents an automated spinal midline delineation method on frontal and sagittal views by using Mask R-CNN. The network detects all vertebrae from C7 to L5, followed by vertebrae segmentation and classification at the same time. After postprocessing to discard outliers, the vertebrae mask centers were regarded as VBCs to get the spine midline by polynomial fitting. Evaluation of the spinal midline on 136 images used root mean square error (RMSE) with respect to manual ground-truth. The RMSE ± standard error values of predicted spinal midlines (C7-L5) were 1.11 mm ± 0.67 mm on frontal views and 1.92mm ± 1.38 mm on sagittal views. The proposed method is capable of delineating spinal midlines on patients with different spine deformity degrees. BiomecAM chair program

International audience; Human pose estimation (HPE) methods based on convolutional neural networks (CNN) have demonstrated significant progress and achieved state-of-the-art results on human pose datasets. In this study, we aimed to assess the perfor-mance of CNN-based HPE methods for measuring anthropometric data. A Vicon motion analysis system as the reference system and a stereo vision system recorded ten asymptomatic subjects standing in front of the stereo vision system in a static posture. Eight HPE methods estimated the 2D poses which were transformed to the 3D poses by using the stereo vision system. Percentage of correct keypoints, 3D error, and absolute error of the body segment lengths are the evaluation measures which were used to assess the results. Percentage of correct keypoints – the stand-ard metric for 2D pose estimation – showed that the HPE methods could estimate the 2D body joints with a minimum accuracy of 99%. Meanwhile, the average 3D error and absolute error for the body segment lengths are 5 cm.

Over the past few years, a series of computational and semantic frameworks have been developed to model and represent the spatial and temporal properties of moving entities. Despite the interest of these contributions, it is recognized that there is still a need for a qualitative reasoning support at the abstract and formal levels. The research presented in this paper introduces a qualitative approach for representing and manipulating moving entities. The model combines topological relations with qualitative distances over a spatial and temporal framework. Several basic movement configurations over dynamic entities are identified as well as movement transitions. The whole approach is illustrated in the context of the analysis of flight patterns.

In a Smart Grid context, the increasing penetration of embedded generation units leads to a greater complexity in the management of production units. In this arti- cle, we focus on the impact of the introduction of decentralized generation for the unit commitment problem (UC). Unit Commitment Problems consist in finding the optimal schedules and amounts of power to be generated by a set of gen- erating units in response to an electricity demand forecast. While this problem have received a significant amount of attention, classical approaches assume these problems are centralized and deterministic. However, these two assumptions are not realistic in a smart grid context. Indeed, finding the optimal schedules and amounts of power to be generated by multiple distributed generator units is not trivial since it requires to deal with distributed computation, privacy, stochastic planning, ... In this paper, we focus on smart grid scenarios where the main source of complexity comes from the proliferation of distributed generating units. In solving this issue, we consider distributed stochastic unit commitment prob- lems. We introduce a novel distributed gradient descent algorithm which allow us to circumvent classical assumptions. This algorithm is evaluated through a set of experiments on real-time power grid simulator. Programme ADEME - Réseaux électrique intelligent - Projet AgentVPP

Structural damages result in nonlinear dynamical signatures that significantly help for their monitoring. A damage type classification approach is proposed here that is based on a parallel Hammerstein models representation of the structure estimated by means of the Exponential Sine Sweep Method. This estimation method has been here extended to take into account for input signal amplitude which was not the case before. On the basis of these estimated models, three amplitude dependent damage indexes are built: one that monitors the shift of the resonance frequency of the structure, another the ratio of nonlinear versus linear energy in the output signal, and a last one the ratio of the energy coming from odd nonlinearities to the energy coming from even nonlinearities in the output signal. The slopes of these amplitude-dependent DIs are then used as coordinates to place the damaged structure under study within a three-dimensional space. A single mass-spring-damper system is considered to illustrate the ability of this space to classify different types of damage. Four types of damage with different severities are simulated through different spring nonlinearities: bilinear stiffness, dead zone, saturation, and Coulomb friction. For all severities, the four types of damage are extremely well separated within the proposed three-dimensional space, thus highlighting its high potential for classification purposes.; International audience; Structural damages result in nonlinear dynamical signatures that significantly help for their monitoring. A damage type classification approach is proposed here that is based on a parallel Hammerstein models representation of the structure estimated by means of the Exponential Sine Sweep Method. This estimation method has been here extended to take into account for input signal amplitude which was not the case before. On the basis of these estimated models, three amplitude dependent damage indexes are built: one that monitors the shift of the resonance frequency of the structure, another the ratio of nonlinear versus linear energy in the output signal, and a last one the ratio of the energy coming from odd nonlinearities to the energy coming from even nonlinearities in the output signal. The slopes of these amplitude-dependent DIs are then used as coordinates to place the damaged structure under study within a three-dimensional space. A single mass-spring-damper system is considered to illustrate the ability of this space to classify different types of damage. Four types of damage with different severities are simulated through different spring nonlinearities: bilinear stiffness, dead zone, saturation, and Coulomb friction. For all severities, the four types of damage are extremely well separated within the proposed three-dimensional space, thus highlighting its high potential for classification purposes.

International audience; Artificial potential field is investigated to provide a high level of synergy between driver and semi-autonomous vehicle. This article presents a framework developed to test the performances of this approach. Stand-alone performances of this system is tested for a lane keeping and cruise control application. Performances are promising and future development is discussed.

International audience; Structural damages can result in nonlinear dynamical signatures that can significantly enhance their detection. An original nonlinear damage detection approach is proposed that is based on a cascade of Hammerstein models representation of the structure. This model is estimated by means of the Exponential Sine Sweep Method from only one measurement. On the basis of this estimated model, the linear and nonlinear parts of the output are estimated, and two damage indexes (DIs) are proposed. The first DI is built as the ratio of the energy contained in the nonlinear part of an output versus the energy contained in its linear part. The second DI is the angle between the subspaces obtained from the nonlinear parts of two set of outputs after a principal component analysis. The sensitivity of the proposed DIs to the presence of damages as well as their robustness to noise are assessed numerically on spring-mass-damper structures and experimentally on actual composite plates with surface-mounted PZT-elements. Results demonstrate the effectiveness of the proposed method to detect a damage in nonlinear structures and in the presence of noise.

International audience; Route planning is an important problem for many real-time applications in open and complex environments. The maritime domain is a relevant example of such environments where dynamic phenomena and navigation constraints generate difficult route finding problems. This paper develops a spatial data structure that supports the search for an optimal route between two locations while minimizing a cost function. Although various search algorithms have been proposed so far (e.g. breadth-first search, bidirectional breadth-first search, Dijkstra's algorithm, A*, etc.), this approach provides a bidirectional dynamic routing algorithm which is based on hexagonal meshes and an iterative deepening A* (IDA*) algorithm, and a front to front strategy using a dynamic graph that facilitates data accessibility. The whole approach is applied to the context of maritime navigation, taking into account navigation hazards and restricted areas. The algorithm developed searches for optimal routes while minimizing distance and computational time.

Traditional Ecological Knowledge (TEK) has been at the centre of mapping efforts for decades. Indigenous knowledge (IK) is a critical subset of TEK, and Indigenous peoples utilise a wide variety of techniques for keeping track of time. Although techniques for mapping and visualising the temporal aspects of TEK/IK have been utilised, the spatio-temporal dimensions of TEK are not well explored visually outside of seasonal data and narrative approaches. Existing spatio-temporal models can add new visualisation approaches for TEK but are limited by ontological constraints regarding time, particularly the poor support for multi-cyclical data and localised timing. For TEK to be well represented, flexible systems are needed for modelling and mapping time that correspond well with traditional conceptions of time and space being supported. These approaches can take cues from previous spatio-temporal visualisation work in the Geographic(al) Information System(s)/Science(s) GIS community, and from temporal depictions extant in existing cultural traditions.

International audience; The motion capture systems are increasingly used for biomedical purposes. In order to recognize and classify the movements, however whole-body movements using passive markers, generate a huge amount of data. Can topological data analysis methods improve the recognition of movements? Can we use the results of this analysis combined with particular types of neural networks to anticipate the continuation of a movement?

International audience; Traditional online or in-process surface profile (quality) evaluation (prediction) needs to integrate cutting parameters and several in-process factors (vibration, machine dynamics, tool wear, etc) for high accuracy. However it might result in high measuring cost and complexity, and moreover, the surface profile (quality) evaluation result can only be obtained after machining process. In this paper an approach for surface profile pre-evaluation in turning process using cutting parameters and radial basis function (RBF) neural networks is presented. The aim was to only use three cutting parameters to predict surface profile before machining process for a fast pre-evaluation on surface quality under different cutting parameters. The input parameters of RBF networks are cutting speed, depth of cut, and free rate. The output parameters are FFT vector of surface profile as prediction (pre-evaluation) result. The RBF networks are trained with adaptive optimal training parameters related to cutting parameters and predict surface profile using the corresponding optimal network topology for each new cutting condition. It was found that a very good performance of surface profile prediction, in terms of agreement with experimental data, can be achieved before machining process with high accuracy, low cost, and high speed. Furthermore, a new group of training and testing data was also used to analyze the influence of tool wear on prediction accuracy.

Part 1: Knowledge Sharing, Re-use and Preservation; International audience; Research on process mining and machine learning techniques has recently received a significant amount of attention by product development and management communities. Indeed, these techniques allow both an automatic process and activity discovery and thus are high added value services that help reusing knowledge to support decision-making. This paper proposes a double layer framework aiming to identify the most significant process patterns to be executed depending on the design context. Simultaneously, it proposes the most significant parameters for each activity of the considered process pattern. The framework is applied on a specific design example and is partially implemented.

This paper presents a temperature compensation method for Lamb wave structural health monitoring. The proposed approach considers a representation of the piezo-sensor signal through its Hilbert transform that allows one to extract the amplitude factor and the phase shift in signals caused by temperature changes. An ordinary least square (OLS) algorithm is used to estimate these unknown parameters. After estimating these parameters at each temperature in the operating range, linear functional relationships between the temperature and the estimated parameters are derived using the least squares method. A temperature compensation model is developed based on this linear relationship that allows one to reconstruct sensor signals at any arbitrary temperature. The proposed approach is validated numerically and experimentally for an anisotropic composite plate at different temperatures ranging from [Formula: see text] to [Formula: see text]. A close match is found between the measured signals and the reconstructed ones. This approach is interesting as it needs only a limited set of piezo-sensor signals at different temperatures for model training and temperature compensation at any arbitrary temperature. Damage localization results after temperature compensation demonstrate its robustness and effectiveness.

12th International Conference, CDVE 2015, Mallorca, Spain, September 20-23, 2015. Proceedings; International audience; With the fastidiously ever-increasing complexity of systems, the relentless, massive customisation of products and the mushrooming accumulation of legal documents (standards, policies and laws), we can observe a significant increase in requirements. We consider the tremendous volume of requirements as big data with which companies struggle to make strategic decisions early on. This paper proposes a collaborative requirement mining framework to enable the decision-makers of an Original Equipment Manufacturer (OEM) to gain insight and discover opportunities in a massive set of requirements so as to make early effective strategic decisions. The framework supports OEMs willing to uncover a subset of key requirements by distilling large unstructured and semi-structured specifications.

The objective of this paper is to present the advantages of Model reference adaptive control (MRAC) motion cueing algorithm against the classical motion cueing algorithm in terms of biome- chanical reactions of the participants during the critical maneuvers like chicane in driving simu- lator real-time. This study proposes a method and an experimental validation to analyze the ves- tibular and neuromuscular dynamics responses of the drivers with respect to the type of the con- trol used at the hexapod driving simulator. For each situation, the EMG (electromyography) data were registered from arm muscles of the drivers (flexor carpi radialis, brachioradialis). In addi- tion, the roll velocity perception thresholds (RVT) and roll velocities (RV) were computed from the real-time vestibular level measurements from the drivers via a motion-tracking sensor. In or- der to process the data of the EMG and RVT, Pearson’s correlation and a two-way ANOVA with a significance level of 0.05 were assigned. Moreover, the relationships of arm muscle power and roll velocity with vehicle CG (center of gravity) lateral displacement were analyzed in order to assess the agility/alertness level of the drivers as well as the vehicle loss of control characteristics with a confidence interval of 95%. The results showed that the MRAC algorithm avoided the loss of adhe- sion, loss of control (LOA, LOC) more reasonably compared to the classical motion cueing algo- rithm. According to our findings, the LOA avoidance decreased the neuromuscular-visual cues lev- el conflict with MRAC algorithm. It also revealed that the neuromuscular-vehicle dynamics conflict has influence on visuo-vestibular conflict; however, the visuo-vestibular cue conflict does not in- fluence the neuromuscular-vehicle dynamics interactions.; International audience; The objective of this paper is to present the advantages of Model reference adaptive control (MRAC) motion cueing algorithm against the classical motion cueing algorithm in terms of biome- chanical reactions of the participants during the critical maneuvers like chicane in driving simu- lator real-time. This study proposes a method and an experimental validation to analyze the ves- tibular and neuromuscular dynamics responses of the drivers with respect to the type of the con- trol used at the hexapod driving simulator. For each situation, the EMG (electromyography) data were registered from arm muscles of the drivers (flexor carpi radialis, brachioradialis). In addi- tion, the roll velocity perception thresholds (RVT) and roll velocities (RV) were computed from the real-time vestibular level measurements from the drivers via a motion-tracking sensor. In or- der to process the data of the EMG and RVT, Pearson’s correlation and a two-way ANOVA with a significance level of 0.05 were assigned. Moreover, the relationships of arm muscle power and roll velocity with vehicle CG (center of gravity) lateral displacement were analyzed in order to assess the agility/alertness level of the drivers as well as the vehicle loss of control characteristics with a confidence interval of 95%. The results showed that the MRAC algorithm avoided the loss of adhe- sion, loss of control (LOA, LOC) more reasonably compared to the classical motion cueing algo- rithm. According to our findings, the LOA avoidance decreased the neuromuscular-visual cues lev- el conflict with MRAC algorithm. It also revealed that the neuromuscular-vehicle dynamics conflict has influence on visuo-vestibular conflict; however, the visuo-vestibular cue conflict does not in- fluence the neuromuscular-vehicle dynamics interactions.

Kansei studies refer to the more and more holistic consideration of the cognitive and affective processes which occur during user experience. In addition, few studies deal with the experience of the designers during the design process, and its influence on the final design outputs. Historically kansei engineering has been firstly focused on the semantic differential approach. Afterwards emotions were integrated into kansei approaches. The semantic differential approach enabled to evaluate products and then to generate automatically design solutions with semantic input data. Thereafter, evaluations have been completed by physiological measurements in order to reduce the subjectivity involved in those evaluations and also to capture some unconscious reactions. This implementation is still in process. Today kansei studies have been much enriched from the three disciplines of design science, psychology and artificial intelligence. The cross influence between these disciplines brought new dimensions into kansei approaches (multisensory design information, personality, values, and culture, new formalisms and algorithms) which lead progressively towards the consideration of a whole enriched kansei experience. We propose in this paper a description of the nature of kansei information. Then we present some major orientations for kansei evaluation. Finally we propose an overall table gathering information about kansei dimensions and formats.; International audience; Kansei studies refer to the more and more holistic consideration of the cognitive and affective processes which occur during user experience. In addition, few studies deal with the experience of the designers during the design process, and its influence on the final design outputs. Historically kansei engineering has been firstly focused on the semantic differential approach. Afterwards emotions were integrated into kansei approaches. The semantic differential approach enabled to evaluate products and then to generate automatically design solutions with semantic input data. Thereafter, evaluations have been completed by physiological measurements in order to reduce the subjectivity involved in those evaluations and also to capture some unconscious reactions. This implementation is still in process. Today kansei studies have been much enriched from the three disciplines of design science, psychology and artificial intelligence. The cross influence between these disciplines brought new dimensions into kansei approaches (multisensory design information, personality, values, and culture, new formalisms and algorithms) which lead progressively towards the consideration of a whole enriched kansei experience. We propose in this paper a description of the nature of kansei information. Then we present some major orientations for kansei evaluation. Finally we propose an overall table gathering information about kansei dimensions and formats.

Arts et Metiers ParisTech built the SAAM driving simulator in collaboration with Renault and Grand Chalon. This research study was realized in the framework of the geDriver project.; International audience; This paper deals with driving simulation and in particular with the important issue of motion sickness. The paper proposes a methodology to evaluate the objective illness rating metrics deduced from the motion sickness dose value and questionnaires for both a static simulator and a dynamic simulator. Accelerations of the vestibular cues (head movements) of the subjects were recorded with and without motion platform activation. In order to compare user experiences in both cases, the head-dynamics-related illness ratings were computed from the obtained accelerations and the motion sickness dose values. For the subjective analysis, the principal component analysis method was used to determine the conflict between the subjective assessment in the static condition and that in the dynamic condition. The principal component analysis method used for the subjective evaluation showed a consistent difference between the answers given in the sickness questionnaire for the static platform case from those for the dynamic platform case. The two-tailed Mann-Whitney U test shows the significance in the differences between the self-reports to the individual questions. According to the two-tailed Mann-Whitney U test, experiencing nausea (p = 0.019 < 0.05) and dizziness (p = 0.018 < 0.05) decreased significantly from the static case to the dynamic case. Also, eye strain (p = 0.047 < 0.05) and tiredness (p = 0.047 < 0.05) were reduced significantly from the static case to the dynamic case. For the perception fidelity analysis, the Pearson correlation with a confidence interval of 95% was used to study the correlations of each question with the x illness rating component IRx, the y illness rating component IRy, the z illness rating component IRz and the compound illness rating IRtot. The results showed that the longitudinal head dynamics were the main element that induced discomfort for the static platform, whereas vertical head movements were the main factor to provoke discomfort for the dynamic platform case. Also, for the dynamic platform, lateral vestibular-level dynamics were the major element which caused a feeling of fear.

This paper examines an important challenge in ultrasonic structural health monitoring (SHM), which is the problem of the optimal placement of sensors in order to accurately detect and localize damages. The goal of this study is to enhance damage detection through an optimal sensor placement (OSP) algorithm. The problem is formulated as a global optimization problem, where the objective function to be maximized is evaluated by a ray tracing approach, which approximately models Lamb waves propagation. A genetic algorithm (GA) is then used to solve this optimization problem. Simulations and experiments were conducted to validate the proposed method on a carbon epoxy composite plate. Results show the effectiveness and the advantages of the proposed method as a tool for OSP with reasonable computation time. Projet AIRCELLE (EPICE/CORALIE)

The paper presents a study of aerodynamic characteristics of a car, which has the simplified geometric shape, so called Ahmed body. Flow around the body and the influence of its rear slant angle on drag are widely studied by numerous researchers. However, small number of studies treats the relationship between drag and lift and this phenomenon is not fully understood. To clarify the relationship between lift and drag, experiments are conducted in the wind tunnel of ENSAM - Paris. The study is carried out for different rear slant angles in order to determine how the drag coefficient varies with lift. The results of experiments are completed by numerical simulations, which permit to obtain the detailed flow field around Ahmed body and to understand better the effect of rear slant angle on drag and lift coefficients.; International audience; The paper presents a study of aerodynamic characteristics of a car, which has the simplified geometric shape, so called Ahmed body. Flow around the body and the influence of its rear slant angle on drag are widely studied by numerous researchers. However, small number of studies treats the relationship between drag and lift and this phenomenon is not fully understood. To clarify the relationship between lift and drag, experiments are conducted in the wind tunnel of ENSAM - Paris. The study is carried out for different rear slant angles in order to determine how the drag coefficient varies with lift. The results of experiments are completed by numerical simulations, which permit to obtain the detailed flow field around Ahmed body and to understand better the effect of rear slant angle on drag and lift coefficients.

International audience; In model-driven engineering, model transformations are con- sidered a key element to generate and maintain consistency between re- lated models. Rule-based approaches have become a mature technology and are widely used in different application domains. However, in var- ious scenarios, these solutions still suffer from a number of limitations that stem from their injective and deterministic nature. This article pro- poses an original approach, based on non-deterministic constraint-based search engines, to define and execute bidirectional model transforma- tions and synchronizations from single specifications. Since these solely rely on basic existing modeling concepts, it does not require the intro- duction of a dedicated language. We first describe and formally define this model operation, called transformation as search, then describe a proof-of-concept implementation and discuss experiments on a reference use case in software engineering.

This paper examines some aspects of the usefulness of interactive tabletop systems, if and how these impact collaboration. We chose creative problem solving such as brainstorming as an application framework to test several collaborative media: the use of pen-and-paper tools, the ''around-the-table'' form factor, the digital tabletop interface, the attractiveness of interaction styles. Eighty subjects in total (20 groups of four members) participated in the experiments. The evaluation criteria were task performance, collaboration patterns (especially equity of contributions), and users' subjective experience. The ''around-the-table'' form factor, which is hypothesized to promote social comparison, increased performance and improved collaboration through an increase of equity. Moreover, the attractiveness of the tabletop device improved subjective experience and increased motivation to engage in the task. However, designing attractiveness seems a highly challenging issue, since overly attractive interfaces may distract users from the task.

In this paper we will present a state of the art of the descriptive and generative models for shape. We will present several different approaches for the manipulation of shape in computational systems: numerical models, graph models, descriptive models. This investigation will lead to a discussion regarding the use of these models for supporting the generation of shapes in the early phases of the design process. ANR GENIUS (TECHLOG-07-010)

International audience; This paper proposes an isomorphism between the number of accessible states of the system including statistical fluctuation and the optimised size of a computer program set-up by combining the Huffman and R.L.E algebrae. At first, a relation between the variation of the free energy in the system and the variation of the programme size used to describe the system was stated. A fluctuation to the equilibrium decreases the size of the programme in relation to the decrease in entropy and the decay of the free energy of the system.