Industrial robots are delivering more and more manipulation services in manufacturing. However, when the task is complex, it is difficult to programme a robot to fulfil all the requirements because even a relatively simple task such as a peg-in-hole insertion contains many uncertainties, e.g. clearance, initial grasping position and insertion path. Humans, on the other hand, can deal with these variations using their vision and haptic feedback. Although humans can adapt to uncertainties easily, most of the time, the skilled based performances that relate to their tacit knowledge cannot be easily articulated. Even though the automation solution may not fully imitate human motion since some of them are not necessary, it would be useful if the skill based performance from a human could be firstly interpreted and modelled, which will then allow it to be transferred to the robot. This thesis aims to reduce robot programming efforts significantly by developing a methodology to capture, model and transfer the manual manufacturing skills from a human demonstrator to the robot. Recently, Learning from Demonstration (LfD) is gaining interest as a framework to transfer skills from human teacher to robot using probability encoding approaches to model observations and state transition uncertainties. In close or actual contact manipulation tasks, it is difficult to reliabley record the state-action examples without interfering with the human senses and activities. Therefore, wearable sensors are investigated as a promising device to record the state-action examples without restricting the human experts during the skilled execution of their tasks. Firstly to track human motions accurately and reliably in a defined 3-dimensional workspace, a hybrid system of Vicon and IMUs is proposed to compensate for the known limitations of the individual system. The data fusion method was able to overcome occlusion and frame flipping problems in the two camera Vicon setup and the drifting problem associated with the IMUs. The results indicated that occlusion and frame flipping problems associated with Vicon can be mitigated by using the IMU measurements. Furthermore, the proposed method improves the Mean Square Error (MSE) tracking accuracy range from 0.8˚ to 6.4˚ compared with the IMU only method. Secondly, to record haptic feedback from a teacher without physically obstructing their interactions with the workpiece, wearable surface electromyography (sEMG) armbands were used as an indirect method to indicate contact feedback during manual manipulations. A muscle-force model using a Time Delayed Neural Network (TDNN) was built to map the sEMG signals to the known contact force. The results indicated that the model was capable of estimating the force from the sEMG armbands in the applications of interest, namely in peg-in-hole and beater winding tasks, with MSE of 2.75N and 0.18N respectively. Finally, given the force estimation and the motion trajectories, a Hidden Markov Model (HMM) based approach was utilised as a state recognition method to encode and generalise the spatial and temporal information of the skilled executions. This method would allow a more representative control policy to be derived. A modified Gaussian Mixture Regression (GMR) method was then applied to enable motions reproduction by using the learned state-action policy. To simplify the validation procedure, instead of using the robot, additional demonstrations from the teacher were used to verify the reproduction performance of the policy, by assuming human teacher and robot learner are physical identical systems. The results confirmed the generalisation capability of the HMM model across a number of demonstrations from different subjects; and the reproduced motions from GMR were acceptable in these additional tests. The proposed methodology provides a framework for producing a state-action model from skilled demonstrations that can be translated into robot kinematics and joint states for the robot to execute. The implication to industry is reduced efforts and time in programming the robots for applications where human skilled performances are required to cope robustly with various uncertainties during tasks execution.

Submitted by Morgana Andrade (morgana.andrade@ufes.br) on 2016-04-19T20:27:17Z No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) TeseDoutoradoCompleta.pdf: 11628770 bytes, checksum: 78cb42f5e620bb943765674da68b0c7e (MD5) / Approved for entry into archive by Patricia Barros (patricia.barros@ufes.br) on 2016-05-16T14:15:50Z (GMT) No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) TeseDoutoradoCompleta.pdf: 11628770 bytes, checksum: 78cb42f5e620bb943765674da68b0c7e (MD5) / Made available in DSpace on 2016-05-16T14:15:50Z (GMT). No. of bitstreams: 2 license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) TeseDoutoradoCompleta.pdf: 11628770 bytes, checksum: 78cb42f5e620bb943765674da68b0c7e (MD5) / CNPq / Os avanços tecnológicos na última década permitiram o desenvolvimento de sistemas de processamento de informação com alta capacidade de armazenamento de dados. Estes avanços na linha de saúde têm evoluíram para o desenvolvimento de dispositivos para aplicações na Bioengenharia e Engenharia Biomédica, no auxílio à compreensão do comportamento fisiológico, diagnóstico, monitoramento, controle e tratamento de variadas alterações biológicas. Juntamente com os avanços tecnológicos, a quantidade e complexidade da informação é cada vez maior, em comparação com a utilidade e compreensão da mesma, representando, para diferentes áreas de conhecimento, um desafio na busca de alternativas viáveis que permitam utilizar os atributos dos sistemas biológicos no desenvolvimento de novas tecnologias para a melhoria da qualidade de vida dos seres humanos. Na atualidade, o desenvolvimento de protocolos de captura de sinais bioelétricos não invasivos está conformando uma opção viável para o diagnóstico de miopatias, reabilitação motora, análise biomecânica, desenvolvimento de Interface Homem-Máquina, e controle autônomo de dispositivos robóticos para pessoas com deficiência motora grave, entre outras aplicações. Em todos os casos, o auxílio de técnicas computacionais como processamento de sinais digitais (DSP), e novos algoritmos baseados em inteligência artificial, abriram a possibilidade de desenvolver técnicas de classificação para o reconhecimento de padrões que podem ser aplicadas na área de biotecnologia para a saúde. A presente tese de doutorado desenvolve protocolos e técnicas de análise de sinais mioelétricas (SME) por eletromiografia de superfície (sEMG) constituídos por “tarefas de atraso instruídas”, aplicados à avaliação motora e reabilitação, que envolve análise e critérios de inclusão-exclusão por anamnese clínica, controle de variáveis no ambiente experimental, captura, aquisição e transformação do sina, digitalização, filtragem, segementação, seleção de características, classificação e reconhecimento de padrões. As aplicações biotecnológicas com SME apresentam uma abordagem quantitativa experimental em forma de estudo de caso. O primeiro estudo de caso desenvolve três protocolos de aquisição para avaliação proprioceptiva do joelho, controle de uma cadeira de rodas robótica por pessoas com deficiência motora grave, e manipulação de um robô móvel por crianças com deficiência cognitiva e motora, utilizando um sensor híbrido (inclinação+sEMG), o qual conformou inclusive uma patente de invenção derivada da presente tese. O segundo estudo de caso desenvolve um protocolo de aquisição SME, para o auxílio ao diagnóstico de fibromialgia utilizando algoritmos para avaliação da fadiga muscular no domínio do tempo (ARV, RMS) e da frequência (MNF, MDF, AIF) com 30%, 60% e 80% de MVC. O terceiro estudo de caso desenvolve um protocolo de aquisição de SME de baixa densidade e baixo nível de contração muscular, com controle do repouso, para o reconhecimento de diferentes gestos da mão, em pessoas saudáveis e com amputação na região do terço distal do cotovelo, avaliando 14 características, 8 no domínio do tempo, 5 no domínio da frequência e Dimensão Fractal (FD), além de várias das sua combinações, as quais foram classificadas com técnicas computacionais de inteligênica artificial como lógica difusa (FL) e redes neurais artificiais do tipo MLP. Os resultados obtidos para o primeiro estudo de caso demonstrou a utilidade da predeterminação de limiares para as variáveis RMS e inclinação obtidas com o sensor híbrido (inclinação+sEMG), melhorando a precisão do senso de posicionamento na análise proprioceptiva do joelho em comparação com um eletrogoniômetro comercial em combinação com o SME. O sensor híbrido facilitou também o controle de uma cadeira de rodas robótica, utilizando o movimento da cabeça para o deslocamento autônomo de pessoas com tetraplegia, assim como, a manipulação autônoma de um robô móvel por pessoas com deficiência cognitiva e motora, os quais obtiveram, com o treinamento, um melhor desempenho na interação com o robô, avaliado pelo índice GAS. No segundo estudo de caso, os resultados obtidos para avaliação da fadiga em pessoas com fibromialgia (FM) indicaram uma relação entre o aumento da carga e a dor muscular, especialmente para 80% de MVC. A regressão linear dos algoritmos RMS, ARV e MNF apresentaram na inclinação (α) e intercepto (β) uma tendência esperada no grupo controle, com regressão linear positiva para características no domínio do tempo e negativas para características no domínio da frequência, para 60% de MVC e 60% do segmento isométrico do SME, obtidos com 20 contrações isotônicas durante a flexão extensão do bíceps braquii (RMS α=1.1319, β=275.706; MNF α=-0.470, β=91.482). No caso de voluntárias com FM, a voluntária N3 apresentou dados com maior relação de tendência esperada da fadiga muscular, para 80% de MVC e 60% do segmento isométrico obtidos durante movimento isotônico do bíceps braquii (RMS α=5,92 β=113,33; MNF α=-1,21 β=96,96). Por último, o terceiro estudo de caso identificou, com UM classificador MLP, e taxa de sucesso de 94,9% seis movimentos de dedos individuais, incluindo repouso, (categoria A), e com 97,5% de taxa de sucesso, sete movimentos que compreendem dedos, punho e agarre (categoria B), ambos os casos, com combinação de características RMS, WL,MAV e ZC. Por outro lado, resultados obtidos por voluntários amputados no terço distal do cotovelo, apresentaram melhores resultados com características no domínio do tempo, em comparação as que incluiram dimensão fractal (DF), com taxas de sucesso de 93,9%, utilizando combinação de características RMS, WL e MAV para a categoria A, e 95,4% de taxa de sucesso, com uma combinação de características RMS, WL, MAV e ZC na categoria B. / Technological advances in the last decade opened up the field for the development of information processing systems with high capacity of data storage. These advances in health have evolved in the development of devices for applications in Bioengineering and Biomedical Engineering, supporting the understanding of the physiological behavior, diagnosis, monitoring, treatment and control of various biological changes. Along with technological advances, the amount and complexity of information is increasing, compared to its usefulness and understanding, representing, for different areas of knowledge, a challenge to find viable alternatives for using the attributes of biological systems in the development of new technologies directed to improve the quality of life of human beings. Currently, the development of noninvasive protocols for capturing bioelectric signals are becoming a viable option for the diagnosis of myopathies, motor rehabilitation, biomechanical analysis, development of Human-Machine Interface, and autonomous control of robotic devices for people with severe motor disabilities among other applications. In all cases, the support of computational techniques, such as digital signal processing (DSP), and new algorithms based on artificial intelligence, has opened the opportunity to develop classification techniques for recognizing patterns which can be applied in biotechnology for health. This doctoral thesis develops protocols and techniques for analysis of sEMG signals, consisting of "instructed delay tasks", applied to the motor assessment and rehabilitation estrategies, involving analysis of inclusion-exclusion criteria for clinical history, control variables in experimental environment, capture, acquisition and processing of sEMG signal, digital group, filtering, segmentation, feature selection, classification and pattern recognition. Biotechnological applications with sEMG signals present a quantitative experimental approach in the form of case studies. The first case study is centered on three acquisition protocols for evaluation of proprioceptive knee, control of a robotic wheelchair for people with severe motor disabilities, and manipulation of a mobile robot for children with cognitive and motor disability, using a hybrid sensor (inclination + sEMG), which is a patent derivate of this thesis. The second case study, develops a protocol for acquisition of sEMG signals in order, to support the diagnosis of fibromyalgia using algorithms for evaluation of muscle fatigue in time domain (ARV, RMS) and frequency domain (MNF, MDF, AIF), with 30%, 60% and 80% of MVC. The third case study, develops a protocol for the acquisition of sEMG signals with low density and low level of muscle contraction, with control of the rest, for the recognition of different hand gestures in healthy and amputees, evaluating 14 characteristics , 8 in time domain, and 5 in frequency domain and Fractal Dimension (FD), with several of their combinations, which were classified with computational techniques of artificial intelligence, such as fuzzy logic (FL) and artificial neural networks of MLP type. The results for the first case study, has demonstrated the usefulness of threshold predetermination as RMS and slope, acquired with the hybrid sensor (inclination + sEMG), improving the accuracity sense of positioning in proprioceptive analysis of the knee compared to a commercial electrogoniometer in combination with sEMG signal. The hybrid sensor also was applied to the control of a robotic wheelchair, using head movements for self-displacement of persons with tetraplegia, as well as autonomous manipulation of a mobile robot by people with cognitive and motor disabilities, which was obtained with training, whose performance in interacting with the robot was evaluated by GAS index. In the second case study, the results obtained for assessment of fatigue in people with fibromyalgia (FM)have indicated a relationship between increasing load and muscle pain, especially with 80% of MVC. The linear regression of algorithms RMS, ARV and MNF havshown in both the inclination (α ) and intercept (β) an expected trend in the control group, with positive linear relationship to characteristics in the time domain and negative characteristics to the frequency domain, with 60% MVC, and 60% of isometric segment of sEMG signal, which were obtained with 20 isotonic contractions during flexion-extension of biceps braquii (RMS α = 1.1319, β = 275 706; MNF α = -0470, β = 91 482). In the case of volunteers with FM, the N3 voluntary presented a behavior with the highest expected trend of muscular fatigue at 80% MVC and 60% of isometric segment, obtained during isotonic movement of biceps braquii (RMS α = 5.92 β = 113.33; MNF α = β = -1.21 96.96). Finally, the third case study, identified, with the MLP classifier, a success rate of 94.9% for six movements of individual fingers, including rest (category A), and 97.5% of success rate for seven movements, including: fingers, wrist and grip (category B), both cases, with a combination of features RMS, WL, MAV and ZC. On the other hand, the results obtained by amputee volunteers showed better results with features in time domain, compared to fractal dimension (DF), with success rates of 93.9% using combination RMS, WL and MAV characteristics for category A, and 95.4% of success rate with combination of RMS, WL, MAV and ZC in category B.

A comparison of the ability of different spatial filters to reduce the amount of crosstalk in a surface electromyography (sEMG) measurement was conducted. It focused on the influence of different properties of the muscle anatomy and detection system used on the amount of crosstalk present in the measurements. An analytical model was developed which enabled the simulation of single fibre action potentials (SFAPs). These fibres were grouped together in motor units (MUs). Each MU has characteristics which, along with the SFAPs, are used to obtain the motor unit action potential (MUAP). A summation of the MUAPs from all the MUs in a muscle leads to the electromyogram (EMG) signal generated by the muscle. This is the first model which simulates a complete muscle for crosstalk investigation. Previous studies were done for single fibres (Farina&Rainoldi 1999; Farina et al. 2002e; Farina et al. 2004a) or MUs (Dimitrova et al. 2002; Dimitrov et al. 2003; Winter et al. 1994). Lowery et al. simulated a complete muscle, but only investigated one spatial filter (Lowery et al. 2003a). This model is thus the first of its kind. EMG signals were generated for limbs with different anatomical properties and recorded with various detection systems. The parameters used for comparison of the recorded signals are the average rectified value (ARV) and mean frequency (MNF), which describe the amplitude and frequency components of an EMG signal, respectively. These parameters were computed for each EMG signal and interpreted to make recommendations on which detection system results in the best crosstalk rejection for a specific experimental set-up. The conclusion is that crosstalk selectivity in an sEMG measurement is decreased by increasing the thickness of the fat layer, increasing the skin conductivity, decreasing the fibre length, increasing the interelectrode distance of the detection system, placing the detection electrodes directly above the end-plate area or an increased state of muscle contraction. Varying the contraction force strength or placing the detection electrodes directly above the tendon area has no influence on the crosstalk selectivity. For most of the conditions investigated, the normal double differential (NDD) detection system results in the best crosstalk reduction. The only exceptions are a set-up with poor skin conductivity where NDD and double differential (DD) performed comparably, and the two simulations in which the muscle length is varied, where the DD filter performed best. Previous studies have found DD to be more selective for crosstalk rejection than NDD (Dimitrov et al. 2003; Farina et al. 2002a; Van Vlugt&Van Dijk 2000). Possible reasons for the contradictory results are the high value of skin conductivity currently used or influences of the muscle geometry. / Dissertation (MEng(Bio-Engineering))--University of Pretoria, 2007. / Electrical, Electronic and Computer Engineering / unrestricted

Surface electromyography records the motor unit action potential signals in the vicinity of the electrode to reveal information on muscle activation. Decomposition of sEMG signals for characterization of constituent motor unit action potentials in terms of amplitude and firing times is useful for clinical research as well as diagnosis of neurological disorders. Successful decomposition of sEMG signals would allow for pertinent motor unit action potential information to be acquired without discomfort to the subject or the need for a well-trained operator (compared with intramuscular EMG). To determine amplitudes and firing times for motor unit action potentials in an sEMG recording, Szlavik's perturbation based decomposition may be applied. The decomposition was initially applied to synthetic sEMG signals and then to experimental data collected from the biceps brachii. Szlavik's decomposition estimator yields satisfactory results for synthetic and experimental sEMG signals with reasonable complexity.

Title: The Appearance of the reflex locomotion elements in the hurdle clearance stride Thesis Objective: The main objective of this thesis is to state the frequency of the presence of reciprocal inhibition (RI) and cross extensor reflex (CI - contralateral inhibition) during hurdle overrun. Methods: The subjects were tested during running over three hurdles with modified distances similar to the 110m hurdles. The height of the hurdles was set to 0,99 m. The muscle activity was monitored by surface electromyography and video was recorded simultaneously. The hurdle clearance was divided into particular phases which altogether formed one cycle. Collected data was analyzed by the software Megawin (rectification, smoothing, and synchronization with camera) and subsequently converted into software MATLAB. Nine cycles of each subject were averaged. Such modified cycle was evaluated consequently. Results: Reflex locomotion (RI and CI) was presented more often in the group with less experience with the hurdles. Co-activation of antagonist muscles was typical for the group consisting of more experienced individuals, rather than RI and CI. Key Words: hurdles, biomechanics, sEMG, reciprocal inhibition, contralateral inhibition

Vid träning av baksida lår kan det vara svårt att förstå hur muskeln aktiveras. Genom att visualisera de myoelektriska signalerna från biceps femoris och semimembranosus till den som tränar kan personen få en bättre mind-muscle connection. I examensarbetet har två teoretiska EMG sensorer skapats och simulerats, kod har skrivits för att filtrera fyra EMG signaler samt överföra dessa över BLE, kraftsensorer är kopplade och kod är skriven för att avgöra den kraftutveckling som sker i en kontraktion av baksida lår. EMG sensorerna har jämförts med SparkFuns MyoWear muskelsensor, OpenBCI Cyton board och BioNomadix BN-EMG2-T. Båda de teoretiska lösningarna anses likvärdiga med ett billigare alternativ till Cryton Board, en flerkanalig lösning till MyoWear, BN-EMG2-T är för dyr att realistiskt implementeras till examensarbetets syfte. Simuleringarna visar att kretsarna behandlar signalen enligt tänkt sätt men det gick inte att bygga en prototyp då en pandemi har begränsat KTH:s verksamhet / One difficulty when training hamstrings is the understanding of how the muscle is activated. Through visualization of the myoelectrical signals from biceps femoris and semimembranosus to the exerciser, a better mind-muscle connection can be achieved. In this bachelor thesis, two theoretical EMG sensors were created and simulated, code to filter four EMG signals and transmit them through BLE was written, also a way to calculate how much force is applied in a hamstring curl was constructed. Both EMG sensors have been compared against SparkFuns MyoWear muscle sensor, OpenBCI Cyton Board and BioNOmadix BN-EMG2-T. The theoretical EMG sensors are interconvertible to a cheaper Cyton Board, a multichannel alternative to MyoWear, the BN-EMG2-T is too expensive to be a realistic alternative for this bachelor thesis attended purpose. Simulations show that the EMG sensors behave as intended but because of a pandemic, a prototype could not be created.

No description available.

A novel exoskeleton robotic system was developed to assist stair climbing. This active demonstrator consists of a motor with a cable system, various sensors, and a control system with a power supply. The objective of this preliminary study is a biomechanical evaluation of the novel system to determine its effectiveness in use. For this purpose, three test persons were biomechan- ically investigated, who performed stair ascents and descents with and without the exoskeleton. Kinematics, kinetics, and muscle activity of the knee extensors were measured. The measured data were biomechanically simulated in order to evaluate the characteristics of joint angles, moments, and reaction forces. The results show that the new exoskeleton assists both the ascent and the descent according to the measured surface electromyography (sEMG) signals, as the knee extensors are relieved by an average of 19.3%. In addition, differences in the interaction between the test persons and the system were found. This could be due to a slightly different operation of the assisting force or to the different influence of the system on the kinematics of the users.

Les systèmes neuromusculaire et musculosquelettique sont des systèmes de systèmes complexes qui interagissent parfaitement entre eux afin de produire le mouvement. En y regardant de plus près, ce mouvement est la résultante d'une force musculaire créée à partir d'une activation du muscle par le système nerveux central. En parallèle de cette activité mécanique, le muscle produit aussi une activité électrique elle aussi contrôlée par la même activation. Cette activité électrique peut être mesurée à la surface de la peau à l'aide d'électrode, ce signal enregistré par l'électrode se nomme le signal Électromyogramme de surface (sEMG). Comprendre comment ces résultats de l'activation du muscle sont générés est primordial en biomécanique ou pour des applications cliniques. Évaluer et quantifier ces interactions intervenant durant la contraction musculaire est difficile et complexe à étudier dans des conditions expérimentales. Par conséquent, il est nécessaire de développer un moyen pour pouvoir décrire et estimer ces interactions. Dans la littérature de la bioingénierie, plusieurs modèles de génération de signaux sEMG et de force ont été publiés. Ces modèles sont principalement utilisés pour décrire une partie des résultats de la contraction musculaire. Ces modèles souffrent de plusieurs limites telles que le manque de réalisme physiologique, la personnalisation des paramètres, ou la représentativité lorsqu'un muscle complet est considéré. Dans ce travail de thèse, nous nous proposons de développer un modèle biofidèle, personnalisable et rapide décrivant l'activité électrique et mécanique du muscle en contraction isométrique. Pour se faire, nous proposons d'abord un modèle décrivant l'activité électrique du muscle à la surface de la peau. Cette activité électrique sera commandé par une commande volontaire venant du système nerveux périphérique, qui va activer les fibres musculaires qui vont alors dépolariser leur membrane. Cette dépolarisation sera alors filtrée par le volume conducteur afin d'obtenir l'activité électrique à la surface de la peau. Une fois cette activité obtenue, le système d'enregistrement décrivant une grille d'électrode à haute densité (HD-sEMG) est modélisée à la surface de la peau afin d'obtenir les signaux sEMG à partir d'une intégration surfacique sous le domaine de l'électrode. Dans ce modèle de génération de l'activité électrique, le membre est considéré cylindrique et multi couches avec la considération des tissus musculaire, adipeux et la peau. Par la suite, nous proposons un modèle mécanique du muscle décrit à l'échelle de l'Unité Motrice (UM). L'ensemble des résultats mécaniques de la contraction musculaire (force, raideur et déformation) sont déterminées à partir de la même commande excitatrice du système nerveux périphérique. Ce modèle est basé sur le modèle de coulissement des filaments d'actine-myosine proposé par Huxley que l'on modélise à l'échelle UM en utilisant la théorie des moments utilisée par Zahalak. Ce modèle mécanique est validé avec un profil de force enregistré sur un sujet paraplégique avec un implant de stimulation neurale. Finalement, nous proposons aussi trois applications des modèles proposés afin d'illustrer leurs fiabilités ainsi que leurs utilité. Tout d'abord une analyse de sensibilité globale des paramètres de la grille HDsEMG est présentée. Puis, nous présenterons un travail fait en collaboration avec une autre doctorante une nouvelle étude plus précise sur la modélisation de la relation HDsEMG/force en personnalisant les paramètres afin de mimer au mieux le comportement du Biceps Brachii. Pour conclure, nous proposons un dernier modèle quasi­ dynamique décrivant l'activité électro-mécanique du muscle en contraction isométrique. Ce modèle déformable va actualiser l'anatomie cylindrique du membre sous une hypothèse isovolumique du muscle. / The neuromuscular and musculoskeletal systems are complex System of Systems (SoS) that perfectly interact to provide motion. From this interaction, muscular force is generated from the muscle activation commanded by the Central Nervous System (CNS) that pilots joint motion. In parallel an electrical activity of the muscle is generated driven by the same command of the CNS. This electrical activity can be measured at the skin surface using electrodes, namely the surface electromyogram (sEMG). The knowledge of how these muscle out comes are generated is highly important in biomechanical and clinical applications. Evaluating and quantifying the interactions arising during the muscle activation are hard and complex to investigate in experimental conditions. Therefore, it is necessary to develop a way to describe and estimate it. In the bioengineering literature, several models of the sEMG and the force generation are provided. They are principally used to describe subparts of themuscular outcomes. These models suffer from several important limitations such lacks of physiological realism, personalization, and representability when a complete muscle is considered. In this work, we propose to construct bioreliable, personalized and fast models describing electrical and mechanical activities of the muscle during contraction. For this purpose, we first propose a model describing the electrical activity at the skin surface of the muscle where this electrical activity is determined from a voluntary command of the Peripheral Nervous System (PNS), activating the muscle fibers that generate a depolarization of their membrane that is filtered by the limbvolume. Once this electrical activity is computed, the recording system, i.e. the High Density sEMG (HD-sEMG) grid is define over the skin where the sEMG signal is determined as a numerical integration of the electrical activity under the electrode area. In this model, the limb is considered as a multilayered cylinder where muscle, adipose and skin tissues are described. Therefore, we propose a mechanical model described at the Motor Unit (MU) scale. The mechanical outcomes (muscle force, stiffness and deformation) are determined from the same voluntary command of the PNS, and is based on the Huxley sliding filaments model upscale at the MU scale using the distribution-moment theory proposed by Zahalak. This model is validated with force profile recorded from a subject implanted with an electrical stimulation device. Finally, we proposed three applications of the proposed models to illustrate their reliability and usefulness. A global sensitivity analysis of the statistics computed over the sEMG signals according to variation of the HD-sEMG electrode grid is performed. Then, we proposed in collaboration a new HDsEMG/force relationship, using personalized simulated data of the Biceps Brachii from the electrical model and a Twitch based model to estimate a specific force profile corresponding to a specific sEMG sensor network and muscle configuration. To conclude, a deformableelectro-mechanicalmodelcouplingthetwoproposedmodelsisproposed. This deformable model updates the limb cylinder anatomy considering isovolumic assumption and respecting incompressible property of the muscle.

This thesis has evaluated the use of Independent Component Analysis (ICA) on Surface Electromyography (sEMG), focusing on the biosignal applications. This research has identified and addressed the following four issues related to the use of ICA for biosignals: • The iterative nature of ICA • The order and magnitude ambiguity problems of ICA • Estimation of number of sources based on dependency and independency nature of the signals • Source separation for non-quadratic ICA (undercomplete and overcomplete) This research first establishes the applicability of ICA for sEMG and also identifies the shortcomings related to order and magnitude ambiguity. It has then developed, a mitigation strategy for these issues by using a single unmixing matrix and neural network weight matrix corresponding to the specific user. The research reports experimental verification of the technique and also the investigation of the impact of inter-subject and inter-experimental variations. The results demonstrate that while using sEMG without separation gives only 60% accuracy, and sEMG separated using traditional ICA gives an accuracy of 65%, this approach gives an accuracy of 99% for the same experimental data. Besides the marked improvement in accuracy, the other advantages of such a system are that it is suitable for real time operations and is easy to train by a lay user. The second part of this thesis reports research conducted to evaluate the use of ICA for the separation of bioelectric signals when the number of active sources may not be known. The work proposes the use of value of the determinant of the Global matrix generated using sparse sub band ICA for identifying the number of active sources. The results indicate that the technique is successful in identifying the number of active muscles for complex hand gestures. The results support the applications such as human computer interface. This thesis has also developed a method of determining the number of independent sources in a given mixture and has also demonstrated that using this information, it is possible to separate the signals in an undercomplete situation and reduce the redundancy in the data using standard ICA methods. The experimental verification has demonstrated that the quality of separation using this method is better than other techniques such as Principal Component Analysis (PCA) and selective PCA. This has number of applications such as audio separation and sensor networks.