Objekt-Metadaten
Forearm surface EMG signals recognition and muscoloskeletal system dynamics identification using intelligent computational methods

Autor :Abdelhafid Zeghbib
Herkunft :OvGU Magdeburg, Fakultät für Elektrotechnik und Informationstechnik
Datum :07.12.2007
 
Dokumente :
Dataobject from HALCoRe_document_00006568
 
Typ :Dissertation
Format :Text
Kurzfassung :The specific aim of this thesis, which considers the biological system "human movement", is presented in two parts. The first part considers the recognition (classification) of measured Electro-myography (EMG) signals of forearm muscles corresponding to hand movements. The second part treats the musculoskeletal system, which is considered by Knee-joint dynamics under non-isometric conditions, in terms of its measured angle between thigh and shank as response for Functional Electrical Stimulation (FES) impulses. This procedure is known as system-identification. Understanding human movement functions is of a great importance in the domain of neuromusculoskeletal systems and biomechanics. Whole-body movement is achieved with help of the interaction between the neuromuscular control signal and musculoskeletal dynamics system. Many elements of the neuromusculoskeletal system interact to enable smooth and coordinated movements. The skeletal system, composed of bones and joint connections with muscles, which complete the musculoskeletal system, apply the necessary driving forces for movement realisation. In this thesis we will focus on human Neuromuscular control signals classification and Musculoskeletal dynamics identification. These complex systems require much knowledge by learning. Hence an improvement of the learning ability, using artificial intelligent methods, is also covered. The information, in the form of nerve impulses, figure 2, travels to and from our central nervous system (brain) along our spinal cord, allows us to coordinate our voluntary movements of our body. Brain electrical impulses, which are transmitted via nerve cells to the muscles, cause the movement of these muscles. These muscles respond by contracting when the brain's electrical signals reach them. These electrical signals can be measured over muscles and they are called electromyography (EMG) signals. Generation of human movement is a complex process, involving the following ways: neural command, neuromuscular signals and finally muscle force. This thesis considers the two last components of movement realisation, which are: 1) EMG Neuromuscular recruitment signals recognition (classification), and 2) Musculoskeletal loading dynamics identification. In this thesis the causal relationships between neuromuscular EMG signals and musculoskeletal dynamics will not be considered. Each part is considered alone. The first goal, is to recognise and classify the EMG neuromuscular signal. This neuromuscular signal represents the Motor Unit Action Potentials (MUAPs) of nerves. The summation of the muscle fiber action potentials from all muscle fibers can be measured with help of electrodes placed on the corresponding muscle as electromyography (EMG) signal. An on-line Algorithm for this part of EMG signals recognition is also developed. The second goal of this thesis is to identify musculoskeletal structure dynamics, which act as actuators producing the joint torques to drive the body (legs and arms). Such movements can be produced using Functional Electrical Stimulations (FES), if the dynamics between FES and joint torques are known. Although this part of study focuses on walking, using quadriceps muscles, the findings can be generalised to other motor control systems such as elbow joint through biceps and triceps muscles. A better understanding of these two components of movement realisation dynamics (musculoskeletal load and neuromuscular recruitment) can help disabled persons in regaining lost function and/or improving their activity of daily living life and for assessing rehabilitation progress. These two components have been studied in this thesis separately. Developing techniques for investigating the relationship between them, in further work, will be of great importance. Such relationship can be illustrated by using the recognition of detected weak voluntary muscle activity, by post-stroke subjects, through electromyography signals (EMG) to control Functional Electrical Stimulation (FES) impulses, which will support the patients to accomplish correct leg or arm movements. These techniques help the investigating of the relationship between the mechanics of movement and the characteristics of the EMG signals The domain of engineers provides efficient technical approaches and tools for biosignals processing and complex dynamic systems identification as muscle, which is the generator of all human body movements. Soft computing includes both neural networks (NN) and fuzzy logic (FL) systems represent intelligent approaches, which are used in this thesis for solving the identification and classification problem of such realistic complex systems in biomedical area. These EMG signals acquired from muscles, through surface electrodes, require advanced computational methods as acquisition, analysis, decomposition, and classification. The purpose of this part is to illustrate the various methodologies and algorithms for all necessary steps used to discriminate the different movements of finger and hand grasps according to their corresponding EMG signals. For the recognition and classification of these EMG signals, a fuzzy-classifier-model algorithm is proposed in this thesis. This classifier-model algorithm, Fuzzy Trimmed Mean Classifier (FTMC) uses the trimmed mean method as tool for input space-set initialisation. The results of this algorithm are compared with other known intelligent computational methods. This first part contains the development of all procedures, starting from EMG signals acquisition till the recognition of their corresponding hand/finger movements, using extraction of relevant features and their classification. The main goal of this first part is to help the patient with the amputated hand to keep the neuromuscular activity of forearm muscles, which will be used to manipulate a myoprosthesis, and to keep the virtual neural activity of the brain related also to this activity of forearm's motor unit potentials. The second component of movement realisation dynamics, which is musculoskeletal dynamics has a great importance. These dynamics are very complex, hence we should look for an effective method that can model this complex motor system. Mathematical modelling methods-based morphological models cannot describe with fidelity such complex dynamics. For this problem an effective and fast hybrid fuzzy Algorithm for modelling is developed and proposed in this thesis. The quadriceps muscles are used because their dimension, which help to choose the desired muscle to be stimulated. The choice of desired muscle to be stimulated is not possible in case of many small muscles that are located together. The parameters of this hybrid fuzzy identifier model are obtained using generated Functional Electrical Stimulation (FES) impulses as an input set, and the measured knee-joint angle as an output set. The efficiency of this fuzzy identifier-model representing non-linear input-output dynamics depends on the fuzzy partition of its input-space (the initialisation of premise fuzzy sets is an important issue in fuzzy modeling). Hence Rapid Prototyping method is introduced in this proposed algorithm to perform this initialisation of premise fuzzy sets. In this proposed algorithm three techniques: Rapid Prototyping algorithm, Gradient Descent method and Least Squares Estimator are combined as a hybrid algorithm to achieve this modelling task. The main issues of this study, concern the knee-joint dynamics identification, are developed for further control-application of the human knee-joint movements by Functional Electrical Stimulation (FES).
Schlagwörter :Biosignals processing, Biological Nonlinear and time-varying systems identification, Electomyograph signals recognition
Pattern classification, Fuzzy logic and neural networks methods
Rechte :Dieses Dokument ist urheberrechtlich geschützt.
Größe :248 S.
 
Erstellt am :03.04.2009 - 08:43:36
Letzte Änderung :22.04.2010 - 09:07:32
MyCoRe ID :HALCoRe_document_00006568
Statische URL :http://edoc.bibliothek.uni-halle.de/servlets/DocumentServlet?id=6568