Study and Analysis of SEMG Signal for Enhancement of Above Shoulder Myoelectric Arm Functionality

Abstract

Upper limb amputation is a traumatic event that can seriously affect the person’s capacity to perform regular tasks and can lead individuals to lose their confidence and autonomy. Prosthetic devices can give relief by acting as substitute the function of missing limb which can help to improve the quality of life of the amputees. In recent years, myoelectric devices have received extensive attraction to provide enhanced degree of freedom over traditional devices. Myoelectric prosthesis is controlled via the acquisition and processing of electromyogram signal produced at the muscles fibre from the surface of body with an array of electrode placed on the residual limb. The acquired signal is a complex one being dependent on the physiological and anatomical property of muscles. The electrodes convert muscles-activity from the torso into information that can be processed by different techniques. The unwanted noise contributes from the electrolyte skin surface, while travelling through the muscles. To make the noisy signal useful, advancement in the detection and processing of the signal becomes a very important requirement in biomedical engineering. The signal has to undergo pre-processing stage consisting of amplification, filtering and adaptive peak detection etc. to reduce the noise level in the raw signal. The different signal processing techniques such as time domain techniques, wavelet coefficients and autoregressive coefficients have been applied to increase the information yield from the EMG signal. Different algorithms to identify the intended movements are available that rely on the feature extraction that provide the user with access to multiple degrees of freedom and have shown great promise in research literature. The identified information of movements is translated into control signal to drive the artificial limb and the force generated by the artificial limb can be varied by the user’s muscles intensity.

The commercially available myoelectric prostheses do not allow to control the transhumeral level and shoulder disarticulation level of amputation. For transhumeral amputee with no muscles-activity or very less muscles-activity in the residual limb there is no intuitive control source for either elbow or hand, therefore controlling the prosthetic device is impossible with existing techniques. As a result, better strategy is required to control a prosthesis for a high-level amputation. Further studies are required to improve the training protocol and analysis of the signal for development of the prosthetic devices for these applications. The main contribution of this thesis to implement the prosthesis based on the EMG signal from the set of shoulder muscles intact in the transhumeral amputee. For this, an overview of the human shoulder muscles anatomy was carried out. The acquired SEMG data with the different shoulder movement are described. Various pre-processing techniques and adaptive peak detection techniques were explored. A new threshold method was developed and applied to filter the unwanted peaks in the pre-processing stage of the SEMG signal. Next step was to investigate the shoulder movements of amputees and non-amputees and compare the EMG activity based on the amplitude level of the signal. Different signal analysis methods such as Fourier transform, short time Fourier transform and wavelet transform were investigated and wavelet transform was applied successfully. The proposed detection scheme focuses on the discrete Daubechies wavelet transform with four decomposition levels. A systematic approach for selecting the optimal wavelet transform method was proposed and demonstrated. A pattern based recognition technique was used with immediate access to four different movements at a time. A set of features was extracted from shoulder muscles of the transhumeral amputee by transforming the wavelet reconstructed coefficients to the new transformed coefficients by using the new proposed transformation method. Subsequently, investigation of classification is presented through various experiments conducted on amputees. The work was carried out with the aim to enhance the robustness in the pattern recognition system to classify different shoulder movements so that it is helpful for making a more reliable and useful device. To classify the different shoulder motions, various machine learning algorithms were compared to select the optimal and efficient algorithm. A data mining Random forest classifier was utilized in this work which was found better than other classifiers. These classification results were evaluated and validated by a prototype using Arduino motor controller for elbow and hand movement.

Myoelectric signal is one of the control signals for controlling the powered prosthesis. A number of commercial products have been developed for these prostheses. But these devices are still insufficient to satisfy the needs of amputee. The precise measurement and analysis of human movements and muscles activity are essential processes in rehabilitation. In India, the maximum amputation rate is for below elbow which is about 52% of the total amputation. Transhumeral amputation is the second largest in upper limb amputation which is about 24% the total. The available literature and systems mostly focus on below elbow amputees and limited work is available on prosthetic design for the shoulder amputees. The researchers have largely ignored the real time SEMG signal for the transhumeral amputees with no activation in the triceps and biceps muscles with the result that a prosthetic device for such amputees is not available off the shelf. The system developed in this work is based on the data from the amputee's muscles activation from different arm movements which allows one to have independent signals required for independent motion of elbow and hand. The system performance has been validated through an actual arm which will be worn by the amputee. The performance of arm has been checked under actual environment. The developed system promises an overall accuracy of more than 90% for correct motion of elbow and hand which shows that its functionality is very close to natural human arm.

This work presents a successful design of an affordable SEMG platform. The effort to achieve this goal will always encourage the researchers into the field of SEMG technologies. Mechanical fabrication using light composite material and electronic assembly using advanced processors can be implemented to make the arm ready for large scale clinical trials. If required, high mechanical functionality in electrode grid to follow the skin surface can be achieved by connecting the wireless sensor to produce better results. High end DSP chip system can be used to handle the large data set and the code that presently exists in MATLAB. The use of energy efficient motors, driving circuits and couplings can further improve the degree of functionalities of the arm.

The main idea of this research is to provide a set of guidelines for researchers and engineers aiming to develop their own low-cost EMG systems applicable in biomechanical, clinical, rehabilitation, sport, and research contexts. EMG signals can be used to generate device control commands for rehabilitation equipment such as robotic prostheses and can be useful in many clinical and industrial applications.