Optimized surface electromyogram (sEMG) signal processing and system identification for smart prosthetic hand control

There are approximately 1.7 million people in the U.S living with limb loss. It is estimated that one out of every 200 people has had an amputation. One might think that with the advances in the fields of medicine and engineering, we would have by now, an "ideal" prosthetic device to help improve the quality of life of amputees. But the stark reality is that we are very far away from achieving that goal. Designing and implementing a fully functional upper-extremity prosthesis is no simple task. This dissertation addresses one of the areas of prosthetic hand design: control of the prosthetic hand using surface electromyogram signals (sEMG), specifically extracting useful information from the sEMG signal to estimate finger force. The choice to use surface EMG signal was made to eliminate invasive signal extraction and to keep the overall cost of the prosthetic low (avoiding surgical implantation of EMG sensors at motor unit locations). Using the signal generated by the skeletal muscles, for controlling the prosthesis, is helpful as the user feels more comfortable using the device. But one of the main disadvantages of using sEMG is that the signal is corrupted by signal from other motor units tiring around the measurement site. Various filters like the Bessel, Butterworth, Chebyshev type I, Chebyshev type II, nonlinear Bayesian filter. Poisson's nonlinear filter and the Half-Gaussian filters were extensively studied and optimized using genetic algorithm (GA) and a hybrid particle swarm optimization with Tabu search (PSO-TS). The objective to optimization was to improve the accuracy of estimation of the sEMG signal and the corresponding forces generated. The entire "system" was modeled as a black box. System identification was used to deduce mathematical models. The most extensively and successful model which captured the signal dynamics was the Hammerstein-Wiener models. Another study was also carried out to optimize the Hammerstein-Wiener model parameters. The research also addresses use of spatial filtering methods for multiple sensor configurations (3x3 sensor array). The spatial filter masks were also optimized using GA to improve the accuracy of estimation. In summary, this research provides a new perspective for developing improved control design algorithms, based on improved signal processing methods, which in turn would improve the control of prosthesis allowing greater flexibility and dexterity to the existing deigns or, for new designs. And, in the process help researchers to get a little bit closer in designing that "ideal" prosthetic hand.