Validation of a new design of dry electrodes for applications in hand prosthetic control using HD-sEMG as a source signal

Current Upper Limb Prostheses (ULPs) are characterized by significant shortcomings that contribute to user dissatisfaction, including limited comfort, excessive weight, and functional limitations. To address the latter issue, a new solution, RPC-Net (Recursive Prosthetic Control Network), was developed in a previous study. RPC-Net translates High-Density surface electromyography (HD-sEMG) signals into hand position, aiming to enhance user experience by providing a more natural control. The aim of this thesis is to further validate RPC-Net as a practical solution and this objective is carried out in two parts. The first part describes a novel dry electrode array design, named the High-Density Electrode Array (HDE-Array). The implementation of dry electrodes aims to offer a more practical alternative to the gel electrodes used in previous work, which are unsuitable for real-life applications due to their setup time and material costs. However, the introduction of dry electrodes may lead to lower performance of RPC-Net, as gel electrodes provide superior signal quality due to better skin contact. To validate the HDE-Array as a viable alternative, the following hypothesis is to be tested: the performance of RPC-Net using signals acquired by the HDE-Array is comparable to that achieved with traditional gel electrodes. The second part aims to verify the hypothesis that RPC-Net is robust to changes in electrode positions and skin condition. Evaluating its robustness is crucial to affirm RPC-Net's suitability for applications in daily life, highlighting its capacity to maintain functionality in real-world scenarios. The second hypothesis is addressed by comparing the performance of RPC-Net when trained and tested on data acquired in the same day to its performance when tested on data acquired on following days. In both cases, the network's performance was assessed using two indicators: Mean Distance (MD in mm) and Mean Pearson's Correlation Coefficient (MPCC in %). Under each hypothesis, t-tests were conducted for both metrics with a significance level of α = 0.05. The results indicate that RPC-Net, when used with the HDE-Array, achieved performance comparable to that observed with the gel electrode setup. Improvements were noted in both mean values of performance indicators when using dry electrodes, with a decrease in MD by 2.89 mm and an increase in MPCC by 2.98%. The t-test results, however, suggest RPC-Net performs better with gel electrodes. Nevertheless, the obtained p-values (p-value for MPCC = 0.072; p-value for MD = 0.058) are close to the set alpha value, indicating that the HDE-Array is indeed a viable alternative. The second hypothesis of this study is rejected. The robustness of RPC-Net could not be demonstrated, as its performance on data acquired on a second day showed a significant decline, with differences of 37.8 mm for MD and 28.11% for MPCC. The t-tests strongly confirmed the null hypothesis, indicating inferior performance of RPC-Net when using EMG signals from the second day (p-values approaching 1). In conclusion, this study underscores the potential of HDE-Array and RPC-Net to revolutionize prosthetic control, offering a more practical, efficient, and user-friendly alternative to current available technologies for the upper limb. However, additional improvements are required to make RPC-Net robust to electrodes shift and changes in skin condition