Etd

Proportional, Simultaneous and Independent Control (PsiCon) of Upper-Limb Prostheses for the Limb-Absent Population

Public

Downloadable Content

open in viewer

The bioelectrical signal generated by muscle activities is the electromyogram (EMG). EMG is a powerful tool utilized in various medical assessments, rehabilitation, and myoelectric control of prostheses and orthoses. Continuing previous study by our group, this dissertation focused on three main topics in upper-limb myoelectric prosthesis control: offline testing of EMG signals and force control under mirrored movement, including direct mechanical comparison of bilateral forces; online testing of prosthesis control performance under different real-time control strategies; and online performance during 2-degree of freedom (DoF) dynamic target tracking. The first topic of the thesis was a study of EMG-force and EMG-target performance during hand-wrist contraction in both able-bodied and limb-absent subjects in force-varying tasks, especially in bilateral mirror tasks. In order to solve the problem that limb-absent subjects have no end-effector force available as the supervisory output in EMG-force models, different kinds of feedback, such as contralateral force and target movement, were provided as alternatives in EMG-force models. We initially tested these alternative models offline on able-bodied subjects and compared to the true EMG-force model which we could measure only on able-bodied subjects, then applied the results to limb-absent subjects to contrast performance. The benchmark errors of relating EMG to dominant force on able-bodied subjects were ~10 percent maximum voluntary contraction (%MVC). Errors when using either contralateral force or target movement in bilateral tracking as alternative outputs were 12–16 %MVC, while no-feedback in unilateral tracking produced errors of 25–30 %MVC. This project explored the influence of different feedback conditions on the models, and effectively provided some calibration protocols than can be adopted in actual prosthetic control. This first topic was extended to document normative force/moment tracking performance during bilateral symmetric hand-wrist tasks. In this aspect of the project, we measured the force /moment from both limbs to test the tracking accuracy under different conditions. This project provided some useful results of how intrinsic characteristics and external visual conditions jointly influenced bilateral synergies on able-bodied subjects, and thus, to better provide data support for the study of limb-absence device control from a biological perspective. The second project involved online testing of prosthesis control to complete standardized (SHAP) tests for both able-bodied and limb-absent subjects. The project integrated the design of prosthesis hardware, software, algorithms and thresholding. Three different control strategies were studied: conventional 2-site control (2 electrodes) with co-contraction switching, 2-DoF direct control (6 or 12 electrodes) and 2-DoF mapping control (6 or 12 electrodes) during three tasks: 1-DoF box-block test, 2-DoF clothespin refined test and 2-DoF doorknob test. The project investigated the actual ability of limb-absent subjects to control a multi-DoF prosthesis, and the superiority of new 2-DoF control algorithms over traditional control methods. The third project explored the feasibility of 2-DoF EMG-force models for the wrist only, and also the wrist-hand, and found the minimum number of electrodes required for hand-wrist control in both 1-DoF and 2-DoF tasks by backward stepwise selection. The results showed as few as two electrodes for 1-DoF and four electrodes for 2-DoF could achieve the best performance, with an average RMS error of 6.0–16.3 %MVC for wrist only, and 8.3–9.2 % MVC for wrist and hand. Other co-participated projects complemented the above projects. The first project focused on efficient training of 2-DoF EMG-force models, finding that a minimum training data length of approximately 40–60 s is the best, and proposed a new universal EMG-force filter across all subjects (as an alternative to subject-specific models or DoF-specific models), which was 15–21% better than original subject-specific models. The second project was a transition between offline testing and prosthesis control. Subjects controlled the cursor on the screen to reach different target position. The count of the number of correct target acquisitions, overshoot rate, path efficiency and stability for each DoF combination were tested to exhibit the performance of real-time 2-DoF control. The third project explored optimization of EMG processing in additive noise with root-difference-of-squares (RDS) and whitening methods. In summary, this thesis provides a systematic study on prosthesis control. It starts with offline EMG-force model and optimization for able-bodied subjects. It then investigates the problems which limb-absent subjects face with respect to alternative feedback selection in bilateral tasks design. Finally, it tested algorithm performance when subjects directly used a prosthesis to complete some daily tasks.

Creator
Contributors
Degree
Unit
Publisher
Identifier
  • etd-21671
Advisor
Orcid
Committee
Defense date
Year
  • 2021
Date created
  • 2021-05-03
Resource type
Rights statement

Relations

In Collection:

Items

Items

Permanent link to this page: https://digital.wpi.edu/show/8p58ph013