ResearchWorks Archive

Walking mode classification through myoelectric and inertial sensors for transtibial amputees

Show simple item record

dc.contributor.advisor Hahn, Michael E en_US Miller, Jason Daniel en_US 2012-09-13T17:36:40Z 2013-03-13T11:04:56Z 2012-09-13 2012 en_US
dc.identifier.other Miller_washington_0250O_10472.pdf en_US
dc.description Thesis (Master's)--University of Washington, 2012 en_US
dc.description.abstract Myoelectric algorithms have the potential to provide motion intent information to a prosthetic system to allow the prosthetic to adapt biomimetically to changes in walking modality. While myoelectric algorithms have been explored extensively for upper limb amputees, little has been done to achieve similar advances for the lower limb. For upper limb amputees, it has been shown that fusion of myoelectric and inertial sensors benefits classification. Myoelectric signals from four muscle groups (tibialis anterior, medial gastrocnemius, vastus lateralis, and biceps femoris) and inertial signals from the lower limb segments were recorded for five non-amputee subjects and five transtibial amputees over a variety of walking modes: level ground walking at various speeds, ramp ascent/descent, and stair ascent/descent. These signals were decomposed into relevant features (mean absolute value, variance, wavelength, number of slope sign changes, and number of zero crossings, maximum, minimum) and used to test the ability of myoelectric classification algorithms for transtibial amputees using either Linear Discriminant Analysis (LDA) or Support Vector Machine (SVM) classifiers. The fusion of myoelectric and inertial signals showed no benefit for amputee subjects and as a result inertial signals were removed from consideration. Detection of all seven walking modes were observed to have an accuracy of 97.9% (± 0.22) and 97.9% (± 1.39) for amputee subjects when using LDA and SVM, respectively. The predominant misclassifications occurred between different walking speeds due to the similar nature of the gait pattern. Stair ascent/descent modalities had the best classification accuracy with 100% (± 0.00) and 99.8% (± 0.29) for amputee subjects when using LDA and SVM, respectively. Transitions into and out of stair modalities could be predicted one gait cycle ahead of the change for 95.0% and 90.0% of all transitions observed in this study when using LDA and SVM, respectively. The robustness of the developed classifier was explored through tests of generalizability and stability under electrode shifting disturbance. While separations between certain modes were not possible through a generalized model or under disturbance, stair ascent/descent classification against all other modalities was consistent. This result highlights the possibility of a stair mode myoelectric control algorithm for lower limb amputees. Future efforts will further explore the robusticity of myoelectric classification under real world conditions such as stability of classification over long periods of time, classifier training methods, and real time capabilities of the classifier. en_US
dc.format.mimetype application/pdf en_US
dc.language.iso en_US en_US
dc.subject electromyography; machine learning; myoelectric control; prosthetic design; transtibial amputee en_US
dc.subject.other Mechanical engineering en_US
dc.subject.other Computer science en_US
dc.subject.other Biomedical engineering en_US
dc.subject.other Mechanical engineering en_US
dc.title Walking mode classification through myoelectric and inertial sensors for transtibial amputees en_US
dc.type Thesis en_US
dc.embargo.terms Restrict to UW for 6 months -- then make Open Access en_US

Files in this item

This item appears in the following Collection(s)

Show simple item record

Search ResearchWorks

Advanced Search


My Account