Development of electromyography-controlled 3D printed robot hand and supervised machine learning for signal classification

Developing a device which resembles the human hand called Anthropomorphic Robotic Hand (ARH) has become a relevant research field due to the needs for the purpose to help the amputees to live their life as normal people. However, the current research state is unsatisfactory, especially in terms of structural design, robot system and the robot control method. In this research, an EMG controlled 3D printed robot hand prototype with improved features and advance hand posture classification method based on EMG signal pattern were proposed. The current state of the robot hand structure development, the structure features do not resemble the human hand functionality with less durability and poor movement capability. In this research, the structural design of the robot hand with five individual actuated fingers and tendon-driven actuator mechanism was designed using the Inventor Professional 2018 software and fabricated it using 3D printing technology. The durability and movement capability of the structure were evaluated through Static analysis (simulation), and validate it through the Load test and Motion capture analysis. As a result, the hand robot structure which made from PLA material can withstand load with 1.5kg while the structure made from ABS material only able to withstand load with 1.4kg in the Static Analysis. After that, the simulation results were validated in the Load Test, and it shows that structure made from PLA material was able to withstand load with 1.7kg while the structure made from ABS material is only able to withstand load with 1.6kg. The result obtained from these experiments shows that the structure made from PLA has better durability than ABS. In another hand, the movement accuracy analysis of the hand robot motion range was performed by comparing the expected motion range and the motion range achieved by 3D printed hand robot. The comparison shows that the similarity percentage achieved is about 72.62% - 98.43%. The accurate motion range and the decent durability were able to achieve by improving the structural design with the tendon-driven actuator mechanism. In the system development aspect, the electromyography (EMG) sensors were applied as the main control interface of the system which used to control the hand robot movement transparently to perform the tasks given. The electronic hardware and hand robot structure were integrated to develop an EMG controlled hand robot prototype, and its functionality was tested through three stages: muscular activity detection only, object detection only and the integration of both detection in an algorithm to control the hand robot structure movement to perform opened hand palm and some grasping postures with two trial for each stage. The tasks were performed without any failure and show the developed robot hand is reliable. Furthermore, the Support vector machine (SVM) and Linear discriminant analysis (LDA) machine learning for the hand posture classification based on the EMG signal pattern were investigated and compared in term of classification performance. The current study of the hand posture classification requires a higher number of EMG sensor used to achieve an accurate classification performance that leads the system to be complicated. In this research, the LDA gives as higher as 85.8% of accuracy with six units of the sensors used compared to SVM which is 85% of accuracy percentage with five units of the sensors used. However, the EMG signal pattern classification was done by SVM has better performance than LDA due to less significant difference in the accuracy percentage, and a fewer number of sensors used by the SVM. The result was achieved with K=15 of fold cross-validation, without PCA and five EMG sensors used that located on the Extensor carpi ulnaris, Extensor digitorum, Extensor carpi radialis, Flexor carpi ulnaris, and Flexor digitorum superficial muscles. In conclusion, the electromyography controlled hand robot prototype was successfully developed with improved features, optimal structural durability, higher accurate movement capability, reliable system and lower number of sensor used with higher accuracy of the signal pattern classification.

Text FK 2020 48 ir.pdf Download (1MB)

Actions (login required)

View Item