Student Projects
Contact Information
University: University of Colorado - Mechanical Engineering
Team Member(s): Jacob Segil
Faculty Advisors: Richard F. ff. Weir, Derek Reamon
Email Address: Jacob.Segil@colorado.edu
Project Information
Title: Design of a Myoelectric Prosthetic Hand Controller Based on Principal Component Analysis
Description:
This project developed a myoelectric prosthetic hand controller that is dimensionally optimized using Principal Component Analysis (PCA). Also, a 3D virtual hand was built with 15 joints that are controllable in real time using muscle activation signals measured using surface electromyography.
Products:
Labview 2010, NI USB-6008, Motion Control ProControl2 electrodes
The Challenge:
In 2005, an estimated 41,000 people were living in the United States with major upper limb loss. Available upper-limb prostheses enhance the abilities of this population with varying degrees of success. Current state of the art upper-limb prostheses use myoelectric control systems. Myoelectric control uses myoelectric signals (MES) generated as a by-product of normal muscle contraction as a control signal for an actuated prosthesis. Today there is a technological bottleneck in the communication between the user and the prosthesis in myoelectric control systems (Figure 1). This bottleneck limits the functionality and intuitiveness of the prosthesis. This project investigates how to overcome the bottleneck between the user and the prosthesis by developing an advanced myoelectric prosthesis controller.
Figure 1. -Comparison between able-bodied and myoelectric control systems
Currently available prosthetic hands cannot be fully utilized because there are fewer control inputs than the number of degrees of freedom (i.e. –joints) that need to be controlled. Based on work from neuroscience literature it was found that grasping tasks involving a number of everyday items could be described by only two patterns of joint movement (derived from Principal Component Analysis). This implies that hand postures involving 15 joints can be controlled using only two degrees of control. Therefore, a prosthetic hand controller can drive grasping postures with only two independent control signals. This is an encouraging finding since current clinical practice indicates two,or three, independent myoelectric signals can be located on the residual limb of a typical person with a transradial amputation (the loss of an arm below the elbow).
This proposed research will advance the field of neural prosthetics by increasing the functionality of myoelectric prostheses. Today, many amputees prefer to use a body-powered hook as opposed to myoelectric prosthetic hand. Current myoelectric hands fail to produce the desired functionality and intuitiveness. For example, a typical myoelectric hand setup today is a pinching hand with a wrist rotator. The hand can only open and close while the wrist can only rotate. In addition, the closing of the hand and the rotation of the wrist cannot occur at the same time. A PCA-based prosthesis controller augments current prosthetic hands by providing a control system that can continuously morph into various postures. This is a significant advancement in the field of neural prosthetics.
The Solution:
This work progressed in several distinct steps as described by Figure 2. The real-time acquisition and processing of (EMG) signals was developed using standard of care two-site myoelectric control schemes. The PCA algorithm was derived to produce 15 joint angles of the hand. A virtual hand with 15 degree of freedom and anthropomorphic size (50th percentile male) was designed to be controlled in real time. All of the development was accomplished using Labview 2010. The following sections discuss the design and development of the PCA-based myoelectric hand controller in more detail.
Figure 2. - PCA Controller project block diagram
EMG Acquisition and Processing
The design and development of a real-time EMG acquisition system was modeled after current standard of care procedures. The standard of care EMG acquisition scheme for a subject with transradial amputation is a two-site myoelectric control system. This is implemented in this controller. The raw EMG signals were filtered and rectified using typical EMG processing technique (band pass filtering, gain, rectification, and enveloping) and as shown in Figure 3. The EMG signals were measured by Motion Control ProControl2 surface electrodes and then acquired using a NI USB-6008.
Figure 3. - Typical EMG Processing Technique
LabVIEW streamlined this process significantly. As shown in Figure 4, multiple Express VI’s provided efficient and effective coding methods to filter the raw EMG signals and envelope the extremely noisy signal. The DAQ Assistant acquired the raw EMG signals. Then a series of filters were applied to the signal using the Filtering Express VI. The signal was first band passed, then a notch filtered at 60Hz to erase any environmental noise, and finally enveloped by using a sliding average. After the EMG signals were processed a thresholding algorithm was applied using the for-loop structure shown in Figure 4.
Figure 4. – EMG Acquisition Block Diagram
PCA Algorithm and Joint Angle Calculation
Following the EMG acquisition and pro cessing, an inverse Principal Component Analysis (PCA) was performed based on empirical grasping data. PCA is a specific type of dimensionality reduction that has properties most advantageous for the application of myoelectric control. The goal of PCA is to compute the most meaningful basis of the dataset . The most meaningful basis vectors (PC1 and PC2) are the axes that correspond to the maximum variance in the dataset as shown in Figure 5. This definition is a key aspect of PCA that separates it from other dimensionality reduction techniques. In this project, each principal component vector can be considered a ‘pattern of activation’ between the 15 joints in the hand.
Figure 5. - The principal components of a 2D dataset
The inverse PCA algorithm takes two control signals utilizes the first two principal components and produces 15 joint angles as an output (Figure 6). This algorithm takes advantage of the dimensionality reduction properties of PCA. The 15 degree of freedom hand has been effective reduced to a two DOF system. The inverse PCA based algorithm and its associated dimensionality reduction differentiates this controller from other multi-function myoelectric prosthesis controllers.
Figure 6 - Inverse PCA Calculation
The PCA algorithm and joint angle calculation was easily implemented using LabVIEW as shown in Figure 7. The input to the algorithm can either be a mouse cursor (x,y) coordinate or 2 EMG signals from the EMG Acquisition SubVI. The calculation takes place by using the principal component matrix produced from the grasping trials. A joint angle offset is implemented in order to produce an anthropomorphic neutral hand posture andthen the 15 joint angles are sent to the Virtual Hand SubVI.
Figure 7. – PCA Algorithm Block Diagram
Virtual Hand
The virtual hand design was modeled with 15 joints and built within the Labview environment using object oriented programming. Each digit has two degrees of freedom (the MCP and PIP joints), the thumb has 4 degrees of freedom (the MCP joint, PIP joint, abduction, and rotation),and the middle, ring, and little fingers each have an abduction degree of freedom. The size of the virtual hand was modeled after a 50th percentile male hand and the joint axes were modeled after anthropomorphic data as shown in Figure 8. A neutral hand posture was derived using empirical data and is used as a baseline input to the virtual hand.
Figure 8 - Virtual Hand model with 15 articulatingjoints
There are many virtual hand software packages available, but using LabVIEW is preferable for many reasons. The biggest advantage of using LabVIEW for this purpose is that the model is controllable in real time. This virtual hand model is also preferable to any other commercially available software because LabVIEW has such a strong data-acquisition interface. This virtual hand VI will be used for years to come by our laboratory for investigations on the design of more functional and intuitive prosthetic hands. Also, the object oriented programming (Figure 9) is flexible enough to redesign the geometry of the hand if necessary. Eventually, we will import a Solidworks CAD file of commercially available prostheses in order to provide an even better visualization.
Figure 9 –Virtual Hand Block Diagram
The Implementation:
The front panel is shown in Figure 10 and 11. The open hand on the left is the controllable hand that is manipulated in real time. The pinching hand on the right is the commanded hand which the user tries to imitate. The waveform on the bottom of the screen displays the real time processed EMG signals from 2 sites on the user’s forearm. The toggle in the upper left corner distinguishes between mouse control and EMG control. The array of Booleans describes the accuracy of the controlled hand to the commanded hand. The menu in the center of the screen toggles the commanded hand into 4 different postures.
Figure 10 – Front Panel
Figure 11. - Video of Front Panel in Use
In addition to the front panel shown, there is a EMG Tuning tab (Figure 12) and a Testing tab (Figure 13 and 14). The EMG Tuning tab is used to tune the EMG sensors for each particular subject. The signal from each electrode depends on many factors including hair, sweat, etc. and must be adjusted before each testing session. In the EMG Tuning tab, the gains and thresholds can be adjusted to accomodate these differences between subjects and to create the most comfortable testing environment for the subject. The Testing tab is a fully automated testing interface. The experimenter enters a script of trials and then the VI iterates through the testing session while saving data after each trial.
Figure 12. - EMG Tuning Panel
Figure 13. - Testing Panel
Figure 14. - Testing Block Diagram
The following videos depict the two different types of control that is supported by this project. Mouse control is used as a troubleshooting interface and as a demonstration of the code (Figure 15). The EMG control scheme mimics the use of a physical prosthesis and is used for multiple research studies.
Figure 15. - Video of Mouse Control
Figure 15. - Video of EMG Control
