Skip to main content

Research Projects


Active-mixing Hotend in Fused Filament Fabrication Learning


Advance functional gradient additive manufacturing technologies.

  • Use active-mixing to provide well-mixed blends of multi-phase>
  • Develop software to facilitate local composition control (LCC) in fused filament fabrication (FFF)..
  • Enable assignment of material composition to facilitate the formation of components with regions of selectable mechanical properties. Characterize material blends with stiffnesses comparable to the bones and soft tissues of a human knee..

Technical approach 

Develop an FFF printer with an active-mixing hotend supported by firmware and software to enable LCC with well-mixed blends of thermoplastic filaments.

    • Identify compatible thermoplastic filaments including one filament with stiffness comparable to the bone, one filament with stiffness comparable to cartilage, and one water-soluble filament to enable printing with soluble support.
    • Design and fabricate active-mixing hotend including three filament inlets and an actuated element supported by bearings and an energized seal.
    • Modify FFF printer to support the use of the active mixing hotend. Modifications include customization of most components including control board replacement and development of custom firmware.
    • Develop software to sort printing in progressive blend ratios with priming blocks to ensure LCC is performed with the assigned compositions having minimal contamination.


Figure 1 – Diagram of patent-pending active-mixing hotend. Actuated element facilitates the blending of three filaments before extruding through the nozzle.



Figure 2 – Images of tensile test specimens composed of blue high-temperature polylactic acid (HTPLA) and white Ninjaflex thermoplastic elastomer (NF).




Figure 3 – Representative stress-displacement curves for blends of HTPLA and NF. The inset graph provides a detailed view of stress-strain curves for strain values below 3% as recorded by an extensometer.



Figure 4 – Summary of tensile test results.



Synthetic ACL development using LCC


Investigate synthetic tissue fabrication using local composition control (LCC) in fused filament fabrication (FFF) to promote effective surgical training in surgical simulation.

  • Develop synthetic tissues with features to biomimic functional gradients between ligaments and bone.
  • Develop synthetic tissues with components having a composition selected to match the modulus of elasticity of its biological counterpart.
  • Increase the ability to customize material structures in synthetic tissue to approach simulation of physiological response to load.


Figure 1 – Computer models of knee components and grip sections to facilitate mounting in tensile test equipment.


Figure 2 – Load-elongation curves of FATCs from tensile tests in anatomical load orientations. The graph includes results from three synthetic FATCs and a load-elongation curve for younger cadaveric FATCs are included for reference, which was reported by Woo et al.


Figure 3 – Single layer preview of FATC print files including LCC and FGs in various stages of processing.



Figure 4 – Representative images of a synthetic FATC specimen during tensile testing. Images were collected by DIC image capture and correspond with notable phases of load response.



Table 1 – Summary of tensile test results from loading synthetic FATC specimens along anatomical and tibial orientations with 30° knee flexion.



Joint Load Simulator

  • Simulate a human maneuver in real-time including multiplanar kinematics, ground reaction forces, and musculotendon forces.
  • Mimic muscle moment arms using anatomical origins for musculotendon actuators.
  • Investigate effects of muscle forces and effective location of the ground reaction force (toe vs. heel) during a squat maneuver.
  • Assess the short-term accuracy and precision errors of The University of Texas at El Paso Joint Load Simulator (UTJLS) during squat maneuvers.

Technical approach 

Develop an in vitro simulator with two robotic manipulators and eight musculotendon actuators:

  • 3 degrees of freedom (DOF) for simulating tibial moments/rotations
  • 6 DOFs for simulating femoral moments/rotations and forces/translations
  • Force and position sensors
  • Pressure sensors and strain gauges for measuring internal joint loads and soft tissue deformation
  • Real-time operating system and controller including control scheme flexibility
  • Actuators providing sufficient acceleration and forces to simulate highly dynamic maneuvers (e.g. drop-jump) Assess accuracy and repeatability of real-time simulation of various squat maneuvers on multiple cadavers during synchronous simulation of ground reaction forces, joint kinematics, and muscle forces.


Table 1 –Simulator design characteristics.


Figure 1 – rendering of the UTJLS. (The color indicates axes of control. The upper manipulator includes three translations (red, blue, green). Both manipulators have three rotations (yellow, teal, brown). The lower manipulator has manufactured components corresponding with a design for translations (transparent).)


Figure 2 – Center of pressure (COP) on medial and lateral tibial plateaus for two squat maneuvers which were repeated three times on the same cadaver. Warm colors are COP paths without muscle forces, and cool colors are COP paths with 50% scaled muscle forces. These pressure measurements are highly sensitive to variations in load and indicate the highly repeatable performance of the UTJLS.



Prosthetic Knee Stumble Detection using Machine Learning


Provide real-time identification of stumbles to support microprocessor operation to reduce the incidence of amputee injury from stumbling.

  • Mount low-cost sensors with a remote battery-operated microprocessor on a prosthetic knee to collect joint kinematic data.
  • Collect a dataset from amputees walking and stumbling.
  • Use a machine learning network to distinguish between walking and stumbling using kinematic data as measured by prosthetic sensors.

Technical approach 

LIMBS M3 prosthetic knee was modified to include two nine degrees of freedom (DOF) inertial measurement units (IMU) with Google Coral microprocessor. The system was configured to record kinematic data during prosthesis use. Three amputees were fitted with the custom prosthetic knee, and kinematic data were collected during gait sessions including induced stumbling. A variety of stumbling was induced including stumbling due to load from a bungee, impact with an obstacle, and stumbling as a result of an uneven surface. A recurrent machine learning network was developed to accurately predict stumbling using long short-term memory (LSTM). Leave-one-out cross-validation was used due to the small dataset of three subjects.


Figure 1 – LIMBS M3 prosthetic knee modified to include nine degrees of freedom inertial measurement units on both the femoral and tibial components of the prosthetic knee. Google Coral microprocessor captures data from sensors to develop kinematic datasets and for the development of machine learning networks.


Figure 2 – Images of the three methods used to induce stumbling during gait.



Figure 3 – Angular velocity of knee flexion with respect to gait phase. Representative gait profiles from two above-knee amputees are plotted alongside a gait profile from a healthy subject as reported by Winters [Winters 1990].



Figure 4 – Example of the angular velocity of the prosthetic knee during stumbling. Redline indicates stumble as identified by lab personnel, and the blue line indicates stumble as identified by recurrent machine learning network.



Additional Research Topics