Design and Mechanical Testing of a Novel Dual-Stiffness Ankle-Foot Orthosis

Passive-dynamic ankle-foot orthoses (PD-AFOs) are a type of biomechanical assistive device that provides support and stability to the ankle joint [1]. PD-AFOs consist of a foot plate, shank cuff, and strut connecting the footplate and cuff. The support and stability provided by PD-AFOs generally come from the size, geometry, and thickness of the strut. Benefits of PD-AFOs include partially replacing lost plantar flexor muscle function during stance [2], increased walking speed [3], increased late stance (push-off) forward ground reaction force (GRF) impulse [4], increased peak dorsiflexion moment (torque) [5,6], and increased peak ankle power generation during push-off [5] when compared to either no PD-AFO or a patient's clinically prescribed AFO.

One way of customizing PD-AFO bending stiffness to an individual is by utilizing a biomechanical metric known as natural ankle quasi-stiffness (NAS). Historically, NAS has been defined as the average increase in sagittal ankle moment divided by the increase in ankle dorsiflexion angle during the loading phase of stance (second rocker), as defined by a linear least squares best fit of ankle moment versus angle [7]. If an individual with a lower limb injury or impairment exhibits decreased NAS compared to a similar healthy baseline, then a PD-AFO's stiffness can be tuned to make up for that difference between the impaired individual's NAS and the healthy baseline NAS [6]. This process yields a PD-AFO with a single bending stiffness that is present throughout its entire range of dorsiflexion. However, a typical, healthy ankle joint does not exhibit single, linear stiffness throughout the entirety of the loading phase [8]. We have observed in prior studies that some PD-AFO users do not fully utilize a PD-AFO's full range of motion during walking [6]. We theorize that this behavior may be due to a PD-AFO being tuned to a single, average stiffness throughout the loading phase [6]. As such, the patients may not be able to initiate dorsiflexion since the PD-AFO may be too stiff during early loading when natural ankle quasi-stiffness (NAS) is relatively low [8].

Updated models of NAS can capture more of the complexities of the natural ankle have been developed. Specifically, these models quantify the dual stiffness properties of the natural ankle (DS-NAS) [9,10]. DS-NAS divides the loading phase into early loading (EL) and late loading (LL) subphases, with separate NAS values. This method of modeling NAS improves model fit compared to standard single NAS by drastically decreasing the root mean squared error (RMSE) [8]. A critical component of DS-NAS is the transition angle or the point at which EL ends and LL begins. The exact transition point can be defined in multiple ways [9,10], varies from stride-to-stride, and varies based on an individual's natural plantarflexion/dorsiflexion posture. Generally, this transition point occurs in the middle of the entire loading phase in terms of percent stance.

Although DS-NAS properties have been examined in the natural ankle, to our knowledge dual stiffness properties have not yet been incorporated into a PD-AFO to create a DS-AFO. We posit that a novel PD-AFO design with dual stiffness properties (i.e., lower stiffness at small dorsiflexion angles and higher stiffness at larger dorsiflexion angles [8–10]) may enable users to more easily initiate dorsiflexion while still providing enough support during late loading to control ankle mechanics. The goal of this study was to design and test a DS-AFO that can: (1) achieve two distinct regions of stiffness when dorsiflexed through a normal range of human motion, just like DS-NAS and (2) has an adjustable transition point between those two stiffness regions to occur at a higher or lower dorsiflexion angle. The DS-AFO should also be able to deform through a normal dorsiflexion range of motion for healthy walking of 12–15 deg dorsiflexion [11–13] repeatedly without breaking. More specifically, our first design objective was that the DS-AFO would be designed such that the increase between EL and LL stiffness will be at least 10%. This first objective sought to determine if the DS-AFO exhibits a substantially different stiffness profile compared to a standard single-strut AFO. Our second design objective was that no significant differences between EL or LL stiffness values would exist if the activation angle at which the DS-AFO's second stiffness region began was shifted by ±1 deg dorsiflexion. This second objective was meant to determine if minor changes in activation angle result in significantly different stiffness values.

The novel DS-AFO consists, like other typical PD-AFO designs, of a foot plate, shank cuff, and a strut connecting the foot plate to the cuff. The novel features of the DS-AFO are a double strut design with accompanying catching mechanism attached to the rear of the cuff. Both struts contribute to the net bending stiffness of the DS-AFO at different periods of the loading phase of stance. The dual stiffness is achieved by utilizing the difference in forward progression of the shank compared to the stationary foot on the ground. A main strut, with stiffness k₁, (also known as early loading (EL) stiffness) is secured to the posterior side of the foot shell and cuff, which wraps around the shank (Fig. 1). A second strut, with stiffness k₂, is also secured to the foot shell just posterior to the first (anterior) strut. The strut is fit snugly into a slot in the posterior of the foot plate, and is secured to the foot plate with two bolts running through the strut and the foot plate. The heads of the bolts rest in a relief on the anterior face of the heel of the foot plate, and are covered with soft fabric so as not to contact the users skin. The bolts are secured with wing nuts for easy removal. The second (posterior) strut is secured to the foot plate in the same way as the anterior strut, but is not secured to the cuff at the top, but instead is free-floating within a catching mechanism attached to the posterior of the cuff, protruding backwards. As the DS-AFO bends through small dorsiflexion angles, the anterior strut is the only engaged strut between the cuff and the foot shell, with stiffness k₁, or EL. The top of the posterior strut does not touch anything at these small dorsiflexion angles, and thus does not contribute to the net stiffness of the DS-AFO. Once a high enough dorsiflexion angle is reached, the catching mechanism touches the top of the posterior strut and forces it to bend along with the anterior strut, resulting in a net bending stiffness between the cuff and foot shell of k₁ + k₂, also known as late loading (LL) stiffness (Fig. 1).

The specific DS-AFO used for this study is modular and adjustable in nature (Fig. 2). All components (foot shell, cuff, anterior strut, & poster strut) can be easily swapped to test numerous strut combinations across a wide range of conditions. Both struts can be changed without removing the foot shell or cuff. Cuff height is set with a friction clamp that secured the anterior strut to the cuff. Additionally, the catching mechanism is adjustable to allow the posterior strut to engage at different dorsiflexion angles. This is essential for achieving consistent posterior strut alignment and activation for different conditions. The foot shell used with the DS-AFO is a plate with a stiff, solid heel cup design. This ensures minimal deformation of the foot plate/strut interface while maximizing the bending of the dual-strut design. In other words, bending of the DS-AFO is isolated as much as possible to only the struts by using this rigid heel cup design. Prior design iterations included other heel geometries, such as an inverted Y heel design, but those included substantial deformations at the heel/footplate interface. Thus, deformation was not limited to just the struts, which limited the dual stiffness effect.

This novel DS-AFO was prototyped in collaboration with Custom Composites, LLC (Cranston, RI), who also manufactured the final devices used for this study.

Testing of the DS-AFO was conducted using a machine called the Mechanical AFO Stiffness Testing Four-Bar (MASTF) (Fig. 3), which is based on the Bi-articulating Reciprocating Universal Compliance Estimator (BRUCE) designed and used by Bregman et al. [14]. The MASTF only contains an ankle joint with adjustable height, in contrast to the BRUCE, which has articulating knee, ankle, and forefoot joints. The MASTF outputs ankle joint angle via an encoder, and linear force from a load cell placed inside the front surface of the cuff. The distance from the ankle joint to the center of the load cell was measured prior to each test condition and used to calculate net moment, which was then used with the joint angle data to calculate stiffness. The MASTF was not designed specifically for the DS-AFO and can be used to measure any AFO's bending stiffness.

For MASTF testing, the DS-AFO foot shell was first secured with clamps to a flat surface. Clamping pressure was placed as close to the heel portion of the foot shell as possible to prevent the heel from lifting during testing and bending the toe region of the foot plate. A horizontal rod, representing the ankle joint axis, sat within bearings whose vertical position could be adjusted. The rod had an encoder in line with it, which measured this “ankle” joint angle. A vertical bar, representing the shank, ran through the cuff along the length of the strut and was fixed to the ankle joint rod. This vertical bar also had a button load cell fixed within a housing that could slide up and down to be placed within the inner front of the cuff. Two other lengths of bars then attached the vertical bar to a motor, which drove the dorsiflexion of the AFO at a constant speed of <1 deg/second and achieved quasi-static testing conditions. As the DS-AFO was dorsiflexed by the vertical bar, the ankle angle was measured by the encoder, and force at the cuff was measured by the load cell. The distance between the ankle joint center and the load cell was measured prior to testing, and measured again after the peak dorsiflexion angle was reached (the difference was always less than 5 mm). We assumed a linear interpolation of the load cell's vertical displacement along the shank bar with respect to dorsiflexion angle between the starting and ending positions throughout the test. This way, an instantaneous moment arm and resultant resistive moment generated by the DS-AFO could be calculated over the range of motion at each point. The difference in resultant moment between using this interpolated moment arm method versus a constant moment arm (either the initial or final load cell position) was less than 0.7% at any given point for every condition. The load cell housing was designed to be able to move up and down on the shank bar instead of being fixed in place because a fixed uni-axial load cell may experience off-axis forces that could not be measured. By allowing the load cell to move slightly, off-axis forces do not develop.

For this study, 10 different DS-AFO configurations were tested with the MASTF in a randomized order, outlined in Table 1. The single strut configuration (“Single”) only used an anterior strut, which was the same anterior strut used for all other configurations. Only the posterior struts and the activation angle were changed across conditions. The strut stiffnesses were chosen based on number of carbon fiber unitape layers, with stiffness increasing as the number of layers increase. The anterior strut used two unitape layers, which was chosen because it was the lowest stiffness strut that could repeatedly be driven through a range of motion of at least 12–15 deg. This total dorsiflexion range of motion was chosen because peak dorsiflexion angle in healthy individuals when walking at comfortable walking speed is approximately 12–15 deg [11–13]. Posterior struts for the other configurations had 4 (low), 7 (med), and 10 (high) unitape layers. These unitape layers were chosen because their average stiffness values were as close as possible to the next stiffest (or less stiff) strut's average stiffness value, plus one standard deviation of the adjacent strut's measurements.

The activation angles for the posterior struts were chosen as 6.5 deg dorsiflexion, which was half of the total dorsiflexion range of motion of 13 deg, then modified ±1 deg to examine if the DS-AFO exhibited significantly different EL and LL stiffness with minor activation angle changes. This ±1 deg threshold was chosen for practical purposes as a reasonable margin of error in the setup process of the DS-AFO. If there are no changes in DS-AFO outcome measures with a shifted activation transition angle, then we know that tuning the activation angle to within ±1 deg of the specified angle should not affect its stiffness properties. Due to slight differences in posterior strut thicknesses, the following standard procedure was used at the beginning of every new condition to set the DS-AFO activation angle. The activation angle of the DS-AFO was set by first bringing the DS-AFO to its neutral, unloaded position. Then, the vertical shank bar and attached load cell were moved such that load cell was pressed against the inside of the cuff with just enough force to prevent the load cell from sliding freely down the vertical shank bar due to its own weight (measured to be ∼0.3–0.5 N). This was done in such a way that no noticeable elastic deformation of the anterior strut was present. The encoder was zeroed and the position of a spirit level attached to the top of the vertical shank bar was noted as a baseline position for that configuration. Then, the MASTF dorsiflexed the DS-AFO until the desired activation angle was reached, which was measured by the encoder. While in this position, the catching mechanism was adjusted such that it lightly touched the posterior strut. At this point, the activation angle was considered set and the catching mechanism was locked in place. Prior to every trial for a given configuration, the AFO was repositioned back to its baseline position noted by the spirit level. This process was repeated for every AFO configuration.

For each of the 10 configurations, the same DS-AFO was tested 10 times each during three separate sessions, totaling 30 tests per configuration. This was based on a Cohen's d = 0.45 from preliminary data, power 1-β = 0.8, and significance level p = 0.05. The purpose of testing in multiple sessions was to account for any variability due to setting up the DS-AFO in the MASTF (e.g., foot plate clamping, cuff tightness, exact load cell placement). The same individual conducted all tests. The DS-AFO's cuff, foot shell, and cuff height, and the MASTF's ankle joint height and anterioposterior distance from the DS-AFO were kept the same across all testing sessions. Repeated measures analysis of variance and Tukey Honest Significant Difference tests were used to detect any significant differences between AFO configurations.

On average, the Single AFO's bending stiffness profile was not perfectly linear. Artificially applying an activation angle of 6.5 deg to the Single condition data and calculating EL and LL stiffness the same way as the DS-AFO configurations yielded a 37.23% increase from EL to LL stiffness, even though there was only one strut (Table 2). This can be observed in Figs. 5–7, in which the Single strut configuration's moment versus angle curve is not a perfectly straight line, but instead slightly increases its slope as dorsiflexion angle increases. This means that even the Single configuration still exhibited some nonlinear stiffness properties from low to high dorsiflexion angles. For the DS-AFO conditions, average increase in stiffnesses from EL to LL was 149.77% (Low), 179.68% (Med), and 262.27% (High), respectively, across all activation angles (Table 2). This was far greater than our required threshold of 10% increase, which the Single AFO also met. Furthermore, those stiffness differences were maintained when the activation angle was shifted by ±1 deg for all Low and Med conditions. EL stiffness for the Single condition was, interestingly, significantly higher than several of the DS-AFO conditions' EL stiffness (Fig. 4(a)). This increased EL stiffness for the Single strut configuration may be due to the strut being more directly secured to the foot plate without the posterior strut possibly contributing to additional deformation in the DS-AFO configurations. Higher stiffness struts and lower activation angles both contributed directly to higher peak moments at 13 deg dorsiflexion (Table 2). Other statistically significant differences were exhibited between EL stiffness across various conditions, mostly between the Single configuration and others. While not statistically significant, there was a trend of EL stiffness increasing as activation angle increased. The differences between the highest and lowest EL stiffnesses across all conditions were less than 0.33 N*m/deg, and less than 0.22 N*m/deg if the Single condition is excluded. For reference, the overall standard deviation of EL stiffness values across all conditions (including Single) was 0.2515 N*m/deg. LL stiffness clearly increased with higher posterior strut stiffness, as expected. LL stiffness was generally not different within each condition group when activation angle changed, except for the High conditions. LL stiffness for the High 5.5 condition was significantly lower than High 6.5 and High 7.5 (Fig. 4(b)), but that average difference was less than 0.16 N*m/deg. For reference, EL stiffness standard deviation across all High conditions was 0.1757 N*m/deg.

Our results supported our first design objective that the increase in stiffness from EL to LL would be at least 10%. This design objective was satisfied by not only all DS-AFO conditions but also unexpectedly by the Single configuration. Our results also generally supported our second design criteria that changing the activation angle by ±1 deg did not affect EL or LL stiffness values, with the High DS-AFO conditions being an exception. This novel design allowed us to independently alter EL and LL stiffness during distinct phases of DS-AFO loading. This separation and control of EL and LL stiffness better match the stiffness requirements of normal, healthy walking than a single strut AFO.

Related to our first design objective, we can clearly increase DS-AFO stiffness from EL to LL above and beyond that of the single strut AFO condition, as evidenced by the >100% increase in stiffness after the transition point for every DS-AFO condition. However, the Single AFO also exhibited some inherent nonlinearity, which was unexpected. The Single configuration does not exhibit a purely linear stiffness slope, but instead exhibits an average increase in stiffness of approximately 37% if EL and LL stiffness is calculated using an artificially applied activation angle of 6.5 deg, the same way as the DS-AFO stiffness values were calculated (Figs. 5–7, Table 2). This increase was small in comparison to the DS-AFO conditions but is clear evidence that the Single configuration's overall stiffness is not perfectly linear. This complex stiffness could come from components of the AFO with lower stiffness than the struts, namely, the cuff, that have to “max out” before the full stiffness of the struts can be realized, or due to nonlinear stiffness behavior of the struts at large deformations. In other terms, the results suggest that this system can be represented by two or three springs: a finite extensible, low-stiffness spring which we will call the “system components” spring (cuff/padding/straps), in series with one or two high-stiffness springs (anterior and/or posterior struts). The system components spring, being finite extensible, ultimately reaches a high enough stiffness at the limit of its extension that it is higher than the stiffness of the struts, and therefore deformation transfers from the system components to the struts. The system components spring deforms much more than the high-stiffness spring at low-loads, then the high-stiffness spring takes over once the low stiffness spring reaches the limit of its deformation. We visually observed, but did not measure, substantial deformation that occurred in the cuff shells and the straps connecting the front cuff to the back cuff. This might mean that at lower dorsiflexion angles, the cuff absorbed a substantial portion of the load applied by the MASTF (or shank). As the cuff deformed more, more load was directly transferred to the struts and more of the true stiffness of the struts was realized. Likewise, portions of the footplate might have deformed or shifted at low dorsiflexion angles due to imperfect clamping, which could have contributed to lower overall stiffness. The existence of deformable components of the DS-AFO that are not the struts is not necessarily a drawback of this DS-AFO design, and is likely present in many other AFOs with similar foot plates and cuffs. Some of these components are required to make devices comfortable and adjustable for users (foam padding, adjustable straps). Including these components as part of the overall measure of AFO stiffness yields a closer approximation to what a user is actually experiencing, instead of isolating certain parts (i.e., the struts) for stiffness testing. Determining the effect of the cuff on net stiffness could be measured by testing just the strut without the cuff. However, this measurement is extraneous if the goal is to understand how users experience the net stiffness of the whole AFO system.

Interestingly, EL stiffness was higher overall for the Single configuration compared to the DS-AFO configurations (Fig. 4, Table 2). Based on our stiffness testing protocol, we can only conclude that the addition of a second posterior strut may disrupt the connection of the struts to the rest of the AFO system and therefore may alter the overall stiffness properties. We did not specifically investigate this, and therefore further research into the strut/foot plate interface may be required to fully understand this phenomenon.

Changing the activation angle by ±1 deg generally did not significantly affect EL or LL stiffness, but there were slight increases in EL present as activation angle increased. For example, among the High conditions, High 5.5 had the lowest EL stiffness and High 7.5 had the highest. This might be explained by the same phenomenon that contributed to the complex stiffness of the Single configuration, in which lower-stiffness components take up proportionally more of the overall load at small dorsiflexion angles. A higher activation angle means that the lower-stiffness components have more time to reach the limit of their finite extensibility before exhibiting their final and higher-stiffness components contribute more to the overall stiffness. The dorsiflexion range between 5.5 deg and 7.5 deg may have included a point at which a critical load was borne by the lower-stiffness components before the higher-stiffness struts contributed more. Generally, these changes are slight and not statistically significant, but this point highlights the importance of cuffs, foot plates, and other AFO components that contribute to the overall stiffness that the device exhibits.

Related to cuff stiffness and the role of components other than the struts that contribute to overall device stiffness, another notable result is that the standard deviation of EL stiffness values was almost always higher than LL stiffness values (Table 2) (High 6.5 is an exception). The unquantified stiffness of the cuff possibly contributes to this higher variability, since EL is the period when the cuff likely deforms most. Additional contributors to EL variability could come from the nature of using the MASTF, which requires the AFO to be driven forward, then returned to approximately the same starting position. The program used to record trial data automatically zeroed the load cell and encoder after every trial. Systematic efforts to mark the starting position of the AFO were taken, including marking a spirit level attached to the vertical driving bar, but it is impossible to return the AFO to exactly the same position every time. The starting position for each condition was chosen to be at such a point that there was only enough force between the load cell and cuff to hold the load cell in place. The existence of any force exerted between the load cell and the cuff, however small, necessarily means that there will be minor deformations in the cuff or cuff padding. Therefore, slightly varied levels of preload and predeformation of the cuff were present at the start of each MASTF test. These likely contributed to the overall increased EL stiffness variability. For LL stiffness, the finite extensible system components spring reached the limit of extension by the time the activation angle was reached (meaning that all deformation has been transferred to the struts), and thus a more pure representation of just the stiffness of the struts.

Stiffness values reported in our results section do not necessarily reflect the stiffness values that an individual might want, or what could be optimal. As such, comparisons to actual DS-NAS values were not conducted. Building a stiffness-customized prescription framework for individuals would be valuable future work, but it was not the goal of this study. Rather, this study provides a proof-of-concept that a DS-AFO can exhibit discrete, tunable stiffness regions throughout a physiologically relevant range of motion.

Overall, the DS-AFO met our design criteria. EL and LL stiffness were clearly different and independently adjustable. Additionally, altering the activation angle by ±1 deg generally did not significantly change EL or LL stiffness values. This study represents a proof-of-concept for passive AFOs with complex stiffness properties that may closely approximate natural ankle dynamic stiffness properties during stance in walking. Future work should consist of testing this DS-AFO with human subjects to understand how varying stiffness during EL and LL separately affects the physiological musculoskeletal system, and should also include streamlining the DS-AFO design to be worn by actual patients in a free-living environment. Designing a DS-AFO that allows users to easily initiate bending to utilize its full range of motion may lead to more positive rehabilitation outcomes or fewer negative gait compensations when using these devices.

The DS-AFO was prototyped in collaboration with Custom Composites, LLC (Cranston, RI), who also manufactured the final devices used for this study. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.

• National Science Foundation's (Grant No. 1940700; Funder ID: 10.13039/100000001).

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

AFO =

ankle-foot orthosis

DS-AFO =

dual-stiffness ankle-foot orthosis

DS-NAS =

dual-stiffness natural ankle quasi-stiffness

EL =

early loading phase of stance

EL-NAS =

natural ankle quasi-stiffness during the early loading phase

LL =

late loading phase of stance

LL-NAS =

natural ankle quasi-stiffness during the late loading phase

NAS =

natural ankle quasi-stiffness

PD-AFO =

passive-dynamic ankle-foot orthosis

SL-NAS =

single linear natural ankle quasi-stiffness during the entire loading phase