Highly sensitive and easy-to-attach wearable sensor for measuring finger force based on curvature changes in an ellipse-shaped finger ring

Technologies for digitizing worker actions to enhance human labor tasks, mitigate accidents, and prevent disabling injuries have garnered significant attention. This study focuses on monitoring the force exerted by the fingers and developing a wearable fingertip force sensor based on a simple elliptical ring structure in conjunction with a commercially available resistive bend sensor. Resembling a ring accessory, the sensor is easy to attach and detach, and exhibits high sensitivity, with a resistance change of approximately 9% for a fingertip load of 1 N. Furthermore, to mitigate crosstalk during finger flexion, we propose a combined configuration employing this ring-shaped sensor alongside another sensor designed for measuring and rectifying finger flexion angles. Additionally, we introduce an empirically derived fitting function and a straightforward calibration procedure to extract the function’s parameters. The proposed system achieves an average RMS error of 0.53 N for force estimations of approximately 5 N, even during finger flexion and postural changes.


Introduction
The digitization and monitoring of workers' activities propelled by advancements in sensor and information processing technologies, foster heightened expectations across a broad Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence.Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.spectrum of applications.These technologies hold promise for mitigating musculoskeletal disorders (MSDs) in factory settings and reducing workplace accidents.Sensors, capable of tracking posture, force, and repetitive motions, serve to identify hazardous behaviors promptly, enabling the implementation of corrective measures to prevent injuries [1][2][3].Moreover, they are envisaged to enhance work efficiency and refine training protocols.The analysis of movement data facilitates the identification of factors contributing to efficiency gains [4] and pinpoints inefficient workflows or postures.Subsequent ergonomic redesigns and tailored training programs optimize movements while reducing fatigue [5][6][7].
In the realm of robotic automation, work monitoring technologies prove efficacious in tasks such as aggregating and analyzing human task data for robot training [8,9], or discerning human intentions to facilitate seamless human-robot collaboration [10,11].However, to harness these technologies, unobtrusive methods for monitoring worker behavior during tasks are imperative.Previous studies have employed various techniques to estimate body movements, including image analysis from external cameras [12,13], data from body-mounted inertial measurement units (IMUs) [14][15][16], and integration of data from both cameras and IMUs [17].Notably, hand movements during work merit particular attention.Researchers have utilized IMUs [18,19] and glove-integrated strain sensors [20,21] for estimating hand movements.
This study emphasized measuring finger force because it is pivotal for quantifying work tasks and understanding MSDs.Notably, the magnitude and frequency of finger force are recognized as significant risk factors concerning MSDs [22,23].Monitoring exerted force, speed, and timing during actual work tasks holds promise for assessing the risk of MSD onset among workers.Recognizing such risks facilitates the implementation of preemptive measures, including instructing improvements in work methods or, if necessary, modifying tasks altogether.However, observing the generation of an external force poses challenges, in contrast to measuring movement.Some studies have directly measured force in fingertip pads by affixing force sensors [24][25][26].Nevertheless, this method encumbers fingertip function and impedes the replication of typical work tasks.Other studies are exploring ultra-thin pressure sensors utilizing micro-or nano-mesh technology, aiming to minimize wearer sensation [27,28].However, these sensors require skin attachment, presenting challenges during application and removal.An alternative approach involves indirectly estimating fingertip force from deformations, such as those occurring in the fingertip itself.For instance, systems have been devised to estimate force from changes in nail color associated with fingertip deformation [29,30], or from flesh deformation on the side of the finger [31,32], preserving tactile sensation by avoiding direct fingertip coverage.Nonetheless, placing additional pressure on the fingertips may induce discomfort for workers and hinder task execution.To address these challenges, a method has been proposed to measure tendon deformation at the finger base using a ring-shaped sensor equipped with a strain gauge [33].This approach enables measurement at the finger base, potentially reducing discomfort and its impact on work tasks.However, firmly attaching the sensor ring to the finger to concentrate deformation on the strain gauge presents difficulties during donning and doffing.Furthermore, the method exhibits low sensitivity (less than 0.1% change in resistance per 1 N) and necessitates an amplification circuit, potentially increasing the size and weight of the hand-worn circuit module.
In this study, the aim was to devise a sensor that is as easy to don and doff as a ring accessory, without necessitating an amplification circuit, all while maintaining high sensitivity.The proposed sensor comprises a simple, flexible, elliptical ring structure coupled with a commercially available piezoresistive bend sensor.With this design, the maximum sensitivity attained was a resistance change of approximately 9% under a 1 N load.This paper describes the sensor's design and structure, along with the methodology employed to estimate force based on the sensor output.

Design
Figure 1 is a schematic representation of the sensor's sensing principle.The sensor mechanism comprises a protrusion exerting pressure on the underside of the finger and a spring forcing the protrusion against the finger (figure 1(a)).Pressure is applied to the palm near the finger's base, where tendons are situated, as indicated by the thick red line in the figure.Upon applying force to the fingertip, tension arises in the tendon, causing the sensor's protrusion to be pushed back, as depicted in figure 1(b).Consequently, the force exerted by the fingertip can be estimated by measuring the deformation of the spring.
Figure 2 shows the sensor's elliptical ring-shaped structure.For the ring to function as a spring when worn on the finger, the diameter H along the minor axis is intentionally smaller than the finger's thickness, aligning with the sensing principle elucidated earlier.The lower end of the ring features an ellipsoidal protrusion measuring 3 mm × 1.5 mm × 1.5 mm in each axial direction and extending 2 mm from the ring's inner surface.Figure 2(b) presents a cross-sectional view of the ring on the finger.When the fingertip applies force, the tendon moves downward, causing the elliptical ring to deform into a more circular shape.In this study, this deformation was quantified as a change in curvature, for which a commercially available bend sensor was affixed to the ring.Commercialized high-sensitivity resistive bend sensors are readily available, typically exhibiting performance (change in resistance due to curvature change) within the range of several percent per meter (1 m −1 ) (e.g.BS-series: Sensia Technology Co., Ltd, MB series: TAIWAN ALPHA ELECTRONIC CO., LTD, Flex Sensor series: Spectra Symbol Corp.).Given the elliptical ring's larger circumference compared to that of the finger, the device can be conveniently donned and removed from the finger, offering the same comfort as a ring accessory.In this design, the diameter L along the major axis is 23.1 mm, the diameter H along the minor axis is 14.7 mm, the thickness T is 0.53 mm, and the width W is 6.6 mm.Notably, these dimensions were primarily tailored to match a finger of diameter 17 mm, as utilized in the experiments.For fingers of varying diameters, adjustments can be made by proportionally scaling the dimensions, as explained in supplementary information 2. In this design, the bend sensor was strategically positioned near the apex of the ellipse, aligned with the major axis, where the change in curvature was measured.It was also hypothesized that the sensitivity could be enhanced by concentrating displacement at the bend sensor's location.This was achieved by reducing the ring's thickness near the apex along the major axis direction, as depicted in figure 2(a).
The impact of partial thinning on concentrated displacement was assessed through finite element analysis (FEA).As figure 3(a) shows, a quarter model of the ring structure was generated, and force was applied to the protrusion area.Throughout this process, the relationship between protrusion displacement and curvature change at the sensing location (apex along the major axis) was examined.
Analyses were conducted for three thickness reductions at the sensing location, t, defined relative to the original ring wall thickness, T, as T × 100%, T × 90%, and T × 80%.Assuming sufficient flexibility of the ring structure, tendon displacement was simplified as a constant displacement applied to the ring.At this stage, sensing sensitivity can be defined as the change in curvature per unit displacement at the sensing location.A comparison of these results is depicted in figure 4. As anticipated, a thinner ring leads to a more pronounced change in curvature, hence, higher sensitivity.Specifically, the design with t = T × 80% exhibited a change in curvature 2.29 times  greater than that of the design with T × 100%.The sensitivity characteristics of prototypes featuring these three thicknesses were evaluated during the experiments conducted in this investigation.

Method
Figure 5 illustrates the appearance of the prototype sensor.The ring structure was manufactured using a 3D printer (Form 3) with gray resin (Formlabs, MA, United States).The bend sensor was the BS-25 (Sensia Technology Co. Ltd, Ibaraki, Japan).Both the ring structure and bend sensor were securely affixed using PTFE tape (5453, 3 M).Due to the sensor's high sensitivity, the change in resistance was measured without a signal amplification circuit.The voltage drop (V out ) across the sensor, connected in series with a fixed resistor (R = 200 kΩ), was monitored at a constant voltage (V D ).The relationship between V out and the sensor resistance (r) is expressed as V out = V D • r/ (r + R).Consequently, the sensor resistance can be calculated as r = R • V out / (V D − V out ).A PicoScope 2205 A oscilloscope (Pico Technology Ltd, United Kingdom) was used for voltage collection.Fingertip force was measured using a force gauge (ZTA-5 N; Imada Co., Ltd, Aichi, Japan).One challenge associated with this sensing method pertains to the potential impact of finger bending.Specifically, it was postulated that finger bending could compress the flesh, resulting in a deformation that increases the finger's apparent thickness, thereby requiring correction.To mitigate this issue, an additional prototype was developed and evaluated, incorporating an extra ring positioned between the first and second joints, as depicted in figure 6.Another bend sensor connects the original ring to this additional ring to gauge finger flexion.

Sensor characteristics
For sensors of thicknesses (t) 100%, 90%, and 80% of the original thickness (T), each sensor was fitted onto a finger, and fingertip load and resistance change were measured five times using a force gauge within a range of approximately 5 N. Considering the potential influence of finger flexion on the output, measurements were conducted in two distinct states: with the finger fully extended and bent at approximately 90 • .
Figures 7(a) and (b) depict the relationship between load and sensor resistance with the finger extended and bent, respectively.Data for the three sensor levels with t at 100%, 90%, and 80% of T are superimposed in the plots.A consistent trend of decreasing resistance with increasing load was observed across all sensor levels.Additionally, the change in resistance was more pronounced near zero load and progressively leveled off as the load increased.This phenomenon likely arises from the tendons being lax near zero load, experiencing significant deformation even with minimal force, whereas at higher loads, the tendons are already taut, leading to less substantial deformation with additional load increments.Based on these observed characteristics, a logarithmic function-based fitting equation was intuitively formulated to estimate sensor resistance (r) based on load (F): where a, b, and r 0 are fitting parameters, with a representing the slope for F = 0, b is a shift parameter such that the lower its value the more rapidly the slope decreases with the load, and vice versa, and r 0 is the initial resistance at F = 0.The results of fitting using the least-squares method with this function are depicted as black dashed lines in the figures, indicating that this function provides a suitable representation of the sensor characteristics.In figures 7(c) and (d) depict graphs where the vertical axis represents the percentage change in resistance relative to the initial resistance value.It was observed that the reduced thickness (t) of the sensing area correlated with a greater percentage change in resistance, regardless of whether the finger was extended or bent.Moreover, the change in resistance was more pronounced in the bent finger state compared to the extended state.Notably, for the sensor with t at 80% of T, significant resistance changes were recorded of approximately 7.1% for the extended state and 9.2% for the bent state at a 1 N load.This resistance change rate of approximately 7%, inferred from the sensor's characteristics (refer to supplementary information 1), corresponds to a displacement of roughly 0.3 mm. Figure 8 presents the results comparing sensitivity for each sensor level and finger state-extended or bent-based on the resistance change rate, defined as the sensitivity at a 1 N load according to previous fitting results.Error bars represent the standard deviations.In the extended finger state, the sensor with t = 80% × T exhibited approximately 1.98 times higher sensitivity than that of the sensor with t = 100% × T. Similarly, the increase in sensitivity in the bent-finger state was approximately 1.71 times, aligning closely with the 2.29 times increase predicted by FEA.
Ring deformation emerges as a critical factor for this sensor.Therefore, as part of the design discussion, we evaluated characteristics when altering the ring's thickness, thereby changing its stiffness.Figure 9(a) illustrates the relationship between ring thickness and sensitivity, whereas figure 9(b) depicts the relationship with initial resistance.These results  indicate that below a certain thickness, ring sensitivity diminishes.Thinner rings may excessively expand from the outset due to finger stiffness, leading to reduced sensitivity.Conversely, initial resistance tends to increase monotonically as deformation caused by fitting the ring on the finger decreases with thicker rings.Given that thicker rings may induce a sensation of compression when worn, it can be inferred that to strike a balance between sensitivity and wearability, the ring should be made as thin (and consequently soft) as possible, within limits that preserve sensitivity.

Load estimation
In this section, we detail the procedures and outcomes of load estimation using the sensor.Figure 10 presents the sensor data alongside the force gauge output in a system featuring two sensors (as depicted in figure 6) during application of pressure by the force gauge.During the experiment, the finger was progressively flexed from approximately 0 • to 30 • , 60 • , and finally, 90 • , while exerting a force of up to 5 N for each step.from the sensor monitoring finger flexion.Figure 10(c) illustrates the load values measured using the force gauge.With increasing finger flexion, it was naturally observed that the resistance value of the flexion sensor correspondingly increased.Simultaneously, a drift in curvature can be observed, where the resistance value decreases despite the load values remaining constant, as depicted in figure 10(c).
Excluding the drift caused by finger flexion, load estimation from sensor data can be achieved using the inverse function of the fitting equation (equation ( 1)) as follows: However, the outcome depicted in figure 10 led to the conclusion that accurate force estimation would not be feasible without correcting for finger flexion.Consequently, based on these findings, a load-estimation formula incorporating the output of the flexion sensor was formulated as follows: where a, b, c, d, and r 0 are fitting parameters, with r representing the resistance of the curvature sensor that captures the tendon movement, and r f denoting the resistance of the flexion sensor.This formula modifies the results estimated with the single sensor (equation ( 2)) by multiplying it by the linear correction factor (−r f + c).In this study, we performed force estimation on three different fingers using equations ( 2) and ( 3) and compared the results.Prior to force estimation, it was imperative to extract the parameters of the estimation formula.For the calibration of parameters corresponding to each finger, we proposed a simple procedure, as illustrated in figure 11: data collection involved incremental loading from 0 N to 5 N with both the finger extended and bent at 90 • , with parameters then extracted using the least squares method.This procedure could be completed in approximately 10 s, which we deemed practically acceptable.
The upper graph in figure 12(a) illustrates the estimated results using a single sensor (fitted with equation ( 2)) compared to the corrected estimate incorporating the flexion sensor (utilizing equation ( 3)), juxtaposed with load measurements from the force gauge.Similar to the experiment in figure 10, loads of up to approximately 5 N were applied at each step while finger flexion increased incrementally.Furthermore, the lower graph in figure 12(b) displays the disparity between the two estimation results and the actual load measurements.Additionally, figures 12(b) and (c) respectively depict the outcomes of conducting the same experiment on different fingers.
When utilizing a single sensor, the estimated load values tend to be lower than the actual values.Particularly when the finger is bent at 90 • , the sensor inaccurately registers a positive load when the actual load is 0 N.This discrepancy underscores the impact of finger flexion on the measurements.Conversely, correction with the flexion sensor consistently yields good agreement with the load measurements.However, a tendency to underestimate the load with the dual-sensor configuration was noted in all experiments (figures 12(a)-(c)) when the fingers were straightened.This is likely attributed to the slight force required to straighten the fingers, which could induce finger deformation and thereby affect the sensor output.Our fingers are indeed typically slightly bent when at their most relaxed and exhibit a degree of tension when either extended or further flexed.Considering this, an improvement may be achieved by adopting a V-shaped correction factor that peaks somewhere between 0 to 90 degrees, rather than a linear model.Figure 13 presents a comparison of the maximum and root-mean-squared error (RMSE) of the load estimation experiments for each finger angle period.The maximum error was 3.67 N for a single sensor and 2.38 N after correction with the flexion sensor, with the measurement range being approximately 5 N. Similarly, the RMSE was 1.07 N for a single sensor and 0.53 N after correction with the flexion sensor.Due to the influence of spikes stemming from the response time differences between the load cell and the sensor, the analysis hereafter primarily focuses on the RMSE (see figure 13(b)).Across all finger angles, the correction applied using the flexion sensor has been demonstrated to reduce errors significantly, with an average reduction of approximately 50% in observed errors.The RMSE, constituting approximately 10% of the measurement range, indicates the sensor's effective performance for indirect load estimation, even at a point distant from the fingertip.An examination of the errors associated with different finger angles indicates that, in the singlesensor configuration, substantial errors occur at both 0 degrees and 90 degrees.In contrast, in the dual-sensor configuration, the errors at 90 degrees are significantly mitigated, demonstrating the effectiveness of the correction applied.A notable aspect of this technology is its ability to leave the fingertips entirely unrestricted, facilitating real-time force monitoring in delicate tasks and distinguishing it from conventional sensors.However, when compared to types incorporating pressure sensors on the fingertips [24][25][26], it may exhibit lower accuracy.Consequently, it may not be suitable for applications requiring precise quantification.The finger angles' influence on load estimation is substantial; therefore, the dual-sensor configuration is recommended for practical use.This configuration, however, introduces the drawback of additional components.For certain specific applications, such as measuring loads at a set pose, the single-sensor configuration may be effective.

Conclusion
A wearable fingertip force sensor was proposed and developed.It merges an elliptical ring structure with a commercially available bend sensor.Comparable to accessories, the sensor is easily attached and detached while achieving high sensitivity, with a resistance change exceeding 9% at a fingertip force of 1 N. Additionally, an average RMSE of 0.53 N was attained when estimating forces up to approximately 5 N.This sensor facilitates unobtrusive force monitoring, enabling the collection of digital monitoring data from the fingertip ultimately aimed at preventing worker injury and enhancing work efficiency.One limitation of this technology is the requirement for an additional sensor to correct finger flexion, necessitating the placement of rings at two locations on the finger.A more streamlined system capable of accounting for bending correction and requiring either a single point or dual points in close proximity would be preferable.Future studies should focus on developing such a simplified system structure.Consideration could also be given to the material of the ring, ideally opting for an elastic material that does not undergo plastic deformation or creep.While ease of prototyping was prioritized in this study by opting for stereolithography 3D printing, the use of engineering plastics with superior mechanical properties may enhance the sensor's characteristics.Furthermore, there is potential for improvement in the fitting function.To mitigate increased errors when the finger is extended, incorporating non-linear corrections into the current linear function of the flexion sensor could prove effective.Currently, measurements are conducted using external wired equipment.Nevertheless, an additional future objective would be the creation of a compact and portable component that integrates circuits for both communication and measurement.This would enable real-world applications such as monitoring factory operations.

Figure 1 .
Figure 1.Schematic of the sensing principle of the proposed fingertip sensor.

Figure 2 .
Figure 2. Structure of the proposed sensor.

Figure 3 .
Figure 3. FEA of the sensor deformation.(a) FEA model, (b) typical deformation of the sensor ring structure.

Figure 4 .
Figure 4. Sensitivity of ring designs with varying wall thicknesses at the sensing region, as determined using FEA.

Figure 5 .
Figure 5. Fabricated fingertip force sensor.(a) Overview of the sensor.(b) under-side view of the sensor attached to a finger (c) lateral view.

Figure 6 .
Figure 6.Sensor system fitted with two bend sensors to measure finger flexion.Views of the fabricated sensor from the (a) top (b) lateral.

Figure 7 .
Figure 7. Relationship between fingertip force and sensor resistance with the finger (a) straightened and (b) bent.The sensor resistance changes when the finger is (c) straightened and (d) bent.All graphs are plotted by superimposing the data obtained from the five trials.

Figure 8 .
Figure 8. Measured sensitivity for varying wall thicknesses at the sensing region.

Figure 9 .
Figure 9.Effect of ring thickness on its (a) sensitivity and (b) initial resistance.

Figure 10 (
a) displays data from the curvature sensor that measures tendon movement, whereas figure 10(b) exhibits data

Figure 10 .
Figure 10.Sensor data when exerting a force of up to 5 N on the force gauge with the fingertip while the finger is progressively flexed from approximately 0 • to 30 • , 60 • , and finally 90 • .

Figure 12 .
Figure 12.Comparison between measured and estimated fingertip force; (a), (b), and (c) show the results for three different fingers.In each figure, the upper and lower time series show, respectively, the measured vs. estimated forces, and the estimation errors.

Figure 13 .
Figure 13.Load estimation error for each finger angle: (a) maximum values, (b) RMS values.The bars show the average values for the three different fingers and the error bars depict the maximum and minimum values.