A one-dimensional touch panel based on strain sensing

Mechatronics xxx (2012) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Mechatronics

journal homepage: www.elsevier .com/ locate/mechatronics

A one-dimensional touch panel based on strain sensing

Chia-Hsing Pi, I-Fang Tsai, Kuang-Shun Ou, Kuo-Shen Chen ⇑Department of Mechanical Engineering, National Cheng-Kung University, Tainan 70101, Taiwan

a r t i c l e i n f o

Article history:Received 4 October 2011Accepted 12 April 2012Available online xxxx

Keywords:Touch sensingStrain gaugeBeam theoryFinite element method

0957-4158/$ - see front matter � 2012 Elsevier Ltd. Ahttp://dx.doi.org/10.1016/j.mechatronics.2012.04.002

⇑ Corresponding author.E-mail address: [email protected] (K.-S. Ch

Please cite this article in press as: Pi C-H et al.j.mechatronics.2012.04.002

a b s t r a c t

A 1-D touch panel based on strain sensing has been designed and realized as a preliminary test protocolfor evaluating smart floor tile design used for indoor localization. By using elementary beam theory, it ispossible to inversely compute the location of the force applied based on the outputs of strain gaugesattached to each end of the touch panel. Detailed 3D finite element analyses are performed to examinethe design concept and to evaluate the effect of boundary conditions and sensor gain differences, as wellas possible strain gauge attachment errors. In typical practical applications, the prediction based on 1-Dbeam mechanics agrees with 3D finite element analysis within 3% error. However, the analytical designapproach breaks down near both beam ends, reducing the effective sensing zone. The experimental dataalso agrees with the analytical results very well. Under an applied force of 1 N, the effective sensing zoneis near 70% and the spatial resolution is between ±0.21 and ±0.37 cm. This resolution is sufficient for afinger touch device and is much higher for smart floor applications. The lessons and conclusion learnedfrom this test stage then serve as the basis for more realistic 2D smart floor tile designs currently under-way for smart building applications.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Touch panels have been widely used as an input interface inmany information science applications for improving the userfriendliness. According to the discrepancy in transduction mecha-nisms, touch sensing can be classified into various categories asresistive [1,2], capacitive [3], ultrasonic [4], and infrared (IR) [5]sensing schemes. Numerous working examples and products havebeen realized (e.g. [6] in healthcare). Using modern microfabrica-tion technology, sensors may be miniaturized and integrated intoa sensing array on a panel for accurate localization.

Strain sensing is a typical resistive sensing scheme and has beenextensively used for structural health monitoring [7,8] and forcetransducers [2,9]. The signal induced by elastic deformation causedby an external load, in conjunction with subsequent Wheatstonebridges and the associated amplification and filtering circuits, isused to measure the local stress/strain and the magnitude of theapplied load. It is also possible to interpret the load-applied loca-tions based on the output of strain gauges. However, comparedto other state-of-the-art transducers, the size of typical straingauges is relatively large. Due to their limited miniaturizationpotential and inability to form high resolution array type elements,it is unlikely that the strain gauge-based touch sensing schemescan be realized for most touch panel applications. However, straingauge-based position sensing may find new applications in

ll rights reserved.

en).

A one-dimensional touch pane

modern smart building related applications [10], where accurateindoor position detection of moving objects and communicationbetween them are critical needs for state of the art smart buildingsand intelligent life applications and various localization schemeshave been proposed, including inertia navigation [11], wiimote IRsensing [12], and received signal strength indicator (RSSI) [13].

In particular, one potential method worth to be explored for in-door localization is to design a ‘‘smart floor’’ to detect the locationof objects. As schematically shown in Fig. 1, this smart floor consistsof numerous floor tiles, each containing a few sensors. Once an objectis located on a tile, its global location can be determined by examin-ing the location of the responding tile and its location within the tile(i.e., the local location) by strain sensing. Compared to the touch pan-els used in portable electronics, this approach requires many largepanels with less positioning resolution. This implies that it wouldbe costly and cannot take advantage of array sensing schemes. As aresult, sensors used for portable electronics may not be ideal candi-dates for this application. By the same token, ‘‘bulky’’ strain gaugesappear to be a promising choice for implementation.

Previously, Schmidt et al. [14,15] proposed a smart floor con-cept by using four supports equipped with individual load cells,which can be treated as an extended version of strain sensing.Using simple force and moment equilibrium in rigid body statics,it is possible to calculate the force-applied location by the mea-sured force signals. This design was successfully realized and dem-onstrated. However, the rigid static model does not allow theexistence of reaction moments at supports (otherwise, the modelbreaks down). In addition, each support must have a load cell to

l based on strain sensing. Mechatronics (2012), http://dx.doi.org/10.1016/

http://dx.doi.org/10.1016/j.mechatronics.2012.04.002

mailto:[email protected]


http://www.sciencedirect.com/science/journal/09574158

http://www.elsevier.com/locate/mechatronics



Fig. 1. The concept of smart floor tile for indoor localization.

2 C.-H. Pi et al. / Mechatronics xxx (2012) xxx–xxx

report the reaction force otherwise the system would be staticallyindeterminate. The former restriction implies that the structure isnot able to resist angular movement and the latter restrictionmeans that the design may require more sensors if more founda-tion supports are required. In this work, based on our primitivework presented [16], instead of using load cells and reaction forcefor localization, a different approach is proposed. In our scheme,localization is accomplished by measuring in-plane strains andsubsequently converting to load-applied location by the mechanicsof the materials. This scheme allows more complicated supportconditions than the previous approach, avoiding their major con-cerns. However, the associated mechanics are much more compli-cated. In most circumstances, no analytical solutions exist andnumerical schemes such as finite element methods must be used.

Fig. 2 shows the fundamental concept of strain-based localiza-tion. Consider a concentration force P applied at a particular

Elastic structure

Strain gauges

s1, (x1,y1)

s2,(x2,y2)sn,(xn,yn)

x

yz

Force P

(xp,yp)

(a)

(b)

Fig. 2. The basic concept of strain-gauge based localization: (a) schematic plot forstrain-gauge touch panel to illustrate the necessary parameters and (b) functionalblock diagram for strain gauge localization.

Please cite this article in press as: Pi C-H et al. A one-dimensional touch panej.mechatronics.2012.04.002

location (xp,yp). It results in a stress/strain distribution over the en-tire structure. This stress/strain information can be monitored at afew locations by attached strain gauges at locations (x1,y1),(x2,y2), . . . , (xn,yn) and their magnitudes depend on the appliedload, the geometry, and the associated boundary conditions. Math-ematically, this is equivalent to an inverse problem for finding theforce applied location (xp,yp) by the known sensor outputs(s1,s2, . . . ,sn), which may not be a trivial task since the mechanicsgoverning this mapping may be complicated and no analyticalsolution is available. In addition, the sensor outputs also dependon the applied force level, a factor which must be excluded for apractical localization scheme. Finally, key performances, such asachievable spatial resolution, also depends on the subsequent cir-cuit design and signal noise level, an issue requiring pre-evaluation.

Although this research is motivated by the needs of indoorlocalization and the ultimate goal is to realize a prototype of smartfloor using strain sensing, in order to demonstrate the conceptualdesign for the proposed methodology and to evaluate the systemperformance with respect to key design and operation parameters,a simplified 1-D touch panel is studied first as a test stage. Throughthe design, implementation, and performance characterization ofthe 1-D panel, important lessons and possible design trade-offsmay be learned, a critical need for guiding the subsequent 2-D floortile design. In addition, the designed 1-D touch panel and the asso-ciated strain sensing scheme may also be applicable for relatedtouch sensing and structural health monitoring applications.

This article presents the design, analysis, and experimentalcharacterization of the 1-D strain sensing-based touch panel in de-tail. In Section 2, the necessary background is addressed and theconceptual design is presented and followed by detailed finite ele-ment analyses presented in Section 3. The implementation andrealization of the system hardware and software are then docu-mented in Section 4, followed by experimental characterizationin Section 5. The important findings and key consequences of thedesign, as well as suggestions for future 2-D floor tile design, arediscussed in Section 6. Finally, Section 7 concludes the paper.

2. Conceptual design

Consider a strain gauge with an initial resistance R0 subjected toan applied elongation strain e0. The resistance change may be ex-pressed as

DR ¼ R0ð1þ 2mÞe0; ð1Þ

where m is the Poisson’s ratio of the material. By incorporating con-stitutive law (e.g., e0 = Er0 for uniaxial loading, where E and r0 areYoung’s modulus and uniaxial stress, respectively), the change inresistance can be found in terms of the applied stress level. Considera half-bridge circuit with all resistors having the same nominalresistance R0 and a Wheatstone bridge under the deflection mode,the relationship between the output voltage DV and the resistancechange DR can be expressed as [17]

DVDR=R0

2ð2þ DR=R0Vcc ð2Þ

For small stress/strain situation, DR� R0 and the resulting stress/strain is linear and proportional to the applied load according tothe theory of linear elasticity.

Now, consider a fixed–fixed beam with a length L subjected to aconcentrated load P, as shown schematically in Fig. 3, with twostrain gauges attached at both ends with the output voltages VA

and VB, respectively. The goal is to find the force applied locationa using the outputs of the two strain gauges. By elementary beam




Fig. 3. Schematic plot of a fixed–fixed beam subjected to a concentrated loading.

C.-H. Pi et al. / Mechatronics xxx (2012) xxx–xxx 3

theory, the bending moment at a particular location x can be ex-pressed as [18]

MðxÞ ¼ �Pab2

L2 þ�Pb2x

L3 ðLþ 2aÞ � Phx� ai1; ð3Þ

where hx � ai1 is the step function representation used in mechan-ics of materials [18]. By Eq. (3), the bending moment at both ends(i.e., at x = 0 and x = L), denoted as MA and MB, may be found by

MA ¼�Pab2

L2 ; ð4aÞ

and

MB ¼�Pa2b

L2 : ð4bÞ

The local stress of a beam is then given by

rðxÞ ¼ MðxÞtI

; ð5Þ

where t and I are the beam thickness and moment of inertia,respectively.

By inserting Eqs. (4) into (5), the stresses at both ends can befound. The strain gauge outputs at both ends are linearly propor-tional to the local stress and can be expressed as

rA ¼ jMa ð6aÞ

rB ¼ jMB ð6bÞ

where j is a proportional constant that depends on beam thickness,inertia, and strain gauge circuitry variables. With some algebraicmanipulation, it can be shown that

VB

VA þ VB¼ MB

MA þMB¼ a

Lð7aÞ

Or, simply

a ¼ LVB

VA þ VBð7bÞ

Using this manipulation, the force applied location can be deter-mined. This result is independent of the applied force level. It is con-venient to define the voltage ratio in Eq. (7a) as a dimensionlessvariable H for use as the performance index, i.e., H � VB/(VA + VB)and a H�, denoted as complementary H, hereafter as

H� ¼ 1�H ¼ VA

VA þ VB; ð8Þ

for the purpose of future performance evaluation. By using H andH�, the location to be found can be expressed as

a ¼ LH ð9aÞ

or

a ¼ Lð1�H�Þ: ð9bÞ

Eq. (7) serves as the fundamental basis of the 1-D touch panel de-sign. Theoretically, using the output of the two strain gauges and


Eq. (7), the load applied location can be determined, independentlyof the level of force applied. However, in reality, this ideal relationfaces several challenges, both mechanical and electrical.

The most fundamental concern is the limitation of the simplebeam equation, which is a highly simplified model from theoryof elasticity. Considerable deviation may occur if the slendernessratio is not sufficient. In addition, the ideal clamped boundary con-dition cannot be realized in real system and the stress near theclamped end of a beam usually exhibits a complicated 3D distribu-tion, which may deviate considerably from the simple beam equa-tion. As a result, it is expected that the ideal relation (i.e., Eq. (7))will break down for a portion of the beam near the clamped edge.Consequently, the effective sensing zone is smaller than the beamlength. Physically, by increasing the slenderness ratio, the effectivesensing zone can be increased. However, this implies that the de-vice size will become either too lengthy or too weak. Finally,attachment and alignment errors, though small, may also affectthe final results. These mechanical issues should be further evalu-ated by finite element analyses before realizing.

The localization scheme is obtained based on the stresses of bothends, the relation will no longer hold once these two sensors have asubstantial difference in sensor gain. Hence, a study may be neces-sary to evaluate the system robustness against possible gain dis-crepancies when developing the design. Meanwhile, the effect ofnoise level should also be examined and its effects on localizationperformance evaluated. Finally, although the ideal relationship(i.e., Eq. (7)) indicates that the localization scheme is independentof the applied load, it is expected that the performance of thescheme will depend on the applied load level. These issues shouldbe clearly characterized via subsequent experimental studies.

3. Finite element analyses

In order to address the key mechanical concerns mentionedabove, two finite element (FE) models were created and analyzedusing the general-purpose finite element package ABAQUS STAN-DARD V6.10 [19]. As shown in Fig. 4, the first model, model A, mod-els the beam using 49,581 nodes and 10,000 C3D20 3D solidcontinuum elements (200 (length) � 10 (thickness) � 5 (width)).Both ends are ideally clamped. The material for the structure isassumed to be a polymer with a Young’s modulus of 2.5 GPa, aPoisson’s ratio of 0.4 and a density of 1200 kg/m3. The strain gaugeoutputs are obtained by examining the longitudinal stress reportedat two small areas near the beam ends. A vertical force is applied atthe mid-line of the beam at the upper surface. In order to furtherinvestigate the boundary compliance effect, a second model (mod-el B hereafter) was established, which adapts the same mesh andmechanical properties of model A with a spring supported bound-ary conditions to mimic the possible compliance from support. Keysimulation results are addressed below. For ease of discussion, theload-applied position is normalized by the beam length.

3.1. Comparison between the 1-D beam and the 3-D FE models

The first issue is to examine the fundamental difference be-tween the 1-D beam and the 3-D FE models using model A. The re-sults are shown in Fig. 5. It can be seen that both results agree witheach other very well for the load-applied position between 0.05and 0.95. The output index difference between the beam modeland the 3-D FEA model is less than 3% within the linear zone, val-idating the original concept. The output index also appears to beindependent from the applied force level in 3D finite element anal-ysis. This is not surprising since the finite element model is alsobased on linear elasticity and the practical sensor resolution isnot included. Finally, for loads applied near the ends, the boundary




E11: normal strain in 1 (x) direction

E33: normal strain in 3 (z) direction

E13: shear strain between 1 and 3 axes

Fig. 4. Mesh plot of the finite element model.

Fig. 5. The relationship between output index H and the touch location obtainedvia finite element simulation.


effects gradually dominate and the 1-D results deviate from the 3Dsimulation as expected.

Fig. 6. (a) The effect of sensor gain difference in the strain gauges in the outputindex H and (b) the percentage errors in H and H� with respect to force location.

3.2. Effect of strain gauge gain differences

Since the only measurable variable for stress/strain is the out-put voltages of the individual strain gauges and the conversion for-mula (i.e., Eq. (7)) is developed under the assumption that both thesensor gains are identical, it is important to examine the robust-ness of the system once the two sensor gains are not identical.The results are shown in Fig. 6a. It can be seen that the output in-dex H offsets slightly once a sensor gain difference exists. How-ever, the influence is not significant. For instance, even for a 25%difference between the two sensor gains, the resulting offset isonly approximately 5% and is essentially constant. However, ifthe results are normalized by the presented output index level,i.e., (DH/H), the story is slightly different. It can be seen that foroutputs with a smaller H, the influence of the gain error becomessignificant due to the small H in the denominator. Since H and H�

are complementary, this concern can be solved by replacing H byH� for H less than 0.5 and using Eq. (9b) for conversion. The shiftof output variable from H to H� can be performed automatically intypical data acquisition (DAQ) systems such as LabView. Using thisapproach, the maximum error occurs if the load is applied at thebeam center and is about 10% for a 25% gain difference, as shownin Fig. 6b. Note that the gain difference levels in this study (i.e.,10–25%) are much larger than what should occur in real applica-tions. This indicates that the gain error should not be a practicalconcern in real service.


3.3. Effect of strain gauge placement and alignment errors

The stress/strain of a beam depends on the bending moment,which is a function of position. As a result, misplacement of a straingauge may result in strain gauge output error. Nevertheless, bysimple error propagation analysis, it can be shown that the straingauge output error (in percentage) is proportional to the position-ing error (in percentage). Practically, the positioning error shouldbe less than 2 mm. For a beam with a characteristic length of31 cm, this error is not significant. This argument is verified bysubsequent finite element analysis (Fig. 7a).




Fig. 7. The effect of strain gauge misplacing: (a) the influence of strain gaugeplacement error and (b) the influence of strain gauge alignment error.

Fig. 8. The effect of boundary compliance on the output index.


From the experimental perspective, it is difficult to align a straingauge exactly parallel to the attached beam underneath. As a re-sult, the misalignment of angles is also investigated. By stress/strain transformation formula (or Mohr’s circle) [18], the misa-ligned stress/strain components can be obtained by giving therotation angle. Results are shown in Fig. 7b. For a misalignment an-gle below 10�, which is much larger than the maximum alignmenterror expected, the resulting variation in H is still very small.

3.4. Effect of boundary compliance

Perfect clamped boundary conditions are not realistic and thepossible boundary compliances may need to be examined byreplacing the clamped boundary with a spring supported boundary(i.e., model B) and the results are shown in Fig. 8. It can be seen thatH exhibits considerable difference around the boundary regions.However, for the portion away from the boundary, the differenceis clearly acceptable for reasonable boundary stiffness.

The above analysis results indicated that the 1-D beam modelused for conceptual design is reasonable and the resulting errorsare acceptable for normal operating conditions. Since the errorresulting from any single component is small, by the root-sum-square (RSS) approach for dealing with error propagation [20],the overall error will be of the same order.

4. System implementation

Based on the conceptual design and the associated finiteelement analyses, the 1-D touch panel is implemented, shown


schematically in Fig. 9a. A plastic beam with a length of 31 cm(with an effective beam length L of 28.5 cm by subtracting thelength for clamping), width of 2 cm, and thickness of 1.2 mm is se-lected as the main structure. Both ends of the beam are clamped bytwo aluminum blocks. Two Kyowa KFG-1-350-C1-11 single axisstrain gauges with a nominal resistance of 350 X are attached nearthe ends of the beam. Two commercially available Kyowa DPM-612A bridge circuit/amplifiers are used to detect and process theoutput signal. The signals are then filtered by a 50 Hz low pass fil-ter before they are acquired by a NI-USB 9215 BNC system with asampling rate of 1 kHz under the LabView environment. Further-more, a user interface written by LabView is also created for pro-viding a better user-machine interaction and visualization.Fig. 9b shows the system pictures, as well as the user-machineinterface of the 1-D touch panel detail shown in Fig. 9c. Noticedthat the length of the strain gauge is 4.8 mm. This finite size willbe considered for evaluation of the panel performance below.

5. Experiments

5.1. Effect of loading magnitude

The effect of the magnitude of the applied load is examined firstsince this issue cannot be evaluated by both analytical and finiteelement models. Several poise weights ranged from 50–150 g(i.e., 0.5–1.5 N) are applied at different locations. Fig. 10a showsthat the results are essentially linear for x/L between 0.2 and 0.8.In this work, x/L = 0 and 1 are corresponding to the left and rightclamped ends. A linear regression analysis indicates that the rela-tionship between H and the force applied location x/L for a loadof 1.5 N can be expressed as

H ¼ 0:9822xLþ 0:0469: ð10Þ

By the same token, at the ends, H varies significantly and is not con-sidered useful information for practical applications. It is possible touse the case with a load of 1.5 N as the referenced value for evalu-ating the effect of the applied loading magnitude. For simplicity, letus define an error index eL(P) as

eLðPÞ ¼Hload¼PN �Hload¼1:5N

Hload¼1:5N� 100%; ð11Þ

which represents the deviation in H away from the nominal value.The results shown in Fig. 10b indicate that the response indeed de-pends on the applied load level. It can be seen that all error indexesare essentially small if the force is applied around the beam center.




Fig. 9. (a) the schematic functional plot of the 1-D touch panel. (b) a typical hardware picture and (c) the user interface and practical operation.


Nevertheless, it remains clear that a larger applied load yields asmaller error and a slightly larger applicable range. However, asthe loading location moves toward either boundary, the error in-creases gradually and the smaller the applied load, the larger theresulting error. The above observations are reasonable based onthe argument of sensor sensitivity. That is, when the load is small,the structure works more like a 3D continuum rather than a simplebeam. The sensitivity to the loading decreases as the applied forcemoves away from the center. In addition, the signal to noise ratiodecreases with a reduction in the applied load. In this design exam-ple, a minimum touch force of 1 N is recommended to yield anacceptable response.

5.2. Effective sensing zone

As explained earlier, the entire span cannot act as the sensingzone due to the boundary effect and the finite size of the straingauges. As a result, it is important to evaluate the effective sensingzone for different loading levels. In this study, the effective sensingzone is defined as the zone where the associated eL is less than 10%,which is a reasonable choice.

Fig. 10c shows that the effective sensing zone increases with theapplied loading level. For example, an applied force of 0.5 N, theeffective sensing zone is only 50%. Once the force is increased to1.3 N, this zone can be increased to 75%. Note that the values ofthe effective sensing zone also depend on the allowed error level.


A stricter requirement for allowable error will significantly reducethe effective sensing zone, and vice versa.

5.3. Device resolution

In order to evaluate the spatial resolution, the time response ofthe obtained signal must be studied first to extract the correspond-ing signal noise level. This will be followed by an error propagationanalysis. The entire process is shown schematically in Fig. 11. Ini-tially, the strain gauge outputs are treated as random variables.With a constant load applied, the mean value and the standarddeviation of both strain gauge outputs are calculated. By standarderror propagation scheme, the mean values and standard devia-tions of H and H� at different locations as well as different forcelevels can be obtained. Statistically, it is possible to cover 95% ofthe uncertainty of the measured H by replacing a given H witha range of ±two standard deviations (r). By using Eq. (10), the po-tential spatial uncertainty or resolution of the touch panel due tothe presence of strain gauge noise may be characterized.

For an applied force of 1.5 N, the standard deviations of thestrain gauge outputs are approximately 0.011–0.013 V. By stan-dard error propagation scheme, the standard deviation of H is cal-culated as 0.004 (applied at center) – 0.01 (applied near both ends).Finally, a complete H versus location data set with the possibleuncertainties included is shown in Fig. 12, in which the resolutionat the center is approximately 0.42 cm (or ±0.21 cm) and is approx-imately 0.73 (or �±0.37 cm) at locations near the panel edge.




Fig. 10. The experimental results of the touch panel: (a) the relation between theoutput index, force location, and applied force magnitude, (b) the percentage errorof H and H� w.r.t. applied load level and force location, and (c) the relationshipbetween the effective sensing zone of the touch panel and the applied force level.

Θ

x/L

Θmean

Θmean +2σ

Θmean -2σ

Θ0

x0x0 - x x0+ ΔΔ x

95% signal confidence level

Localiza�on uncertainty

Fig. 11. Schematic plot of determining the device spatial resolution.

Fig. 12. The experimental output index with 95% confidence level versus loadapplied locations.


However, there is a trade-off between the confidence level and thedevice resolution. The above result is based on a ±2r (i.e., 95% con-fidence level) estimation [20]. Depending on the targeted applica-tions, a further reduction in confidence level could increase theapparent spatial resolution and vice versa.

6. Discussion

6.1. Performance summary

In this work, a 1-D touch panel based on strain sensing is de-signed and realized as a preliminary test protocol for smart floortile design used for indoor localization. By using elementary beam


theory, it is possible to inversely compute the force applied loca-tion based on the output of two strain gauges attached at the endsof the touch panel. Detailed 3D finite element analyses are per-formed to examine the design concept. The 1-D beam mechanicscalculation agrees quite well with the 3D finite element analysisand the experimental data agree with analytical results very well.However, the design approach breaks down near the beam endsand the effective sensing zone is reduced in reality. Under an ap-plied force of approximately 1 N, the effective sensing zone is near70% and the spatial resolution is between ±0.21 and ±0.37 cm. Thisresolution is sufficient for a finger touch device and is much higherfor smart floor applications. Meanwhile, instead of using analyticalbeam model, one may also like to use the simulation results from3D FEA as the conversion basis for localization for further enhanc-ing the effective sensing zone.

6.2. Contributions

For touch sensing devices, this strain gauge sensing designprovides a simple, low cost solution for applications without highprecision requirement. Modern state-of-the-art touch panels usearray-type sensing elements built into the entire touch panel areato detect the touch location. While that approach offers remarkablespatial resolution for many portable devices, it may not becost-effective for applications requiring much larger areas. Thestrain-gauge based design only needs a few strain gauges and theassociated bridge and amplification circuits. Although the spatial




Fig. 13. Possible extension of the 1-D touch panel design: (a) 2-D strain gauge touch panel, (b) 2-D touch panel by combining several 1-D design.


resolution may not match an array-based touch panel, the signifi-cant reduction in cost and system complexity coupled with a rea-sonable resolution implies that strain gauge-based designs couldbe advantageous in larger area applications such as indoor localiza-tion. The experimental data indicate that the performance of thestrain sensing scheme should be sufficient for addressing the needsof the abovementioned larger area applications. To the best of ourknowledge, this is the first detailed performance investigation ofstrain gauge touch panel via analytical, simulation, and experiment.In addition, the conclusion drawn from the mechanical analysis maybe useful for other applications related to strain gauge sensing. As aresult, we believe that this work provides useful information for thefields of touch sensing, strain gauge measurement, and smart build-ing related applications.

6.3. Perspective and next generation design

One natural next step of this work is to extend the 1-D touchpanel to a 2-D touch panel, work that is currently underway. Asshown in Fig. 13a, instead of using a 1-D beam, a rectangular plateis designed and supported along all four edges. In this design,strain gauges are mounted at the ends of the mid-lines. The load-applied position may be obtained by the measured strain usingplate theory [21]. Originally, the idea was to use an individual pairof strain gauges to detect the location in x and y directions sepa-rately. However, it was found that the coupling between the twoaxes may not be ignored. Thus, this approach could yield consider-able error or lead to poor spatial resolution. To address this issuewithout adding hardware complexity, a novel recursive algorithmis proposed to improve the spatial resolutions based on the inter-action of the two sets of data. Refer to [22] for our preliminary re-sults. By the same token, it is also possible to design a 2-D planecomprising various 1-D sensing elements demonstrated in thiswork for certain specific applications (Fig. 13b). Using this ap-proach, the presented 1-D design can be fully exploited.

High quality bridge circuits and amplifiers may be far moreexpensive than the strain gauges themselves. The present designconnects each strain gauge with its own bridge and amplifier, whichis not feasible for the entire indoor localization, where many floortiles are required. Nevertheless, since the sampling rate requiredis low (on the order of 10 Hz or less), this implies that the systemneeds only two bridge/amplifier units, along with additional multi-plexers and switches to sequentially process the signal with a rea-sonable fast sampling rate (e.g., 10 kHz). This will significantly


reduces the cost of the most expensive component while maintain-ing performance.

Another advantage of strain sensing is that the surface may notneed to be flat. Theoretically, the mapping between the appliedforce and its location to the response of stress/strain at particularlocations is unique once the structures are defined. This feature im-plies that it is not necessary to design a floor in the traditional waydue to the limitations of the sensing system, meaning that revolu-tionary smart architectural designs are possible.

7. Summary and conclusion

Touch sensing plays an important role in modern human-ma-chine interfacing for intelligent life applications. In this work, a 1-D touch panel based on strain sensing is designed and realized asa preliminary test protocol for smart floor tile design used for indoorlocalization. The force-applied location is obtained by the measuredstrain gauge signals with the conversion from elementary beammechanics. Three-dimensional finite element analyses are also per-formed to validate the design concept. We find that in typical prac-tical applications, the 1-D beam mechanics calculation agrees quitewell with the 3D finite element analysis, but the design approachbreaks down near the beam ends and the effective sensing zone isreduced. Further, the experimental data also agree with analyticalresults very well. In particular, under an applied force of 1 N, theeffective sensing zone is near 70% and the spatial resolution is be-tween ±0.21 and ±0.37 cm and may be further improved by increas-ing the touch load or by reducing the noise level. This resolution issufficient for a finger touch device and is sufficiently high forupcoming smart floor applications.

However, it must be pointed out that the success of the conceptproposed in this work relies on the accuracy of solid mechanicsconversion models. In most situations, no analytical solutions areavailable and extensive finite element analyses must be performedin order to find the accurate conversion. Nevertheless, the lessensand conclusions of this test stage may serve as the basis for smartfloor tile designs currently underway for low cost and reasonableaccurate indoor localization systems for smart building applica-tions in the future.

Acknowledgements

This work is supported by National Science Council under Con-tract Numbers: NSC 99-2815-C-006-107-E and NSC 98-2815-C-006-041-E for undergraduate research.





References

[1] DeVisser BC. Towards an industry standard for resistive touch panel testing. In:Proceedings of the third Americas display engineering and applicationsconference, ADEAC; 2006, p.63–4.

[2] Kim D-K, Kim J-H, Kwon H-J, Kwoi Y-H. A Touchpad for force and locationsensing. ETRI J 2010;32:722–8.

[3] Kim H-K, Lee S, Yun K-S. Capacitive tactile sensor array for touch screenapplication. Sens Actuat A: Phys 2011;165:2–7.

[4] Katsuki T, Nakazawa F, Sano S, Takahashi Y, Satoh Y. A compact and high opticaltransmission SAW touch screen with ZnO thin-film piezoelectric transducers. In:Proceedings of the IEEE ultrasonics symposium, vol. 1; 2003. p. 821–24.

[5] Walker G. A cornucopia of touch technology. Inform Display 2006;22:14–20.[6] Connolly C. Touch-sensitive skins for Japanese health care robots. Sensor Rev

2009;29:104–6.[7] Tegtmeier FL. Strain gauge based microsensor for stress analysis in building

structures. Measurement 2008;41:1144–51.[8] De Backer H, Outtier A, Van Bogaert P. High precision strain gauge

measurements in areas of high stress concentrations of orthotopic platedbridge decks. Insight-Non-Destruct Test Condition Monitor 2007;49:384–9.

[9] de Silva JG, de Carvalho AA, de Silva DD. A strain gauge tactile sensor for finger-mounted applications. IEEE Trans Instrum Measur 2002;51:18–22.

[10] Jeng T. Advanced ubiquitous media for interactive space. a framework. In:Computer aided architecture design future, Part 6. London: Springer; 2005. p.341–50.

[11] Evennou F. Advanced integration of WiFi and inertial navigation systems forindoor mobile positioning. EURASIP J Appl Signal Process 2006;2006:11.

[12] Chen P-W, Ou K-S, Chen K-S. IR indoor localization and wireless transmissionfor motion control in smart building applications based on Wiimote technology.


In: Proc SICE 2010, 49th annual conf of the society of instrument and controlengineers of Japan; 2010. p. 1781–85.

[13] Sugano M. Indoor localization system using RSSI measurement of wirelesssensor network based on ZigBee standard. Proceedings of the sixth IASTEDinternational multi-conference on wireless and optical communications; 2006.p. 503–8.

[14] Schmidt A, van Laerhoven K, Friday A, Gellersen H-W. Context acquisitionbased on load sensing. In: Proc 4th international conf on ubiquitous computing(UbiComp 2002), Goteborg, Sweden; 2002. p. 333–50.

[15] Schmidt A, Strohbach M, van Laerhoven K, Gellersen H-W. Ubiquitousinteraction – using surfaces in everyday environments as pointing devices.Lecture Notes Comput Sci 2003;2615:263–79.

[16] Pi C-H, Ou K-S, Chen M-H, Chen K-S. Analysis, simulation, and experimentalinvestigations of a one dimensional touch panel based on strain sensing. In:Proc 50th annual conf of the society of instrument and control engineers ofJapan (SICE 2011), September 2011, Tokyo, Japan. p. 1954–59.

[17] Figliola RS, Beasley DE. Theory and design for mechanical measurements. 4thed. New York: John Wiley & Sons; 2006.

[18] Hibbeler RC. Mechanics of materials. 6th ed. Pearson Prentice Hall; 2005.[19] ABAQUS STANDARD user manual. Version 6.10, Stimula; 2009.[20] Beckwith TG, Marangoni RD, Lienhard JH. Mechanical measurements. 5th

ed. Addison Wesley Longman; 2004.[21] Timoshenko SP, Woinowsky-Krieger S. Theory of plates and shells. McGraw-

Hill International Edition; 1959.[22] Pi C-H, Tsai I-F, Ou K-S, Chen K-S. Development of intelligent floor based on

strain sensing. In: 27th National conference of the chinese society ofmechanical engineers, Taipei, Taiwan; December 2010.




A one-dimensional touch panel based on strain sensing

Documents