vibration

Article Accelerometer Based Method for Tire Load and Slip Angle Estimation

Kanwar Bharat Singh 1,* and Saied Taheri 2

1 Tire Intelligence, The Goodyear Tire & Rubber Company, Colmar-Berg L-7750, Luxembourg 2 Department of Mechanical Engineering, Virginia Tech, Blacksburg, VA 24061, USA; [email protected] * Correspondence: [email protected]



Received: 9 February 2019; Accepted: 24 April 2019; Published: 28 April 2019


Abstract: Tire mounted sensors are emerging as a promising technology, capable of providing information about important tire states. This paper presents a survey of the state-of-the-art in the field of smart tire technology, with a special focus on the different signal processing techniques proposed by researchers to estimate the tire load and slip angle using tire mounted accelerometers. Next, details about the research activities undertaken as part of this study to develop a smart tire are presented. Finally, novel algorithms for estimating the tire load and slip angle are presented. Experimental results demonstrate the effectiveness of the proposed algorithms.

Keywords: smart tire system; signal processing; tire load; tire slip angle; accelerometer

1. Introduction The application of sensor technology in tires has been extensively studied over the last decade by many researchers [1–6]. The stimulus for the research and application of sensor technology in tires was the Bridgestone/Firestone recall of 14.4 million tires in 2000 [7]. Because of the recalls, US legislators included tire pressure monitoring (TPMS) as part of its TREAD Act [8–10]. Thereafter, European countries (via United Nations regulation UNECE-R64) adopted the TPMS legislation for new models beginning in November 2012, and for all vehicles beginning in November 2014. South Korea has adopted a similar TPMS legislation to that of Europe. Likewise, countries such as Japan, China and India are in the process of investigating TPMS technology as well. Although classical TPMS sensors are mounted on the valve stem inside the wheel, there is a great level of interest in placing the sensor directly on the tire. A tire mounted sensor will not only fulfill the basis TPMS functionalities, i.e., measure cavity air pressure and temperature, but also presents opportunities to sense key attributes related to the tire contact patch. This would be achieved through the inclusion of an additional sensing element (e.g., an accelerometer) and advanced digital signal processing (DSP) algorithms. Running sophisticated DSP algorithms requires advanced low power microchips. The Internet of things (IoT) digital revolution is resulting in an exponential growth in the computing power of embedded microchips thus giving great impetus to the development of smart tire technology. Large-scale usage of IoT sensors across several industries is driving the economies of scale. The availability of a low-cost sensing solution becomes more critical in the case of tires, knowing they are largely a commoditized product. Throughout this paper, the term smart tire technology explicitly refers to a configuration wherein the sensor is mounted on the tire inner liner. The smart tire is expected to have capabilities to monitor in real-time the tire forces, road surface conditions, tire slip conditions, temperature, pressure and the tire tread depth (Figure1). Being able to leverage a plethora of information, a smart tire is expected to stimulate the development of

Vibration 2019, 2, 174–186; doi:10.3390/vibration2020011 www.mdpi.com/journal/vibration Vibration 2019, 2 175

advanced traction, braking and stability control systems for improving the vehicle safety and theadvanced advancedperformance traction, traction, [11–14]. braking braking and and stability stability control control systems systems for for improving improving the the vehicle safetysafety andand performanceperformance [[11–14].11–14].

Figure 1. Smart tire system sensor output. FigureFigure 1.1. Smart Smart tire tire system system sensor sensor output. output. There have been several architectures proposed in the past few years pertaining to smart tires There have been several architectures proposed in the past few years pertaining to smart tires (FigureThere have2): been several architectures proposed in the past few years pertaining to smart tires (Figure2): (Figure 2): (i) The APOLLO program—Developed a 3-in-l intelligent tire [15]; (i) The(i) (ii) APOLLOThe TheAPOLLO FRICTI@N program—Developed program—Developed program—On-board a 3-in-l a 3-in-l intelligent system intelligent for tire measuring [ 15tire]; [15]; and estimating tire-road friction (ii) The(ii) FRICTI@NThe [16];FRICTI@N program—On-board program—On-board system system for measuring for measuring and estimatingand estimating tire-road tire-road friction friction [16]; (iii) Contact(iii)[16]; areaContact information area information sensing (CAIS) sensing system (CAIS) being system developed being developed by Bridgestone by Bridgestone [17]; [17]; (iv) Cyber(iii) (iv)Contact Tire Cyber being area Tire developed information being developed by sensing Pirelli by Tires (CAIS) Pirelli [18 sy]; Tiresstem [18]; being developed by Bridgestone [17]; (v) Intelligent(iv) (v)Cyber Intelligent tireTire system being tire beingdeveloped system developed being by Pirelli developed by ContinentalTires by[18]; Continental AG [19]; AG [19]; (vi) Intelligent(v) (vi)Intelligent Intelligent tire system tire tiresystem being system being developed being developed developed by Hyundai by Continentalby Hy Motorundai Co, AG Motor Mando [19]; Co, Corp, Mando and Corp, Corechips and Corechips [20];

(vii) Goodyear(vi) Intelligent[20]; Dunlop’s tire system chip-in-tire being [ 21developed]. by Hyundai Motor Co, Mando Corp, and Corechips (vii)[20]; Goodyear Dunlop’s chip-in-tire [21]. (vii) Goodyear Dunlop’s chip-in-tire [21].

FigureFigure 2. Smart 2. Smart tire system—products tire system—products and concepts.and concepts. Figure 2. Smart tire system—products and concepts. The keyThe valuekey value proposition proposition for thefor usagethe usage of smart of smart tires tires in vehicles in vehicles is the is potentialthe potential they they offer offer in in improvingimprovingThe key performance value performance proposition of vehicle of vehicle for control the co usagentrol systems. ofsystems. smart This Thistires would wouldin bevehicles of be even of is even morethe morepotential relevance relevance they in the offer in case the in case ofimproving autonomousof autonomous performance vehicles vehicles (AVs), of vehicle (AVs), considering co considntrol eringsystems. their their high This high safety would safety requirements. be requirements. of even more relevance in the case of autonomousThisThis paper paper vehicles first first presents (AVs),presents a comprehensiveconsid a comprehensiveering their overview high overview safety of therequirements. of state-of-the-artthe state-of-the-art in the in fieldthe field of smart of smart tire technologytireThis technology paper and first thereafterand presents thereafter presentsa comprehensive presents details details of overview the of work the work undertakenof the undertaken state-of-the-art at the at Center the Centerin the for Tirefield for Tire Research of smart Research (CenTiRe)tire technology(CenTiRe) to develop to and develop thereafter a smart a smart tire. presents tire. details of the work undertaken at the Center for Tire Research (CenTiRe)The mainThe to maindevelop contributions contributions a smart of tire. this of paperthis paper are as are follows: as follows: The •main Literature contributions review of of this the paper different are assignal follows: proces sing techniques used by researchers to extract Literature review of the different signal processing techniques used by researchers to extract • • Literaturemeaningful review information of the different from signal tire proces mountedsing techniques sensors, specificallyused by researchers from tire to extractmounted meaningfulmeaningfulaccelerometers. information information from tire from mounted tire sensors,mounted specifically sensors, fromspecifically tire mounted from accelerometers.tire mounted •accelerometers. Details about research activities undertaken as part of this study to develop a smart tire. • Details about research activities undertaken as part of this study to develop a smart tire. Vibration 2019, 2 176

Details about research activities undertaken as part of this study to develop a smart tire. • Novel algorithms for estimating the tire load and slip angle. • 2. Literature Review There are several signal processing algorithms proposed in literature for estimating the tire load and slip angle. Table1 presents a summary of the di fferent techniques proposed by researchers.

Table 1. State-of-the-art literature review—analysis of tire accelerometer data.

Estimated State Signal Used Underlying Physics Reference Use an empirical model between tire contact patch [22–25] length and tire load Use an empirical model correlating the radial [26] acceleration amplitude to the tire load Use an empirical model to describe the shape of the radial acceleration signal—Vertical load is treated [27] as an unknown parameter and is estimated used an EKF observer Use Principal Component Analysis, to describe smart tire signal variance by means of a few linear [28] projections. Projections are correlated linearly with Tire vertical force Radial acceleration the contact forces Use a model between tire contact patch length, tire load and the tire operation conditions (toe) for a [29] respective tire type and/or respective physical properties of the tire Assess the real-time applicability of a tire vertical force estimator on a [30] standard platform used for rapid control prototyping on vehicles Use a model correlating the tire ground contact time to tire load, where the contact time is extracted [31] from the differential acceleration waveform Monitor change in the footprint length at the Radial acceleration centerline and shoulder positions to estimate the [32] tire slip angle Tire slip angle Correlate lateral acceleration amplitude with the [33] tire slip angle Lateral acceleration Extract lateral deformation from the acceleration [34] signal and correlate that with the tire slip angle

In view of the above literature review, the following conclusions can be drawn:

Among the three axes, the radial acceleration signal has been exploited the most by researchers. • Most of these applications require data to be sampled at high rates. For instance, for tire-road • friction sensing, the sensor signal needs to be sampled at 5–10 kHz. Both time and frequency domain signal processing techniques have been extensively used. • Current tire pressure monitoring (TPMS) chips available in production vehicles cannot support • such complex signal processing algorithms. Hence, an application specific IC would to be required. Since these sensor systems are expected to survive the tire service life, they would require • a power source capable of supporting the high sampling rate data processing and high frequency data transmission. Vibration 2019, 2 177

3. Development of a Smart Tire System Based on their proven reliability for tire application, it was decided to use the tri-axial micro electro mechanical systems (MEMS) accelerometers in this study (Table2). These sensors o ffer the advantage of their small dimension, low weight, high reliability, robustness and reliable performance even in harsh and hostile environments.

TableTable 2. 2.Tri-axial Tri-axial accelerometer accelerometer characteristics characteristics [35 [35].].

AccelerometerAccelerometer Characteristics Characteristics ± Range Range 5000 g 5000 g ± Sensitivity Sensitivity 1 mv/g 1 mv/g Frequency responseFrequency response 1.2–10 kHz 1.2–10 kHz Resonance frequencyResonance frequency 30 kHz 30 kHz Mass Mass 3.0 g 3.0 g Dimensions 17.7 9.02 9.14 mm Dimensions × × 17.7 × 9.02 × 9.14 mm

TheThe smart smart tire tire was was developed developed by by placing placing the the tri-axial tri-axial accelerometer accelerometer on on the the inner inner liner liner of of a a tire tire (Figure(Figure3a). 3a). The The accelerometer accelerometer was was placed placed on on the the centerline centerline of of the the tire tire footprint. footprint. The The accelerometer accelerometer wireswires were were connected connected to to a a slip slip ring ring placed placed at at the the wheel wheel center. center. Figure Figure3b 3b shows shows the the final final assembly assembly of of thethe instrumented instrumented tire tire with with the slipthe ring.slip Wiresring. Wires from the from slip the ring slip were ring connected were connected to the data acquisitionto the data system.acquisition Extensive system. on-road Extensive tests on-road were conducted tests were usingconducted the in-house using the mobile in-house tire mobile test rig tire shown test inrig Figureshown3c. in Figure 3c.

(a) (c)

(b)

FigureFigure 3. 3.Smart Smart tire tire application: application: ( a(a)) Sensor Sensor mounted mounted on on the the tire tire footprint footprint centerline; centerline; ( b(b)) instrumented instrumented tiretire assembly assembly with with a a high-speed high-speed slip slip ring; ring; (c ()c mobile) mobile tire tire test test rig. rig.

ItIt is is noteworthy noteworthy to to mention mention that that the the tri-axial tri-axial accelerometer accelerometer was was enclosed enclosed in in latex latex (rubber) (rubber) and and thethe latex latex part part was was glued glued to to the the inner inner liner. liner. This This formed formed a a stronger stronger bond bond with with the the tire tire and and reduced reduced the the chanceschances of of the the accelerometer accelerometer coming coming loose loose when when the tirethe istire rotated is rotated even even at high at speeds.high speeds. In addition, In addition, duct tapeduct (in tape combination (in combination with cyanoacrylate with cyanoacrylate adhesive) wasadhesive) used to was secure used the to wires secure of the the accelerometers wires of the insideaccelerometers the tire. This inside prevented the tire. the This wires prevented from pulling the onwires the from accelerometer pulling on and the yanking accelerometer it out while and theyanking tire was it out rotating. while the tire was rotating. AnAn example example of of the the measured measured acceleration acceleration signal signal (unfiltered) (unfiltered) for for one one wheel wheel turn turn is is shown shown in in FigureFigure4. 4. These These measurements measurements were were done done on on a 17-incha 17-inch SUV SUV tire tire (size: (size: P245 P245/65R17)./65R17).

Vibration 2019, 2 178

(a) X axis (b) Y axis (c) Z axis

FigureFigure 4.4. Signal for one wheel turnturn fromfrom aa tiretire attachedattached tri-axialtri-axial accelerometer.accelerometer.

SharpSharp peakspeaks inin thethe circumferentialcircumferential andand radialradial accelerationacceleration signalssignals areare indicatorsindicators ofof thethe leadingleading andand trailingtrailing edgesedges ofof thethe tiretire footprint.footprint. MoreMore explanationsexplanations aboutabout thethe measurementmeasurement setupsetup andand thethe rawraw accelerationacceleration signalssignals recordedrecorded cancan bebe foundfound inin ourour priorprior publicationpublication [ 36[36]].. TheThe tiretire load an andd slip slip angle angle estimation estimation were were treated treated as asthe the lead lead application applicationss for this for thisstudy. study. The Themotivation motivation for considering for considering these these tire tirestates states is twofold: is twofold: • First, with information about individual tire loads from the tire mounted sensors, the vehicle First, with information about individual tire loads from the tire mounted sensors, the vehicle • inertial parameters, namely, vehicle mass, yaw moment of inertia and axle distances from inertial parameters, namely, vehicle mass, yaw moment of inertia and axle distances from the the vehicle center of gravity, can be precisely estimated, thus making the vehicle controller vehicle center of gravity, can be precisely estimated, thus making the vehicle controller robust robust against variations in vehicle model parameters. against variations in vehicle model parameters. • Second, it is well known in literature that the vehicle sideslip angle can significantly reduce Second, it is well known in literature that the vehicle sideslip angle can significantly reduce the • the risk of accidents through the effective design and implementation of advanced chassis risk of accidents through the effective design and implementation of advanced chassis control control systems. However, in production vehicles, the sideslip angle is difficult to measure systems. However, in production vehicles, the sideslip angle is difficult to measure within the within the desired accuracy level because of high costs and other associated impracticalities. desired accuracy level because of high costs and other associated impracticalities. Having a direct Having a direct measurement of the tire slip angle from a smart tire would enable the design measurement of the tire slip angle from a smart tire would enable the design of novel control of novel control systems using the tire slip angle as a feedback signal instead of the vehicle systems using the tire slip angle as a feedback signal instead of the vehicle sideslip angle. sideslip angle. SpecificSpecific detailsdetails of of the the signal signal processing processing algorithms algorithm developeds developed as partas part of thisof this study study are givenare given in the in subsequentthe subsequent subsections. subsections.

4.4. AlgorithmAlgorithm DetailsDetails

4.1.4.1. TireTire LoadLoad EstimationEstimation AA commoncommon approachapproach proposedproposed inin literatureliterature toto estimateestimate thethe tiretire loadload isis toto exploitexploit thethe relationshiprelationship betweenbetween tiretire load load and and the the tire tire contact contact patch patch length. length. A parametricA parametric study study was was conducted conducted to quantify to quantify the influencethe influence of tire of load tire andload other and other relevant relevant parameters parameters on the on tire the contact tire contact patch length. patch length. The patch The length patch waslength extracted was extracted from the from tangential the tangential acceleration acceleration signal by identifyingsignal by identifying the leading the and leading trailing and edge trailing of the tireedge using of the a peaktire using detection a peak algorithm. detection More algorithm. details regardingMore details the regarding patch length the extraction patch length algorithm extraction can bealgorithm found in ca an previous be found publication in a previous by thepublication authors [by13 ].the authors [13]. AsAs shownshown inin FigureFigure5 5,, the the tire tire load load and and inflation inflation pressure pressure were were seen seen to to have have a a significant significant impact impact onon thethe patchpatch lengthlength ofof thethe tire.tire. • Higher load results in a longer contact patch length • Higher inflation pressure results in shorter contact patch length Vibration 2019, 2 179

Higher load results in a longer contact patch length • Higher inflation pressure results in shorter contact patch length • These results are in-line with results previously published in literature. These results are in-line with results previously published in literature.

FigureFigure 5. 5.Impact Impact of of load load and and inflation inflation on on the the tire tire contact contact patch patch length. length Tire. Tire Size: Size: P245 P245/65R17/65R17. . Figure 5. Impact of load and inflation on the tire contact patch length. Tire Size: P245/65R17. TheseWhat results was not are so in-line unexpected with results and previouslyin fact not captured published in in the literature. published literature is the strong correlationWhatWhat was was seen not not between so so unexpected unexpected the tire and wearand in instate fact fact not(i.e. not, captured remainingcaptured in in tread the the published publisheddepth) and literature literature the tire is contactis the the strong strong patch correlationcorrelationlength (Figure seen seen between 6 between). Lowering the the tire thetire wear treadwear state statedepth (i.e., (i.e. remaining results, remaining in tread a dec tread depth)rease depth) andin the the and tire tire the contactcircumference tire patchcontact length (i.e.patch, a (Figurelengthshrinkage6 ).(Figure Lowering of the 6). tire Lowering the radius) tread depthandthe consequentlytread results depth in a decreaseresults a decrease in in a theindec the tirerease contact circumference in the patch tire length. circumference (i.e., a shrinkage (i.e. of, a theshrinkage tire radius) of the and tire consequently radius) and aconsequently decrease in the a decrease contact patchin the length. contact patch length.

Figure 6. Impact of remaining tread depth on the tire contact patch length . Tire Size: P245/65R17. FigureFigure 6. 6.Impact Impact of of remaining remaining tread tread depth depth on on the the tire tire contact contact patch patch length. length Tire. Tire Size: Size: P245 P245/65R17/65R17. . This poses an issue for the load estimation model. The load model will need to have knowledge of theThisThis tire poses poses wear an anstate, issue issue which for for the theby load itselfload estimation estimationpresents several model. model. technical The The load load challenges model model will will considering need need to to have have the knowledge knowledgecomplexity ofofof the thetypical tire tire wear wearmodels state, state, required which which by forby itself itselfestimating presents presents the several several remaining technical technical tread challenges challengesdepth. considering considering the the complexity complexity ofof typical typicalTo overcome models models required required this challenge, for for estimating estimating alternative the the remaining signal remaining features tread tread showing depth. depth. a strong correlation with the tire loadingToTo overcome overcome state were this this studied. challenge, challenge, Knowing alternativealternative that a signal tire deflects features features vertically showing showing in a a stronga strong linear correlation correlationmanner as with the with theload the tire is tireloadingincreased, loading state state it were was were studied. decided studied. Knowing to extract Knowing that the thata peaktire a deflects tire radial deflects vertically displacement vertically in a linear o inf thea linear manner tire manner from as the the asload radial the is loadincreased,acceleration is increased, it signal. was it was decided This decided requires to extract to double extract the integrating the peak peak radial radial the acceleration displacement displacement signal o off the theto retrieve tire from from displacement. the the radial radial accelerationaccelerationUnfortunately, signal. signal. accelerometers This This requires requires have double double an integrating unwantedintegrating thephenomenon the acceleration acceleration called signal signal drift, to to retrieve retrievecaused displacement. bydisplacement. a small DC Unfortunately,Unfortunately,bias in the acceleration accelerometers accelerometers signal. have haveThe an presencean unwanted unwanted of the phenomenon phenomenon drift can lead called called to large drift, drift, integration caused caused by by aerrors. a small small If DC DCthe biasbiasacceleration in in the the acceleration acceleration signal from signal. signal.a real accelerometer The The presence presence ofis of integrated the the drift drift can withoutcan lead lead toany to large largefiltering integration integration performed, errors. errors. the Ifoutput If the the accelerationaccelerationcould become signal signal unbounded from from a a real real over accelerometer accelerometer time. To solve is is integrated integrated the problem without without of anydrift, any filtering filteringa high performed,-pass performed, filter was the the output usedoutput to couldcouldremove become become the unbounded DC unbounded component over over time. of time. the To solveaccelerationTo solve the problem the signal. problem of drift, This of a isdrift, high-pass done a high by filter - carefullypass was fil usedter extracting was to remove used theto theremoveacceleration DC component the DCsignal ofcomponent the per acceleration wheel of turn the signal. and acceleration then This applying is done signal. by the carefullyThis high is pass done extracting filter by carefullyin the conjunction acceleration extracting with signal the the accelerationintegration operationsignal per ( Figure wheel 7 turn). and then applying the high pass filter in conjunction with the integration operation (Figure 7). Vibration 2019, 2 180

per wheel turn and then applying the high pass filter in conjunction with the integration operation (Figure7).

Figure 7. Algorithm for retrieving radial displacement from the acceleration signal. FigureFigure 7. 7. Algorithm Algorithm for for retrieving retrieving radial radial displacement displacement from from the thethe acceleration accelerationacceleration signal. signal.signal. By filtering before integrating, the drift error was eliminated, and the radial displacement ByByBy filtering filteringfiltering before beforebefore integrating, integrating,integrating, the thethe drift driftdrift erro errorerrorr was waswas eliminated, eliminated, and andand the thethe radial radialradial displacement displacement successfully extracted (Figure 8). successfullysuccessfullysuccessfully extracted extractedextracted (Figure (Figure(Figure 88). 8).).

((aa()a) ) ((bb()b) ) 25

Acceleration Acceleration 25 25 AccelerationAcceleration Acceleration Displacement AccelerationDisplacement Displacement Displacement DisplacementDisplacement

200 20 200 2020 200200 200 200100

0 100100 0

] 2 0 0 ]

2 0

] 0 ] -100 2 ] 2 ] 2 -200 2 -100-100-200 15 -200-200 -200 1515 -200-300 -300-300

100 10

100100 Acceleration [m/s Acceleration

Displacement [mm] Displacement 1010

Acceleration [m/s Acceleration Displacement [mm] Displacement Acceleration [m/s Acceleration Displacement [mm] Acceleration [m/s Acceleration Displacement [mm] Acceleration [m/s Displacement [mm] Acceleration [m/s Displacement [mm]

50 5050 5 5 5

0 0 50 100 150 200 250 300 3500 0 0 0 0 5050 100100 150150 200200 250250 300300 350350 0 50 100 150 200 250 300 3500 0 Rotation Angle [] 0 50 100 150 200 250 300 350 RotationRotation Angle Angle [° []°] 0 50 100 Rotation150 Angle200 [] 250 300 350 RotationRotation Angle Angle [° []°]

FigureFigure 8. 8.( a()a) Integration Integration without without filtering; filtering (;b (b) integration) integration with with filtering. filtering. FigureFigure 8. 8. ( a(a) )Integration Integration without without filtering; filtering; ( b(b) )integration integration with with filtering. filtering. Finally,Finally, the the amplitude amplitude of of the the peak peak radial radial displacement displacement of of the the tire tire was was extracted extracted from from the the Finally,Finally, thethe amplitudeamplitude ofof thethe peakpeak radialradial dispdisplacementlacement ofof thethe tiretire waswas extractedextracted fromfrom thethe displacementdisplacement curve. curve. The The radial radial displacement displacement shows shows the the near near linear linear correlation correlation with with the the tire tire load load and and displacementdisplacement curve. curve. The The radial radial disp displacementlacement shows shows the the near near linear linear correlation correlation with with the the tire tire load load and and inflationinflation pressure pressure (Figure (Figure9). 9). inflationinflation pressure pressure (Figure (Figure 9). 9).

LoadLoad Sensitivity Sensitivity PressurePressure Sensitivity Sensitivity 2424 14.614.6

14.4 2222 14.4

14.214.2 2020 1414 1818 13.813.8 1616 13.613.6 1414 Radial Displacement[mm] Radial Displacement [mm] Displacement Radial Radial Displacement[mm] Radial Displacement [mm] Displacement Radial 13.413.4

1212 13.213.2 400400 500 500 600 600 700 700 200200 210 210 220 220 230 230 240 240

TireTire Load Load [ k[kg]g] TireTire Inflation Inflation Pressure Pressure [ k[kppaa]] Figure 9. Impact of load and inflation on the tire vertical displacement. Tire Size: P245/65R17. FigureFigureFigure 9. 9.9. Impact Impact of of load loadload and andand inflation inflationinflation on onon the thethe tire tiretire ve verticalverticalrtical displacement. displacement.displacement. Tire TireTire Size: Size:Size: P245/65R17. P245P245/65R17./65R17.

It is important to mention that it would be incorrect to generalize that the tire behavior will ItIt isis importantimportant toto mentionmention thatthat itit wouldwould bebe incorrectincorrect toto generalizegeneralize thatthat thethe tiretire behaviorbehavior willwill always be linear under all loading and inflation pressure conditions. The linear behavior observed is alwaysalways be be linear linear under under all all loading loading and and inflation inflation pressure pressure conditions. conditions. The The linear linear behavior behavior observed observed is is specifically under the range of loads and inflation pressures considered in this study. specificallyspecifically under under the the range range of of loads loads and and in inflationflation pressures pressures consi considereddered in in this this study. study. Vibration 2019, 2 181

It is important to mention that it would be incorrect to generalize that the tire behavior will alwaysMore be linear interestingly, under all in loading comparison and inflation to the pressuretire contact conditions. patch length, The linear the amplitude behavior observedof the peak is specificallyradial displacement under the showed range of a loadsmuch and higher inflation sensitivity pressures to the considered tire load (Table in this 3 study.). Moreover, the radial displacementMore interestingly, shows negligible in comparison sensitivity to theto the tire tire contact tread patchdepth. length, This makes the amplitude radial displacement of the peak a radialmore displacementattractive feature showed in comparison a much higher to the sensitivity contact patch to the length tire loadand hence (Table was3). Moreover, the preferred the radialfeature displacementfor this study. shows negligible sensitivity to the tire tread depth. This makes radial displacement a more attractive feature in comparison to the contact patch length and hence was the preferred feature for thisTable study. 3. Sensitivity analysis—impact of load, pressure and tire wear change on the tire radial displacement and contact patch length. Table 3. Sensitivity analysis—impact of load, pressure and tire wear change on the tire radial displacement and contact patch length. Load (kg) Pressure (kPa) Tread Depth (mm)

Range Tested 350Load kg– (kg)680 kg Pressure200–240 (kPa) kPa Tread2mm Depth–8mm (mm) Range Tested 350 kg–680 kg 200–240 kPa 2mm–8mm % Change in the Peak Radial Displacement 80–90% 10–15% negligible % Change in the Peak Radial Displacement 80–90% 10–15% negligible % Change in the Contact % Change in the Contact Patch Length 40–50%40–50% 10–15%10–15% 15–20%15–20% Patch Length

A model capturing the dependencies between the tire radial displacement, load and inflation A model capturing the dependencies between the tire radial displacement, load and inflation pressure was fit (Figure 10). A polynomial fit with the second-order in pressure and the first-order in pressure was fit (Figure 10). A polynomial fit with the second-order in pressure and the first-order in the load provided a good model fit. the load provided a good model fit. 푃푒푎푘 푟푎푑𝑖푎푙Peak 푑 radial𝑖푠푝푙푎푐푒푚푒푛푡 displacement= 푝=00p +00 푝+10p10∗ 푙표푎푑load ++ p푝0101 ∗pressure푝푟푒푠푠푢푟푒+ + ∗ ∗ (1) p11 load pressure + p02 pressure2 (1) 푝11 ∗ ∗푙표푎푑 ∗∗ 푝푟푒푠푠푢푟푒 + 푝02∗ ∗ 푝푟푒푠푠푢푟푒2

Figure 10. Model fitting. Tire Size: P245/65R17, speeds tested: 25, 50, 75, 100 kph, camber = 0 deg. Figure 10. Model fitting. Tire Size: P245/65R17, speeds tested: 25, 50, 75, 100 kph, camber = 0 deg. Equation (1) can be rewritten into a standard parameter identification form as follows: Equation (1) can be rewritten into a standard parameter identification form as follows: 푦(푡)y=(t휑) ^=푇ϕ (ˆT푡)(t⋅)휃(θ푡)(t) · where : where: (Peak Displacement p00 p01 pressure p02 pressure2) ( ) = − − ∗ − ∗ 2 (2) (y푃푒푎푘t 퐷푖푠푝푙푎푐푒푚푒푛푡− 푝00(p11 − pressure푝01∗푝푟푒푠푠푢푟푒+p10) − 푝02∗pressure ) 푦(푡) = ∗ (푝11ϕ∗푝푟푒푠푠푢푟푒T(t) = 1 + 푝10) (2) 휑푇(푡) = 1 θ(t) = is the unknown variable (i.e. the tire load) 휃(푡) = 𝑖푠 푡ℎ푒 unknown variable (i.e. the tire load) TheThe unknownunknown parameterparameterθ ((t푡))can can bebe identifiedidentified inin real-timereal-timeusing usingthe the parameter parameter identification identification approach.approach. TheThe recursiverecursive leastleast squaressquares (RLS)(RLS) algorithmalgorithm [[37]37]provides provides aa method method toto iteratively iteratively update update thethe unknown unknown parameterparameter atat eacheach samplingsampling timetime toto minimizeminimize thethe sumsum ofof thethe squares squares of of the the modeling modeling error using the past data contained within the regression vector, φ(푡). The performance of the RLS algorithm was evaluated through experimental road vehicle tests. The vehicle maneuver was straight driving with intermittent gas pedal presses (Figure 11). Vibration 2019, 2 182 error using the past data contained within the regression vector, ϕ(t). The performance of the RLS algorithm was evaluated through experimental road vehicle tests. The vehicle maneuver was straight driving with intermittent gas pedal presses (Figure 11).

(a)

1000

500 ] 2 0

-500

-1000

Acceleration [m/s -1500

-2000 59 59.1 59.2 59.3 59.4 59.5 59.6 59.7 59.8 59.9 60 Time [sec] (b) (c)

1000 25 Displacement (mm) = 21.53 Acceleration Displacement 500 20 ] 2 0 15

-500 10

-1000 5 Acceleration [m/s Acceleration Displacement [mm] Displacement

-1500 0 Predicted tire load 1490 lbs. Actual Tire Load (measured)= 1455 lbs. -2000 -5 0 50 100 150 200 250 300 350 Error=2.6% Rotation Angle [°]

Figure 11. Prediction performance. (a) Raw measurement—straight-line driving; (b) feature extraction; (cFigure) RLS prediction 11. Prediction model. performance. (a) Raw measurement—straight-line driving; (b) feature extraction; (c) RLS prediction model. The RLS model was found to predict the tire load within the first 20 wheel rotations. The estimationThe RLS error model was 2.6%. was Forfound the saketo predict of completeness, the tire load the within same datasetthe first was 20 alsowheel used rotations. to evaluate The theestimation performance error ofwas a contact2.6%. For patch the lengthsake of basedcomplete loadness, estimation the same model. dataset The was estimation also used errorto evaluate was foundthe performance to be 5.3%. So,of a radialcontact deflection patch length based based model load was estimation found to be model. superior The than estimation the contact error patch was lengthfound based to be model.5.3%. So, a radial deflection based model was found to be superior than the contact patch length based model. 4.2. Tire Slip Angle Estimation 4.2. UsingTire Slip the Angle procedure Estimation described in Figure7, the lateral displacement profile of the tire footprint was extractedUsing the from procedure the lateral described acceleration in Figure signal 7, at the di fflaerentteral tiredisplacement slip angles profile (Figure of 12 the). tire footprint wasThe extracted following from observations the lateral acceleration were made: signal at different tire slip angles (Figure 12). The following observations were made: The maximum lateral displacement of the tire footprint increases as the slip angle increases. • • The maximum lateral displacement of the tire footprint increases as the slip angle increases. The slope of the initial linear part of the lateral displacement profile also increases as the slip • • The slope of the initial linear part of the lateral displacement profile also increases as the slip angle increases. angle increases. Both of these features were found to be linearly correlated with the tire slip angle (Figure 13).

Vibration 2019, 2 183 Slip angle=0° Slip angle =2°

Acceleration Acceleration Slip angle=0° Displacement Slip angle =2° Displacement

100 Acceleration Acceleration Displacement Displacement ] ] 2 2 50 40 20 100 0 0 2.5 -20 ] ] 2

2 2 -40 50-50 40 Acceleration [m/s Acceleration Displacement [mm]

Acceleration [m/s Acceleration -60 Displacement [mm] 20 1.5 0 -1000 2.5 -20 1 2 -40 0.8 -50-150 Acceleration [m/s Acceleration Displacement [mm]

Acceleration [m/s Acceleration -60 0.6 Displacement [mm] 0.5 0.4 1.5 0.2 -100 0 0 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 1 ° ° 0.8 Rotation Angle [ ] Rotation Angle [ ] -150 0.6 0.5 0.4 0.2 0 Slip angle =4° 0 0 50 100 Slip150 angle200 =6°250 300 350 0 50 100 150 200 250 300 350 ° Rotation Angle [°] Rotation Angle [ ] 200 5 Acceleration Acceleration Slip angle =4° Displacement Slip angle =6° Displacement

4 4 200 100 5 Acceleration Acceleration Displacement Displacement ] ] 2 2 50 3 4 4 100 0 0 ] ] 2 2 2 50 2 -50 3 Acceleration [m/s Displacement [mm] Acceleration [m/s Displacement [mm]

0 0 -100 1 2 2 -50 -150 Acceleration [m/s Displacement [mm] Acceleration [m/s Displacement [mm]

-100 -200 0 0 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 3501 Rotation Angle [°] Rotation Angle [°] -150

-200 0 0 Figure0 12.50 Extraction100 150 of 200lateral250 displacement300 350 from the lateral0 acceleration50 100 150signal200 at different250 300 tire350 slip Rotation Angle [°] Rotation Angle [°] angles. Figure 12. Extraction of lateral displacement from the lateral acceleration signal at different tire Figure 12. Extraction of lateral displacement from the lateral acceleration signal at different tire slip slipBoth angles. of these features were found to be linearly correlated with the tire slip angle (Figure 13). angles. Peak Deflection Sensitivity Initial Slope Sensitivity Both of these5 features were found to be linearly correlated70 with the tire slip angle (Figure 13). 4.5 Peak Deflection Sensitivity 60 Initial Slope Sensitivity 54 70 50 4.53.5 60 43 40 50 3.52.5 30

32 Slope Initial 40 20 2.51.5 30 10 Lateral Displacement [mm] Displacement Lateral

21 Slope Initial 20 1.50.5 0 0 2 4 6 0 2 4 6 10 Lateral Displacement [mm] Displacement Lateral 1 Tire slip angle [deg] Tire slip angle [deg] 0.5 0 0FigureFigure 13. 13.2Correlation Correlation4 between between extracted 6extracted features feat0 ures and and2 the the tire tire slip slip4 angle. angle. 6 Tire slip angle [deg] Tire slip angle [deg] AA multiple multiple linear linear regression regression model model (Equation (Equation (3)) (3)) was was trained trained using using both both these these signal signal features features and found to estimateFigure the 13. tire Correlation slip angle between with an extracted accuracy feat ofures+/ and0.2 the deg. tire slip angle. and found to estimate the tire slip angle with an accuracy of +/−− 0.2 deg. A multiple𝑇𝑖𝑟𝑒 linear𝑠𝑙𝑖𝑝Tire slip regression angle 𝑎𝑛𝑔𝑙𝑒= p model00 + p01 (Equation= lateral𝑝00 displacement+ (3)) 𝑝01 was ∗ trained𝑙𝑎𝑡𝑒𝑟𝑎𝑙+ p02 usinginitial both slope 𝑑𝑖𝑠𝑝𝑙𝑎𝑐𝑒𝑚𝑒𝑛𝑡𝑙𝑜𝑝𝑒 +𝑝02∗𝑖𝑛𝑖𝑡𝑖𝑎𝑙𝑠 these+ signal + features ∗ ∗ (3)(3) and found to estimate the tire slip angle 𝑝03 withp03 ∗an 𝑝𝑟𝑒𝑠𝑠𝑢𝑟𝑒 accuracypressure of +/− 0.2 deg. ∗ 𝑇𝑖𝑟𝑒 𝑠𝑙𝑖𝑝 𝑎𝑛𝑔𝑙𝑒 = 𝑝00 + 𝑝01 ∗ 𝑙𝑎𝑡𝑒𝑟𝑎𝑙 𝑑𝑖𝑠𝑝𝑙𝑎𝑐𝑒𝑚𝑒𝑛𝑡𝑙𝑜𝑝𝑒 +𝑝02∗𝑖𝑛𝑖𝑡𝑖𝑎𝑙𝑠 + (3) 𝑝03 ∗ 𝑝𝑟𝑒𝑠𝑠𝑢𝑟𝑒 Vibration 2019, 2 184

It is noteworthy to mention that the correlation between the lateral displacement of the footprint and the tire slip angle only holds true when the tire is deforming. Once the tire force saturates and the footprint starts to slide, this relationship does not hold true anymore. Future work will focus on understanding the lateral acceleration signal during high slip angle maneuvers and extracting signal features that correlate with the slip angle even when the tire is sliding.

5. Conclusions This paper provides a review of relevant works about the different signal processing techniques proposed by researchers to extract the tire load and slip angle information from tire mounted accelerometers. Details about research activities undertaken as part of this study to develop a smart tire are presented. Novel algorithms for the tire load and slip angle estimation are presented. The tire load is estimated by extracting the peak radial deflection of the tire from the radial acceleration signal. The tire slip angle is estimated by extracting the peak lateral deflection and the slope of the lateral deflection curve from the lateral acceleration signal. Multiple linear regression models are used to model the relationship between the identified signal features of interest and tire load, slip angle. It is noteworthy to mention that the motivation for using linear regression models is the low computational burden imposed while doing the real-time implementation of these models in the vehicle electronic control unit (ECU). Non-linear models are expected to give better fits; however, they also need more compute resources, which poses implementation challenges. This is expected to improve as the next generation vehicles are expected to come equipped with more sophisticated processing units to enable advanced driver-assistance system (ADAS) functionalities. Future work will focus on using the tire load and slip angle information to enhance the robustness of the vehicle state estimators. Key challenges for the large-scale industrialization of smart tire technology are as follows:

For the installation of sensors in a tire, many problems will need to be considered, such as the • compatibility of the sensors with the tire rubber i.e., stiffness issues, robustness of sensor in the harsh tire environment, wireless transmission of gathered data in an energy efficient manner, economic issues relating to the use of expensive sensors in a comparatively inexpensive product, the tire and finally meeting the power requirements of all the electronic components. Most studies have assessed the benefits of a smart tire for control systems in a simulation • environment. It remains to be foreseen how much of this benefit can be captured in the real-world, considering that most smart tires use a single point sensing solution transmitting data at a low frequency. It is also concluded that the current tire pressure monitoring chips available in the production • vehicles cannot support complex signal processing algorithms required to extract signal features from tire mounted sensors. Hence, an application specific integrated chip (ASIC) would be required.

Author Contributions: Conceptualization, K.B.S. and S.T.; Methodology, K.B.S. and S.T.; Validation, K.B.S; Writing—Original Draft Preparation, K.B.S.; Writing—Review and Editing, K.B.S. Funding: This research received no external funding. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Sabbioni, E.; Kakalis, L.; Cheli, F. On the impact of the maximum available tire-road friction coefficient awareness in a brake-based torque vectoring system. SAE Tech. Pap. 2010.[CrossRef] 2. Braghin, F.; Cheli, F.; Melzi, S.; Sabbioni, E.; Mancosu, F.; Brusarosco, M. Development of a cyber tire to enhance performances of active control systems. In Proceedings of the 7th EUROMECH Solid Mechanics Conference (ESMC 2009), Lisbon, Portugal, 7–11 September 2009; pp. 1–9. 3. Cheli, F.; Leo, E.; Melzi, S.; Sabbioni, E. On the impact of ‘smart tyres’ on existing ABS/EBD control systems. Veh. Syst. Dyn. 2010, 48, 255–270. [CrossRef] Vibration 2019, 2 185

4. Erdogan, G.; Hong, S.; Borrelli, F.; Hedrick, K. Tire Sensors for the Measurement of Slip Angle and Friction Coefficient and Their Use in Stability Control Systems. SAE Int. J. Passeng. Cars Mech. Syst. 2011, 4, 44–58. [CrossRef] 5. Cheli, F.; Sabbioni, E.; Sbrosi, M.; Brusarosco, M.; Melzi, S.; D’alessandro, V. Enhancement of ABS Performance through On-Board Estimation of the Tires’ Response by Means of Smart Tires; SAE Technical Paper; SAE International: Warrendale, PA, USA, 2011. 6. Cheli, F. Cyber Tyre: A Novel Sensor to Improve Vehicle’s Safety; SAE Technical Paper; SAE International: Warrendale, PA, USA, 2011. 7. National Highway Traffic Safety Administration. Proposed New Pneumatic Tires for Light Vehicles. FMVSS No. 139. Available online: https://www.nhtsa.gov/sites/nhtsa.dot.gov/files/tp-139-02.pdf (accessed on 25 April 2019). 8. 106th Congress. Transportation Recall Enhancement Accountability, and Documentation (TREAD) Act. Available online: https://www.congress.gov/106/plaws/publ414/PLAW-106publ414.pdf (accessed on 25 April 2019). 9. National Highway Traffic Safety Administration. Federal Motor Vehicle Safety Standards; Tire Pressure Monitoring Systems; Controls and Displays; National Highway Traffic Safety Administration: Washington, DC, USA, 2000. 10. 70 FR 18135—Federal Motor Vehicle Safety Standards; Tire Pressure Monitoring Systems; Controls and Displays—Content Details—05-6741. Available online: https://www.govinfo.gov/app/details/FR-2005-04-08/ 05-6741 (accessed on 19 April 2019). 11. Singh, K.B.; Arat, M.A.; Taheri, S. Enhancement of collision mitigation braking system performance through real-time estimation of tire-road friction coefficient by means of smart tires. SAE Int. J. Passeng. Cars Electron. Electr. Syst. 2012, 5, 607–624. [CrossRef] 12. Arat, M.A.; Singh, K.B.; Taheri, S. An intelligent tyre based adaptive vehicle stability controller. Int. J. Veh. Des. 2014, 65, 118–143. [CrossRef] 13. Singh, K.B.; Arat, M.A.; Taheri, S. An intelligent tire based tire-road friction estimation technique and adaptive wheel slip controller for antilock brake system. J. Dyn. Syst. Meas. Control 2013, 135, 031002. [CrossRef] 14. Arat, M.A. Development and Improvement of Active Vehicle Safety Systems by means of Smart Tire Technology; Virginia Polytechnic Institute and State University: Blacksburg, VA, USA, 2013. 15. Apollo Main Page. Available online: http://virtual.vtt.fi/virtual/proj3/apollo/ (accessed on 19 April 2019). 16. EU Project Friction. Available online: http://friction.vtt.fi/ (accessed on 19 April 2019). 17. Sensing Technologies. Bridgestone. Available online: https://www.bridgestone.com/technology_innovation/ cais/ (accessed on 21 April 2019). 18. Pirelli, Schrader Team up for Computerized ‘Cyber Tire’. Available online: https://www.motorauthority.com/ news/1044745_pirelli-schrader-team-up-for-computerized-cyber-tire (accessed on 19 April 2019). 19. Continental’s Intelligent Tire. Available online: https://www.continental-tires.com/transport/products/ continental-itire (accessed on 19 April 2019). 20. Jo, H.Y.; Yeom, M.; Lee, J.; Park, K.; Oh, J. Development of Intelligent Tire System; SAE Technical Paper; SAE International: Warrendale, PA, USA, 2013. 21. Goodyear Dunlop’s Chip-In-Tyre Shows Latest in Intelligent Tyre Technology: Tyrepress. Available online: https://www.tyrepress.com/2014/03/goodyear-dunlops-chip-in-tyre-shows-latest-in-intelligent-tyre- technology/ (accessed on 19 April 2019). 22. Mancosu, F.; Brusarosco, M.; Arosio, D. Method and System for Determining a Tyre Load during the Running of a Vehicle. U.S. Patent 7404317B2, 29 July 2008. 23. Schillinger, J.; Von Lutz, R.; Cyllik, A.; Lehmann, J. Sensor Transponder and Procedure for Measuring Tire Contact Lengths and Wheel Load. U.S. Patent 7536903B2, 26 May 2009. 24. Wagner, M.; Troeger, H. Method for Monitoring the Load of Vehicle Tires. U.S. Patent 8742911B2, 3 June 2014. 25. Tebano, R.; Audisio, G.; Fioravanti, A.P. Method and System for Estimating the Load Acting on a Tire. U.S. Patent 20130261991A1, 3 October 2013. 26. Brusarosco, M.; Mancosu, F.; Arosio, D. Method and System for Determining a Tyre Load during the Running of a Motor Vehicle. U.S. Patent 7945361B2, 17 May 2011. Vibration 2019, 2 186

27. Teerhuis, A.; Schmeitz, A.; van de Molengraft-Luijten, L. Tire state estimation based on measured accelerations at the tire inner liner using an Extended Kalman Filter design. In Proceedings of the 4th International Tyre Colloquium, Guildford, UK, 20–21 April 2015. 28. Krier, D.; Zanardo, G.S.; del Re, L. A PCA-based Modeling Approach for Estimation of Road-tire Forces by In-tire Accelerometers. IFAC Proc. Vol. 2014, 47, 12029–12034. [CrossRef] 29. Dharamshi, P.; Haas, T.; Kretschmann, M.; Spotka, J. Method and System for Determining a Wheel Load Acting on a Tire of a Vehicle. U.S. Patent 20170355234A1, 14 December 2017. 30. Schmeitz, A.J.C.; Teerhuis, A.P. Robustness and Applicability of a Model-Based Tire State Estimator for an Intelligent Tire. Tire Sci. Tech. 2018, 46, 105–126. [CrossRef] 31. Masago, T. Tire Load Estimation Method and Tire Load Estimation Device. U.S. Patent 20190025113A1, 24 January 2019. 32. Mancosu, F.; Brusarosco, M.; Arosio, D. Method and System for Determining a Cornering Angle of a Tyre during the Running of a Vehicle. U.S. Patent 7552628B2, 30 June 2009. 33. Mancosu, F.; Brusarosco, M.; Arosio, D. Method and System for Determining a Cornering Angle of a Tyre during the Running of a Vehicle. U.S. Patent 8024087B2, 20 September 2011. 34. Matsuzaki, R.; Kamai, K.; Seki, R. Intelligent tires for identifying coefficient of friction of tire/road contact surfaces. Sens. Smart Struct. Technol. Civ. Mech. Aerosp. Syst. 2015, 9435, 94350A. 35. PCB Model 356A01. Available online: http://www.pcb.com/Products/model/356A01 (accessed on 21 April 2019). 36. Singh, K.B. Development of an Intelligent Tire Based Tire-Vehicle State Estimator for Application to Global Chassis Control; Virginia Tech: Blacksburg, VA, USA, 2012. 37. Vahidi, A.; Stefanopoulou, A.; Peng, H. Recursive least squares with forgetting for online estimation of vehicle mass and road grade: theory and experiments. Veh. Syst. Dyn. 2005, 43, 31–55. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).