TRANSPORT AND ROAD RESEARCH LABORATORY Department of Transport RESEARCH REPORT 301 @ . THE MERLIN LOW-COST ROAD ROUGHNESS MEASURING MACHINE by M A CUNDILL, Crown Copyright 1991, The work describad in this report forms part of the programme carried out for the Overseas Development Administration, but the views exiressed are nol net Administration. Extracts irom the text may tv: reproducud, except lor comme source is acknowledged, | Overseas Unit Transport and Road Research Laboratory Crowthorne, Berkshire, RG11 6AU 1991 . ISSN 0266-5247 ily those of the ial purposes, provided the eelDk i i ie Li Sk CONTENTS Abstract 1. Introduction 2, Roughness measuring instruments 3. The MERLIN 3.1. Principle of operation 3.2 General description 3.3. Method of use 34 Practical details 4. Calibration equations 5. Accuracy of masurement 6. Discussion 7. Acknowledgements 8, References ‘Appendix A: Simulation of performance A4. The International Road Roughness Experiment A2 Simulation results AB Alternative procedures and designs AB.1 Choice of machine length A3.2 Measurement of data spread Page 2 1 12 7 7 €THE MERLIN LOW-COST ROAD RCO MEASURING MACHINE ABSTRACT The roughness of a road's surtace is an important ‘measure of road condition and a key factor in determining vehicle operating costs on poor quailty surfaces. 1 ‘report descrives a simple roughness measuring machine Which has been designed especially (01 use in dovoloping ‘counties. Its called MERLIN - a Machine for Evalualing Roughness using Low-cost INstrumentation. The device can be used either lor direct measurement of lor calibra ing response type instruments such as the vehicle: ‘mounted bump integrator. It consists of a metal rame 1.6 ‘atres fong with a wheel atthe front, afoot a the rear ‘and a probe mid-way between them which rests on the road surlace. The probe is attached to a moving arm, a the other end of wifich is a pointer which moves over a Chart. The machine is wheeled along the road and at regular intervals the position ofthe pointer is raccrdied on the chart to builé up a histogram. The width oft histogram can be used lo give a good eslinate ot oughness in terms ol the international Roughness Index. Calibration ofthe device was carried out using computer simulations of is operation on road profiles measured in the 1962 International Road Roughness Experiment Merling are in use in a number af davelaping countries, ‘They can usually be made locally at a current cost of \ypically 2508 US. 1. INTRODUCTION ‘The longitudinal unevenness of a road's surtace (nor ‘mally termed its roughnass) is both a good measure of {he road's condition and an important determinant of vehicle operating costs and ride quality. Within develop: ing counties, there is particular interest in the effect on vehicle operating costs. A number of studias (Hide etal 1975, Hide 1982, CARI 1982, Chosher & Harrison 1987) ‘have shown how roughness can iniluence the cost of vehicle maintenance, the extent of tyre damage anc Vehicle running speeds (and hence vehicle productivity) Reliable measurement of oad roughness is therelore Seen as an important actvily in road network manage: ‘ment. Several dillerent road roughness scales have been established and a variely of roughness measuring ‘machines have been developed. However, it was lel hat there was a need, paniculary within developing coun- tries, for a new simple type of measuring instument which could be used either dirécily to measure roughnes over a limited part of the road network oF for calibrating other roughness measuring equipment, particularly the very widely used vehicle mounted bump integrator UGHNESS 2. ROUGHNESS MEASURING INSTRUMENTS Roughness measuring instruments can be groupe io three diferent classes. The simplest in cancel ae the stallc road pratile moasuring devices such as the rod and level, which measure surlace undulations 3 regutaw intervals. Unortunately, these devices are very slow i use and thoro can be a considerable amount of calcul tion involved in deriving roughness levels from ihe measurements taken, ‘Two recent devices which work on a similar principle but are semi-automated are the TARL Abay beam (Abaynay: ‘aka 1964) and the modified ‘Dipstck proliler (face Company). With both of these instruments, the suttabe Lndulations are measured trom a static Felerimces and dais fed directly into a microprocessor tad ty necessary calculations. They produce high quitity ‘results, bul Ihoy are relaively slow in operation and expensive The second class of instrument isthe dynamic profile ‘measuring device, such as the TRAL high-speed pill ‘ometer (Stil and Jordan 1980). In these instants, surface undulations are measured with respect Wo a ‘moving platform equipped with some means ol compen- sting for plattorm movement, so thatthe true road profile ‘can be derived. This is then converted fo roughness indices by automatic data procassing. These devices can operate al high speeds and give good qualiy rwsulls. bul they are very expensive, they are not usually suitable for very rough roids and thay have to be careluly mi tained, Finally, hove are the response-type road roughness ‘measuring systems (RTARMS), These measure the ‘cumulative vertical movernents of a wheel or axle with respect to the chassis of a vehicle as it vavels slang the toad. in the case ot a standard device such as the lowed hth whee! bumnp integrator (Bl} (Jordan and Yours 1980), the response is used directly as & roughness, index. In other non-standard devices, such as the vehicle-mounted Bi, tho rasponse is Converted fo a standard roughness measure by calibration. Ths towed ith wheel Blis expensive and needs carelul operation, The vehicle: mounted BI, however. is much cheaper and ‘can perform well as long as i is correctly used ant is calibrated cegulary The standard roughness scale which has been used lor many years by the Overseas Unit of TAAL i is studies on vehicle operating costs and pavement deteriaralion is the output of the Hlth wheel Bl lowed a1 32 kmm, How:-] fever, another seal which is now being widely uscd is the International Roughness Index (Sayers otal 1906) 11 scale, which is derived from road prolite data by a tatly ‘complex mathematical procedure, represents the vertical movement of a wheel with respect to a chassis in an idealised susponsion sysiem, when traveling along tive road al 80 km/h. As with the BI scal, itis measured in farms of nits of vertical movement of the wheel per unit length of road, and is normally quoted in metres per kilometre. Traditionally, the BI scale is normally quoted ia milimetres per kilometre, 3. THE MERLIN The new instrument which has been developed is a variation of the static profile measuring device. It's a ‘manually operated instrument which is wheeled along the oad and measures surface undulations at regular intervals. Readings are easily taken and there is a ‘graphical procedure for data analysis so thal road roughness can be measured on a siandard roughness Scale without the need for complex calculation. is ppaticular attractions for use in the developing world are that it is robust, inexpensive, simple to operate, and easy to make and maintain, ‘The device is called MERLIN, which is an acronym for a ‘Machine for Evaluating Roughness using Low-cost INstrumentation. i was designed on the basis of a Log 1 0.9m Figure 1 me {compuiter simulation of is operation on road swat Inieasured in the Intemational Road Roughness | ‘mont (Sayers ot al 1986a). Details of this simulat ‘avon in Appendix A. 3.1 PRINCIPLE OF OPERATION ‘The principle of operation is as follows. The dew two feet and a probe which ast on ihe oad surle the wheel-tack whose roughness is to be meas foot are 1.8 metres apart and the probe lies mid ‘between them (see Figure 1). The device measy vertical displacement between the road surface ‘probe and the cantre point of an imaginary ine jc {wo points whero the road surtace is In contact w two feet. This displacement is known as the ‘mie deviation’ ( measurements ara taken at siccessive interva road, then the rougher the road surface, the gr: variability of the displacements, By plotting the di ‘ments a8 a histogram on a chart mounted on they, ‘ment, itis possible to measure their spread and {*, been found to correlate well with road roughness, ‘measured on standard roughness scales. ? The concep! of using the spread of mid-chord det 5 a means of assessing road roughness is not n example, two roughness indices, Ql, and MO, ha ‘proposed by other researchers and are describes ‘Sayers et al (1986a), They are each based on thy Leg 2 Measurement nf mid-chard dnwintion‘ean square values ol two mid-chord deviations with fllerent base lenqths andl have been suggested a. standards which can be calculated relatively easily too ‘ard protiles measured by rod and level. Jowever, the Mein operates by using just one base tengih, the machine measures mid-chord deviations without the need for red and level, the variabity ofthe rid-chord deviations is determined graphically and very ite calculation is invelved to determine coughoess 2.2 GENERAL DESCRIPTION Figura 2 shows a sketch ol the Merlin. For ease of iperation, a wheel is used as the front leg. while the rear {gis a rigid metal rod. On one side of the rear logis & ‘horler stabilising lag which prevents the device from ling over when taking a reading. Projecting behind thr rain rear log are two handles, so that the dovice looks in ‘ome ways ike a very long and slender wheelbarrow. he probe is attached to a moving arin which is weighted thal the probe moves downwards, ither until aches the road surlace or the arm reaches the lil of 's traverse, Atthe olhar end of the arm is attached! toinier which moves over the prepared data chart. The Front foot (with marker in contact with the road) ‘arm has a mechanical amplification often, 80 thal a movornent ofthe probe of one mitimeter: will produce a. movement af the pointer of one centimetre. The char consists of a series al columns, cach S.nvm wide, and divided into boxes. Ihe radius ofthe wheel isnot unitorm, there wal be a variation in the fongty of the font leg lon one measure iment fo the nat arid this wil give rise to accuracy 1 The ‘Merln’s results, To overcome this, a mark is painted on the rin of the whee! and all measurements ave taken with the mark al ts clasest proximity fo the road, The wheel is, then said fo be in ils “normal position 3.3. METHOD OF USE The recommended procedure to determine the rough hess ola siyetch of rond isto ake 200 measurements al ‘regular intervals, say once every wheel revotution. At each measuring point, the machine is rested on the coad withthe wheal in its normal position and the rear foo}, probe, and siabilisar in contact with the road surlace. The Dperalor thon eocards the position of the point on te chart with a erons in the appropdate column and, t keep fp rncord of the total number of abservations, makes. {r035 in the ‘ally box" on the chat Moving arm Stabiiser M2 | | 5 éRi ul ‘ eo oe ol fee eos oe cle a ‘The handles of the Merlin are then raised so that only ths wheel remains in contact with the road and the machine is moved forward fo the next measuring point where the process is repeated. The spacing between the measuring points does not matter, as iong as the readings are always taken with the wheel in the normat position. ‘Taking measurements at cogular intervals should both produce a good average sample over ihe whole length of the section and reduce the risk of bias due to the opera. tor tending to avoid particularly bad sections of road Figure 3 shows a typical completed chart When the 200 observations have been made, the chart is emoved trom the Mattia, The positions mid-way between tho tanth and the eleventh crosses, countiag in rom each ‘end of the distribution, are marked on the chart below the ‘columns. It may be necessary to interpolate between column boundaries, as shown by the lower mark of the Testscomon__c7 Wenpam ss oan 2.6.90 oreraTOR See ela ehche bet Debs abuixbebebe stabetabet ola eee fate be Bele elsaelx six ec ex ee el ee eee e exaimlo, The spacing betwaen the two mark Imeasired in mimes an hi the rough Mes seale, Road foughness, in terms of the onal Roughness Index or a8 measured by 4 ‘toe! bump ategeator, can tn be datenin of ths equations given in Seton & 3.4. PRACTICAL DETAILS Plates 1 and 2 show the Merlin. For ease of r the nai bam, the central and rear legs, the ‘arm, the stabiliser and the handles are all ma Slee! tubing of square cross-section, 25 x 25 Wall mickiess of 1.5 mm, Joints are welded + possible, though the stabiliser and handles a: bolts so that thay can be eemoved for easier | tion, To strengthen the joints between the ma the logs, additional siruts are used. The whee « 0-88 mm Q fers xk a Figure 3. Typical completed chartSESE EES ERE EEE EEE TUES TEE SEEN TEES SEES SRE SE ESESEEIE SE SIE SE CIES SSIES EOE SESE SEES SEES SEE ES SES ES SES ESSEE ESE EESSEEESIS SESS EE SES EESSSEESSSSESSSSESESES TSE SEIES ESSEC ESE SESES ESE CTESESESMESESSESIESECSETECE SES OETE CISTI CISTI TIETTIS any type of common bicycle wheat mounted in a pair of j front forks and with a iyre which has a tally smooth wend pattem. ‘To reduce sensitivity to road surface micro-texture, the probe and the raar foot are both 12 mm wide and ‘founded in the plane ofthe wheel irack to a radius of 4100 mm, The rounding also tends to Keep the point of ; contact of tha prabe with the road in the same vertical line. The pivot is made trom a bicycle wheel hub and the ann betwaen the pivet and the weight is stepped to avoid ‘grounding on vary rough roads, weariness ing tough an elongated holo: a sysiom which bath vertical ang lataral adjustment. The varical sition af the probe must be sot so that the pointer is - clase 10 tho mice ofthe chart whan the proba displace: montis zero, or the histogram wil nol bo central. The 1 latera! position of he probe has tobe adjusted so that its | traverse pacsos cantally through the in jining the bottom ofthe tyra and the rae foot. if na, iwi be found 5 that when the machina ted from sido to side, tho pointer moves. When coteaty adjusted, fearing the tmachine over to ene side so thatthe stabiliser rasts on the road has its effect on tho position of the pointer Bofors usa, the meichanical amipliication’af the arm should be chacked using a small callration block, ‘ypically 6 mm thick. Insartion ofthe block under the probe should move the pointer by 60 min and any discrepancy has fo be allowad for. For example, if the pointer moved by only 57 mm, then the value of D fi ‘measured on the chart should be increased by a factor of Ite ao racommandad that ehack is catiod ott before ne after each set of measuremons fo ensura that there has been no unwanted movement of teal pats such as : the raar fot or the probe mounting. The check is carriad ‘vt by returing he machine fo a precisely datined posi along the road and making sure hat he same Doinisr ceading is obtained. 1, ywhen making measurements on a very rough road, ] ‘more than 10 readings ara at ether limit of the histogram, the probe should be-rernoved and attached to the alternative fixing point which is provided. This is twice as . far from the pivot and reduces the mechanical amplfica- tion of the arm 16, halving the width of the distribution. . Values of D raad from the chart are gcaled using the calibration procedure described earfler. Although the * spacing between the probe and the two feet is no longer (0.9 metres in this case, the errors introduced are small 4 and can be ignarad. 4. CALIBRATION EQUATIONS ‘The relationships between the Merlin scala and the and IRI scales are given below, For all road surfaces: IRI = 0.593 + 0471 0 42> D> 312 (24> IRl> 15,9) whore IR isthe roughness in terms of the Internat Roughness Index and is measured in matras per motre and D is tho roughness in terms of the Mei and is measured in milimetres, al 803 + 47.50 ' 42 > D> 12 (1,270 > BI> 16,750) where Blis the rqughiess as measured by a fith bump integrator bed at 32 kr and is measure rmilimatras per kilomatt, When measuring on the Bt scale, greater accuracy be achieved by using the following relationships 10 different surface Wypos. . Asphaltic concrete Ble 57442090 ‘ 42 .Ar iai90/5Roughness (mmvien) International Roughness Index e = = a : i : = ° Te ce au om 18.000 — 100 . ‘5000 es S a P : Mestin © frm) Figure 4. Calibration relationshipsbok met ee Be Gavel [2s-t104+ 44.00 tee 08} i if Bat 82 ka/h (roen/tam) : | po 200 30 400 Mertin © (mm) Figure §. Calivation relationships for Bl ~ Gifferent surface types4,000 3,000 2.000 81 (mm/km) 1,000 © Resuits from . tost sections Cattration relationship Ble 574+29.90 L L L el ° 20 40 6a 80 100 Merlin D (mm) Figure 6 . Calibration check on asphattic concreteJORDAN, PG and YOUNG, J G (1980). Developments inthe calibration and use of the Bump-Integrator for ride assessment, TAAL Supplementary Report 604; Trans port and Road Research Laboratory. Crowthorne. ‘SAYERS, W S et al (1986a). The Intemational Road Roughness Experiment; establishing correlation and a calibration standard lor measurements. World Bank Technical Paper Number 45, The World Bank, Washing: ton D.C. SAYERS, W S et al (19868). Guidalines for conducting ‘and calibrating road roughness measurements. World Bank Technical Paper Number 46. The World Bank, Washington 0.C STILL, P B and JORDAN, P G (1980). Evaluation of the TAAL high-speed profilomeler, TAR. Laboratory Report 922: Transport and Road Research Laboratory, Crowthorne. APPENDIX A: SIMULATION OF PERFORMANCE AA THE INTERNATIONAL ROAD ROUGHNESS EXPERIMENT {In 1982, a major study, the International Road Rough: ness Experiment (IRRE), was carriod out in Brasilia (Gayors et al 1986a) to compare the performance of a ‘umber of different road roughnoss measuring machines ‘and to calibrate their measures to a common scale. As ‘par of this study, the machines were run over a series of test sections 320 metras long, for four types o road surface - asphaltic concrete, surface treated, gravel and ‘earth, Ona of the Instruments used in the study was an ‘early version of the TARL Abay Beam. This employed an aluminium beam, 3 metres in length, supported at each fend by adjustable tripods which ware used for levelling. Fuunning along the beam was a sliding carriage which hhad at its lower end a wheel of 250 mm diameter which was in contact with the road surface. A linear transducer inside the carriage measured the distance between the bottom of the wheel and the beam to the nearest mili metre and this was recorded at 100 mm intervals along tha road. By successively relocating the beam along the Jongth of the road section and repeatedly leveling the , beam, the racordings provided a continuous sampling of the road profile, Data from the Abay beam were available for 27 of the test wheel paths, These are listed in Table At together with roughness on tha IRI scale as computed trom the ‘beam road profile data and roughness on the Bi scale as ‘measured by a fit wheel bump integrator towad at 32 anh, As can be seen, there aro eight paths on asphaltic concrete roads, five on surtace treated roads, seven on ‘gravel surfaces and seven on earth surfaces. Rough: asses range tram 2.44 mvkm on tho Il scale (1,270 ‘aynvlan 09 the BI scale} forthe best asphaltic conerete ‘surlace to 15.91 mvkmn (16,750 mravkin on the Bl scale) for the worst earth surlace. Figure A1 shows, as an example, the road profile as measured by the Abay Beam along 50 metres ol two of the test sections. The first is an asphaltic concrete road in relatively good condition, while the seconduis a gravel surtace in fair condition. As might be expected, compared to the asphaltic concrete, the gravel suriaca shows a much greater presence of shor wavelength undulations, To help visualise the Merlin’s response, the Figure alsa shows the machine's lengih, 1.8 metres, on Ihe same scale A.2. SIMULATION RESULTS Given these road profiles, it was possible to carry out a computer simulation of the performance ol a Merlin, Neglecting the small effects due tothe fact that the Merlin {s not operated in a horizontal positon, i itis assumed that the rear foo! is placed al a horizontal distance of X ‘metres from the start of the section, then the probe would be al a distance ol (X + 0.9) meves fom the start and the Iront foot at a distance of (X + 1.8) metres. lt the corresponding vertical distances at these points are Y,, Y, and Y,, then the pointer on the Murtin will be displaces from the Zero position by an amount d, givan by Mx(Y,-0.5x(%, + Ye) a where M is the mechanicat ampiilication provided by the moving arm, usually close to 10 Placing tha Mertin at successive positions along the road is simulated by using suecossively increasing values of X, ‘Tabulating the values of d into different § mm ranges corresponds to making crosses in the columns of the chart, and once 200 observations tiave baen mada, O can be deduced from the labulation, using the process of counting in ten observations trom each and of the distribution and interpolating where necessary For each of the test sections, four simulation runs were carried oul. in each tun, a Meri reacing was taken every 1.5 meties, so that the observations covered almost the ‘entte fast Section. inthe fist run, the starting point wass atthe beginning of the test section. Subsequent runs Slated al 0.4, 0.8 and 1.2 metres tom the beginning Table A2 shows the results ofthese simulations. Values ol D for each of the four runs per section are denoted as D,,D,, , and D,, The Meriin’s operation is essentially a statistical sampling of the road profile and the values of D show a statistical scattor with an average coetlicient of variation of eight per cent. To reduce the ellects of this, scaler, mean valves of the lour simulation runs are used in the analyses, ‘A plot of roughness on the IRI scale against D for each of the test sections is shown in Figure A2. As can be seen, the points are a good fi to a linear regression passing lose to, bul not through, the origin, Table A3 gives theTABLE At Test Sections Secin Surface Section Whoo! 1R al no. ‘yp0(1) code (2) veack (3) (enk) (enavkrm) 7 7 oa nS «76 3095 | 2 nO 0 os 5.80 365 ! 3 AG 05 NS 5.68 4050 4 AG 05 08 653 4390 R) 5 8 AG 08 NS. 6.96 4605 o 6 AC 06 os 8.29 5370 nO 10 NS 329 1850 4 *s iB os baa ‘210 st ot 08 451 280 10 st 0 08 527 3705 u st 05 os, 7.00 1920 12 st 06 NS an 2250 13 st 06 os 341 2728 4 cn 01 NS. 3.03 2010 18 Ga 05 NS 350 5075 16 GR 05 08 982 8095 7 cA o7 NS ait 2310 18 GR 07 08 74 5025 19 GR 2 NS 11.65 8545 20 GR 12 08 143i 12225 21 EA o NS 439 2035 Py EA o 08 a7 3865 23 EA 93 NS 6.03 4315 24 EA 03 08 8.03 8385 25 EA 08 NS 1591 18750 » 26 EA " NS. 778 6085 ar EA " 08 1078 10055 1. AG = Asphalt concrete ST = Surface treated GR = Gravel EA = Earth 2. _Asused in he IRRE 3. NS = Nearside = Right OS = Offside LetAephatic am Seaton ACIONS ie $0 --—————_—_—__-___ ~ — I nm fe} nat i : Yr 25 pe oe ° 70 * 30 4 Ee 757 a as Grave! ten Section GROTOS 2 30 Heoteontl distance (im) °_Fig.A1 Examples of text section profilesTABLE A2 Simulation Resuls Secin Surface in al D (mmm) m0 ypott) vin) fmm) 2, 2 2 Moan ' ac 4753095 i a | ye ay 2 AG 590 3465 913-975 tka. 3 AG 50a 4050 75 850.0 aR 4 aC. 6534990 67 res 125 tanta 5 nc 595 45,05 71 1801813 targa 5 AG B23 $970 150 19001625 16831785 7 ac. 3291850 SOE cre toa ge. | ano a a 24a 1270 8 529 aa? aN 9 st 451 3200 50 BABS 299 wag 10 st 5273705 wos 1075 akg " st 700 4920, TO Cn 2 sr att 2280 soé 636592838 3 ST Ba Bras ee i ee a “4 cn 393 2010 m2 783797558 15 GR 8505875 jaa 1002) | | tees = ees) te 16 cr 8928095, 2050 180.0242 tae 19.9 ” GR ai 2910 87 813025780 10 GR 704 5025, 17S 408 15001550458 19 Gn 165545 250825580 40.8 20 Gh Wat 13225 m0 27572818 21 EA 439 2995 e000 an SA Tw 2 EA 4723865 858 1900 87S 23 EA 503 4a15 2014212331050 att 24 EA 8030385 1570-1858 = 150017081609, 25 EA 159116750 75° 390020000 tg 26 EA 7786085 178 1750163871703 a En 1078 10055 2150 225 TS 08S i Ac) Asphalic concrete ST = Surlace vaated GR = Gravel EA Earth {09ression cooticients together with thor standatd errors, ‘The cootticient of determination (Ais over 0.98. Hence iW appears thatthe Marlin canbe used! as a fily accurate ‘means of measuring roughness onthe IRI scale. Figure AS shows a similar pot or roughness on the Bl ‘scale, Once again, te points can be fitted fo a linear ‘egression passing close tothe origin. However, he ft to the line isnot as good as forthe IR scale and the oelicient of determinations lower al jus! under 0.92. In par, this was fo be expected since the Bl value was {termined independently using a dynamic measuring ‘device whoteas the (Rl and Merin valvns were bat ‘computed fom the same static pete data, Howave, this ' not the ful explanation and better corcelation ean be achioved witha Matin of eilecent engi as described in Section A. Upon closer examinaion of Figure A, ean be seen that ‘hore are consistent ditoreaces etween the resus or the diferontsurtace types. Tho analysis can therelore be ‘improved by considering tho ditaret surface types ‘separately andthe result of doing so is shown in Figure Ad, Table AS ists the rogression coetiient, The Coefficient of determination ranges trom 0.944 on asphal- ‘ie conerete surfaces fo 0.987 on aurtace Weated soc- Yon. A3 ALTERNATIVE PROCEDURES: AND DESIGNS. ‘The simulations describad so far have used one sampling procedure, a Matin of one pariciar size and one ‘method of data analysis. n lac, the choice of these was. bbased on other considerations and the results of other simulations. ‘The Metin samplos the oad surlace ata number of poinls and the accuracy with which roughness can be ‘deduced clearly depends upon the qualiy and size othe ‘sample Ii was fel thal he best way af ensuring an unbiased result was to have a systematic sample wih recordings taken al regular ineals. The sample size[ im 609 -0,04r10 Metin D (rn) Fig.A2 Relationship between | R {and OTABLE A3 Rosults ofthe Regression Analyses. (Roughness = A, + A, 0) Roughness Surtace Number of scale ype (1) As) A.) Re sections iF ‘a 0.593 0.0471 “0.988 27 (mk) (0.185) (0.0012) al a 29 75 oar ar (mnvaeny (423) (28) a ac. 574 209 oar 8 (mmm) (401) (97) a st 132 ane 0.987 5 (mnvkm) (220) (2.5) 8 GR +1134 440 0.967 7 (mnvien (676) (6) at EA 2230 504 oars 7 (convien (797) (44) 1. AC = Asphatic concrete SY. = Surface treated Gi = Gravel EA = Eath 2, Bracketed values ave one standard error (200 observations) was chosen as a practical upper mit fom the point of view of managing the data handling and limiting the length of time tab to measure D. A3.1 Choice of machine length ‘The choice of machine length was examined by simula Ing Mertins of lengths ranging trom 0.6 to 3 metres. Using the same procedure as that described above, and not distinguishing between the differant types of surface, linear regressions wore derived relating the value of roughness on the two measuring scales to 0 for each Merlin length, Figura AS shows the R values for these regressions. On tho IRI scale, the best correlations are between 1.4 and 2.6 metres. The highest value occurs at around 1.8 ‘metres and so this was chosen as the standard Mein length. Reducing the length below 1.4 metres causes a ‘sharp decrease in correlation, Turning to the results for the Bl scale, the answer is quite dillerent, Here the best correlation is more sharply defined and occurs at a Merlin length of one metre. The degree of correlation is not as good as the best IRI valve, but this is to some extent explained by the fact that the Bi value was detormined by independent measurement. The uso of a ono-metre Mertin is an atvactive concept, since it would be considerably more portable than the 1.8 ‘metre version. However, it would be a much poorer prodictor of IRI and in practice it would be necessary 10 istinguish between tha dillerent surface types to reduce ome of the uncertainty ‘The underlying reason for the results of this analysis can be explained by considering tho irequency sensilvilies of the Mertin and the IF and Bi scales. The Mevdin has a {undamental frequency response to surtace waves of wavelength equal to ts own bse longtn, while the [Rt ‘and Bl scales are particularly sensitive to surface waves which would stimulate the natural vibrations of a vehicle whee! (al about 10 Hz) and a chassis (at about 1 Hz}, A180 km/h, the speed used for the IRI scale, the natural vibration of the wheel would be'stimulated by surface waves of around 2.2 metres and the chassis by waves of ‘around 22 metres, Al 32 kmh, the speod used for the 6} scala, the equivalent surtace wavelengths are 0.9 motes and 9 metres respectively. Hence it appears thal the Correlation analysis has selected Merlin lengths such that the wavelength of the fundamental requency is close 1. the wavelongih of the surtace waves which would stimulate the natural vibration of the wheel 3.2 Measurement of data spread Finally, the choice of method for determining the data Spread should be described. Measuring the limits for a Ccortain central percentage ol the data points is an _ci 20000 16000 i 8 BL at 32 kan/h (mmv/km) 100 200 300 400 Meriin D (mm) Figure A3. Relationship between Bl and DS@08}#NS JUaJ9}1P 40} Q PUR Ie UBEMYeq YSsUORRIOY * py dnb) {ta} ao44 (uy) @ uneyy Cor coe we oh or 00@ oe, o01 Y locos 0008 0001 ooo Joost ooost oo Were cone looowe (uous) ta (wyum) yg eae 10005 wp vf cf coat cooo, € . -fooost | ~——fo0081 [[os°62 +715 -i] ned lovee L. (won) (unyuau a powwon eowng 23910000 opoudoyy‘moni. Hf working to the Bl scale and using a single ‘elalionship forall surface types, systematic errors are much larger. The IMS residual eror for single measure: ‘ments was 21 per cent and this reduced only slightly to 19 por cent for lour measurements The benatits of multiple measurements are greater when ‘using separate BI relationships for each surtace type: tne MS residual errors ranged from 9 10 15 per cent for singlo measurements compared with 4 to 19 per cent for ‘mulipla measuremonts. Tho relatively large error fot asphaltic concrete compared to surface treated roads Could wal! rfloct the more limited roughness range for the latter and that the true relationships are non-tinesar, When estimating roughnoss for a vehicle, the normal procedure is o assume thal the combined roughness lor the two whee! tracks can be equatod to the mean o the individual tracks, although this doos give tise to a small ‘error. Hence, in practice, a minimum of two sets of Merlin observations are required, The roughness o! the individ ual whaol tracks can diffar considerably. Beating in mind the above limitations, it is normally better to calibrato an RTRRMS device ata larger number of sigs than make many repeat measurements al the same site, Moreover, particularly if working on the 8l scale, these sitos should have similar surfaces to those on which the RTRAMS is to be used. A number of other practical points should be considered when measuring roughness or calibrating an ATRRMS and a uselul guide is provided by Sayers et al (19860). ‘As a simple cross-chack on performance, roughness values on the Merlin scala were measured for a series ol asphaltic concrete test sections on the TRAL experimen- tal track. Four measurements were taken on each section ‘and the mean values are shown plotted in Figure 6 agalnot the roughness of each section on the BI scale as ‘measured with the Abay beam (Abaynayaka 1984). The Figure also shows the Meri __ asphaltic conerate roads as given in equation 3. As can 'be so0n, the points lie very close to the calibration line ‘and while the check is by no means comprehensive, it does lend strong suppor tothe results derived from the simulation, 6. DISCUSSION ‘The reason for designing the Merlin was to provide a dovice which is easy to use and reasonably accurate and yet can be manufactured and maintained with the limited resources availabla within developing countrios. Experi: ‘ence indicates that it has been successful in meeting these objectives. A number of the machines have been ‘mada at TRAL and shipped overseas, while other units hhave been made overseas from drawings provided by the Laboratory. To dato, Merlins have been used in 11 Gravel EA = EarthCoetfcient of Determination (R 2) ‘Moxie Length (r) Figure AS, Rouginess measuring accuracy for Merins of cterent long attractively simple procedure in the field and requires a ‘minimum of calculation. To decide what percentage ‘would give the bost answers, the performance of a Merlin over the test sections was again simulated. This time, the machine length was fixed at 1.8 metres and the rough: ness was measured on the IRI scale. Linear regressions were carried out betwoan D values, dorived using diferent data percentages, and roughness. Table A4 shows the resulting values of Fi, from which it can be seen that, of the values tasted, 90 per cent, which corresponds to counting in 10 crosses from each end of the distribution, appeared to be the best choice . TABLE A4 Elect of Data Limits on Correlation Porcentage Count fromadge RY ofdata > of distribution 95 5 “0.992 90 10 0.983 85 18 0.966 80 20 asa\® [3 FA QUALITY AND TEST SYSTEMS: TALLY BOX. vere HG of PG) ie LTT — Fl ap ae rer TF BP EF Fe SF SF BF Ss se SF SF eS S&S ee YE