Research Study Austin, Texas 78712

Transcription

1 TECHNICAL REPORT ST ANOARO TIT LE PAG E 1. Reprl N. 2. Gvernmenl AccelSln N. 3. Recipient'. Catalg N. CFHR F 4. Titl. and Subtitle 5. Reprt Date A STUDY OF THE REIATIONSHIPS BE'lWEEN VARIOUS CIASSES OF ROAD-SURFACE ROUGHNESS AND HUMAN RATINGS OF RIDING QUALITY 7.' Authrl,) Hugh J. Williamsn, W. Rnald Hudsn, and C. Dale Zinn 9. Perfrming argn; llin Nme nd Addre.. August Perfrming Organ; ltin Cde 8. Perfrming Organi lalin Reprt N. Research Reprt 156-5F 10. Wrk Unit N. Center fr Highway Research 11. Cntract r Granl N. The University f Texas at Austin Research Study Austin, Texas Type f Reprt nd Perid Cvered ~~~----~--~--~~~ ~ Final 12. Spn.ring Ag.ncy Name nd Addr Texas State Deparbnent f Highways and Public Transprtatin Transprtatin Planning Divisin P. O. Bx 5051 Austin, Texas Suppl.m.ntry Nle. 14. Spn.ring Agency Cde Study cnducted in cperatin with the U.S. Deparbnent f Transprtatin, Federal Highway Administratin. Research Study Title: "Surface Dynamics Rad Prfilmeter Applicatins It 16. Abatracl In this study, statistical relatinships between each f several classes f rad rughness and human panel ratings f riding quality are develped. Fr this purpse, the rughness is categrized thrugh digital filtering methds n the basis f wavelength. Lngitudinal and transverse surface effects are als studied. Multiple regressin analysis is used t relate the panel ratings t rughness as a whle and t the individual types f rughness. By using the mdels s develped, ne can btain fr any given rad sectin a measure f riding quality crrespnding t each f a set f imprtant aspects f rughness. Use f the mdels is demnstrated by analyzing the rughness f an illustrative rad sectin just befre, just after, and a year after an verlay. 17. K.y Wrds present serviceability rating, serviceability index, rad rughness, digital filtering, regressin analysis, Surface Dynamics Rad Prfi10meter 18. Distributin St.t.m.nt N restrictins. This dcument is available t the public thrugh the Natinal Technical Infrmatin Service, Springfield, Virginia S.curity Clallif. (f thl. r.prt) 20. Security CI.lllf. (f thl. pa,.l 21. N. f Pag Pric. Unclassified Unclassified 216 Frm DOT F (S U)

2

3 A STUDY OF THE RELATIONSHIPS BETWEEN VARIOUS CLASSES OF ROAD-SURFACE ROUGHNESS AND HUMAN RATINGS OF RIDING QUALITY by Hugh J. Williamsn W. Rnald Hudsn C. Dale Zinn Research Reprt Number 156-5F Surface Dynamics Rad Prfilmeter Applicatins Research Prject cnducted fr The Texas Highway Department in cperatin with the U. S. Department f Transprtatin Federal Highway Administratin by the CENTER FOR HIGHWAY RESEARCH THE UNIVERSITY OF TEXAS AT AUSTIN August 1975

4

5 The cntents f this reprt reflect the view f the authrs, wh are respnsible fr the facts and the accuracy f the data presented herein. The cntents d nt necessarily reflect the fficial views r plicies f the Federal Highway Administratin. This reprt des nt cnstitute a standard, specificatin, r regulatin. iii

6

7 PREFACE This is the fifth reprt presenting results frm Research Prject , "Surface Dynamics Rad Prfi1meter Applicatins," which was initiated t carry ut the implementatin and peratin f the Prfi1meter in field and research applicatins. A variety f highway rughness studies have been perfrmed during the prject, including the fllwing: (1) A methd was develped fr calibrating the Mays Rad Meter n a peridic basis. Cntrl prcedures were develped fr determining when a given Mays Meter needed calibratin. (2) A mdel was develped t cmpute a serviceability index, which estimates a human panel rating f riding quality. The serviceability index is calculated in terms f rughness amplitudes and is much less time-cnsuming than a direct evaluatin by a human panel. (3) A pilt-study analysis f the characteristic rughness patterns n asphalt pavements f different types and ages was perfrmed. The crrelatins between rughness and ther types f distress were als studied. (4) Anther pilt study was perfrmed t demnstrate methds f characterizing the rughness patterns n bridge decks and the adjining pavement. Three bridge sites were analyzed in detail with respect t specific effects, such as the bump at the end f the bridge and waves cincident with the bridge spans, and with respect t rughness amplitudes n the bridge and n the adjining pavement. These studies have all been dcumented in interim prject reprts. Additinally, rughness measurements have been made fr use in ther research prjects and fr specific field applicatins. This service has been prvided fr the Texas Highway Department and Texas Transprtatin Institute as well as fr the Center fr Highway Research. The purpse f this reprt is t present the results f a study f the relatinships between varius types f rughness and human ratings f riding quality. The authrs appreciate the suggestins made by the Texas Highway Department representative, Mr. James L. Brwn. The prgramming supprt by Mr. Randy Wallin and Mr. Jack O'Quin and the engineering cnsultatins and v

8 vi SDP measurements made by Mr. H. H. Dalrymple and Mr. Nel Wlf are als greatly appreciated. Hugh J. Williamsn W. Rnald Hudsn C. Dale Zinn August 1975

9 LIST OF REPORTS Reprt N , "Crrelatin Study f the Mays Rad Meter wi th the Surface Dynamics Prfi10meter," by Rger S. Walker and W. Rnald Hudsn, discusses a study f the crrelatin between measurements made with the Mays Rad Meter and the Surface Dynamics Prfi10meter and, based n this study, prvides a set f calibratin, peratin, and cntrl prcedures fr peratin f the Mays Rad Meter using serviceability index values frm the prfi10meter as a measurement standard. Reprt N , "The Use f Spectral Estimates fr Pavement Characterizatin," by Rger S. Walker and W. Rnald Hudsn, discusses the general uses f rad prfile spectral estimates fr pavement characterizatin. A mdel fr predicting serviceability index based n rad prfile amplitude estimates is als described. Reprt N , "Analysis f Characteristic Rughness Patterns in Pavements and the Relatinship Between Rughness and Pavement Distress," by Hugh J. Williamsn and W. Rnald Hudsn, discusses the characterizatin and cmparisn f the rughness n pavements f differing types and ages. The applicatin f digital filtering is treated, and pilt study results are presented. Reprt N ,"The Characterizatin f Rad Rughness n Bridge Decks and the Adjining Pavement," by David B. Law, Hugh J. Williamsn and W. Rnald Hudsn, discusses the characterizatin and cmparisn f the rughness n bridge decks and the adjining pavement. Several methds f analysis are presented fr analyzing varius cmpnents f rughness. Reprt N F, '~ Study f the Relatinship Between Varius Classes f Rad Surface Rughness and Human Ratings f Riding Quality," by Hugh J. Williamsn, W. Rnald Hudsn, and C. Dale Zinn, discusses the develpment, frm rad prfile data and human panel ratings, a set f regressin mdels fr evaluating the severity f several types f rughness. is used t analyze the rughness n a wavelength basis. Digital fi1tec~ng vii

10

11 ABSTRACT In this study, statistical relatinships between each f several classes f rad rughness and human panel ratings 0f riding quality are develped. Fr this purpse, the rughness is categrized thrugh digital filtering methds n the basis f wavelength. Lngitudinal and transverse surface effects are als studied. Multiple regressin analysis is used t relate the panel ratings t rughness as a whle and t the individual types f rughness. By using the mdels s develped, ne can btain fr any given rad sectin a measure f riding quality crrespnding t each f a set f imprtant aspects f rughness. Use f the mdels is demnstrated by analyzing the rughness f an illustrative rad sectin just befre, just after, and a year after an verlay. KEY WORDS: present serviceability rating, serviceability index, rad rughness, digital filtering, regressin analysis, Surface Dynamics Rad Prfilmeter. ix

12

13 SUMMARY A detailed study f the relatinship between rad rughness and human panel serviceability ratings is presented in this reprt. This type f analysis is necessary t determine what types f rughness rad users find mst bjectinable. Withut this knwledge, it wuld be much. mre difficult t assciate meaning with physical rughness measurements. Rad prfile data, btained by using the General Mtrs Surface Dynamics Prfi1meter, were analyzed n the basis f (1) lngitudinal effects, which cause a passing vehicle t accelerate vertically, and transverse effects, which cause a vehicle t rtate slightly, r "rll" abut a lngitudinal axis (a transverse 'wave" ccurs when ne whee1path rises and falls relative t the 1 ther ) and (2) the lengths f the surface waves. Wavelength bands f 4 t 10, 10 t 25, 25 t 50, and 50 t 100 feet (1.219 t 3.048, t 7.620, t 15.24, and t meters) were studied. Because f high-frequency nise prduced by the tape recrder which was used at the time f the measurements, but which has since been replaced by a mre accurate recrder, rughness waves shrter than 4 feet (1.219 meters) culd nt be analyzed. The fllwing results were btained by determining what prprtins f the rad-t-rad variatins in PSR culd be '~xp1ained" r predicted in terms f all types f rughness cmbined and in terms f rughness with the specific wavelength bands listed abve. fr cncrete and asphalt pavements. Separate studies were perfrmed (1) Fr bth the cncrete and asphalt cases, multiple crrelatins f.91 were achieved by predicting PSR in terms f all rughness types cmbined. Thus, verall rughness and PSR are very clsely related. 1The wavelengths fr lngitudinal and transverse waves and bth measured lngitudinally alng the radway. xi

14 xii (2) The characterizing measures f the rughness with 4 t 10-ft (1.219 t meter) wavelengths have almst as much predictive value as all the rughness types cmbined. Crrelatins f.86 were btained fr bth types f pavement fr the 4 t 10-ft (1.219 t meter) wavelengths. (3) The strength f the relatinship between PSR and rughness within specific wavelength bands decreases steadily as the wavelength increases. The decrease is nt drastic, hwever, since crrelatins f.75 and.68 fr cncrete and asphalt pavements, respectively, were btained fr the 50 t 100-ft (15.24 t meter) case. Additinal regressin studies in which transverse and lngitudinal rughness are treated separately and analyses f certain rughness prperties f cncrete and asphalt pavements are als included in the reprt. Several applicatins f the regressin mdels are discussed, and their use in befre and after-maintenance studies is demnstrated by the analysis f measured prfiles.

15 IMPLEMENTATION STATEMENT New capabilities have been develped fr analyzing the rad rughness measurements btained by using the General Mtrs Surface Dynamics Prfilmeter. These capabilities include (1) digital cmputer methds fr calculating characterizing measures f varius types f rughness and (2) statistical methds fr evaluating the different physical measures n the basis f their relatinships t human panel ratings f riding quality. Several areas f applicatin f these methds are discussed in the reprt, and their use in befre and after-maintenance studies is demnstrated by the analysis f measured prfiles. xiii

16

17 TABLE OF CONTENTS PREFACE LIST OF REPORTS ABSTRACT SUMMARY. IMPLEMENTATION STATEMENT v vii ix xi xiii CHAPTER 1. INTRODUCTION Purpse Backgrund and Applicatins Scpe f Reprt CHAPTER 2. LITERATURE SURVEY The Serviceability Rating Serviceability Predictin: SI Predictin f PSR in Terms f Pwer Spectra. 6 Digital Filtering and Amplitude-Frequency Distributins. 8 Assciatin f Meaning with the Regressin Cefficients in Estimating PSR CHAPTER 3. THE DETAILED RELATIONSHIP BETWEEN SERVICEABILITY AND SURFACE ROUGHNESS Calculatin f Rughness Measures. Illustrative Test Case Illustrative Test Case CHAPTER 4. PROPERTIES OF THE SI MODELS Crrelatin with PSR Standard Errr Number f Terms in Mdel. Brief Cmments n the Rughness Prperties f Asphalt and Cncrete Lngitudinal and Transverse Rughness Studied Separately Cmments n Differing Methds fr Relating PSR t the Cmpnents f Rughness... Cmparisns Between the Existing SI Mdel and the New Mdel Validatin xv

18 xvi CHAPTER 5. SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS Backgrund Summary and Cnclusins f this Study 58 Recmmendatins 60 REFERENCES APPENDICIES Appendix 1. Appendix 2. Appendix 3. Appendix 4. Appendix 5. SI Regressin M~els 67 Cmments n a Pssible Future Study f the Relatinship Between PSR and Vehicle Speed 83 Rughness Data Used as Predictr Variables in this Study. 91 Appraches fr Develping a Single SI Regressin Mdel which is Usable fr Sensitivity Analysis 109 An Investigatin f Signal Prcessing Techniques fr the Purpse f Characterizing Rad Rughness 117 THE AUTHORS

19 CHAPTER 1. INTRODUCTION PUrpse The bjective f this study is t develp methds fr quantifying the severity f the imprtant aspects f rad rughness; fr this purpse, the rughness is categrized n the basis f wavelength and n the basis f lngitudinal versus transverse effects. The methds f evaluatin are develped by statistically analyzing (thrugh multiple regressin techniques) the relatinship between human panel ratings f riding quality and rad-surface measurements. The calculatin frm a rad-surface prfile f characterizing measures f the varius aspects f rughness is a necessary preliminary step t the statistical analysis described abve. The relative merits and the pitfalls f a number f mathematical techniques, including pwer spectral analysis and digital filtering, are discussed frm the standpint f characterizing the imprtant classes f rughness. Backgrund and Applicatins The evaluatin f the cnditin r quality f a rad is an essential jb f the highway engineer. It is nly thrugh this type f evaluatin that a number f functins, all f which are required if high-quality rads are t be prvided with reasnable expenditure f public mney, can be perfrmed. The fllwing list f applicatins illustrates the imprtance f an adequate methd f evaluating rad cnditins. (1) New Pavement Evaluatin. It is necessary t have a set f specificatins fr accepting new pavements. The specificatins prvide guidelines fr use during cnstructin and, thus, prevent many pssible misunderstandings between the gvernmental agency respnsible fr the cnstructin and the staff perfrming the field wrk; withut a frmal set f acceptance criteria, the supervising agency wuld have cnstant difficulties in cmmunicating its criteria fr acceptance t the cnstructin cntractrs. Mrever, it wuld be difficult withut the specificatins t accept r reject new cnstructins n a cnsistent basis. Clearly, the frmulatin 1

20 2 f a set f acceptance criteria requires a frmally defined methd f evaluating the quality r cnditin f the pavement. (2) Allcatin f maintenance fr existing pavements. It is necessary t evaluate existing pavements in rder t diagnse prblems and prescribe maintenance. The allcatin f maintenance resurces can be made in a beneficial and efficient way nly if the pavement evaluatins n which the allcatin is based are bth meaningful and cnsistent. An evaluatin is meaningful if it reflects r characterizes prperties f the pavement which determine the pavement's present r future ability t serve the public. (3) Pavement research. Thrugh research, imprvements can be made t existing methds fr pavement design, cnstructin, and rehabilita- ~ tin practices. Because f the cmplexity f the pavement system, these bjectives cannt be achieved thrugh purely theretical studies. Thus, evaluatins f existing pavements are needed t supply data fr research purpses; the netwrk f public rads prvides a tremendus labratry fr this purpse. These cmments apply t the special area f rehabilitatin. Pavement evaluatins immediately befre, immediately after, and at successive pints in time fl10wing rehabilitatin f a selected set f pavement prjects can be made t answer questins such as the fllwing: (a) (b) (c) D the rehabilitin methds actually crrect the cnditins they are designed t crrect, r d certain prblems, such as surafce rughness with certain ranges f wavelengths, remain after the wrk is dne? Which rehabilitatin practices are mst effective fr crrecting which types f rad surface deteriratin? What are the cntinuing benefits f rehabilitatin? What types f prblems tend t recur within a reasnably shrt time after the maintenance is cmpleted? These three questins, stated abve in general terms, relate t all aspects f the rehabilitatin prcess which affect the ultimate quality f the wrk. The peratinal prcedures, chice f equipment and material, and design parameters, such as thickness f an verlay, are examples f areas fr evaluatin in a pavement rehabilitatin research prgram. It is clear that if evaluatins made ver perhaps a severalyear perid were t be cmpared, a cnsistent means f perfrming the evaluatins wuld be a critical factr. An example f an analysis f a pavement's cnditin befre and after an verlay is discussed in the fllwing chapter. The three areas discussed abve are intended t illustrate the extensive practical need fr the evaluatin f pavements. The research prblems invlve sme very cmplex physical prcesses and will cntinue t be studied in the

21 3 future. The peratinal needs, such as the assessment f rad cnditins 1nrder t prescribe maintenance, will exist as lng as paved rads are in use. Scpe f Reprt Perhaps the tw mst cnmn types f pavement evaluatin are the assessments f cnditins which relate t (1) prbable future deteriratin and (2) the present ability f a rad t serve the public. Bth types f evaluatins cme int play in all f the three areas discussed abve. The subject f the research reprted herein is the develpment f new methds f the secnd type. The new methds are designed t give a detailed characterizatin f the rad-surface quality and, thus, t prvide the practicing highway engineer with imprved capabilities fr perfrming functins in the three areas discussed abve. The methds are based n the predictin f human-panel evaluatins f the quality f a rad by using descriptrs f different aspects f rad rughness as predictr variables. Thus, several predictive mdels are develped, each relating t a certain aspect f riding quality. In Chapter 2, the current state f the art and basic cncepts f present pavement-surface evaluatin are discussed by summarizing several published papers and reprts which are clsely related t this study. In Chapter 3, the nature and practical advantages ver existing appraches f the new mdels are discussed alng with illustrative test cases. The test cases demnstrate the types f physical insights which can be gained abut a rad's cnditin by using the mdels and serve as the basis fr discussin f certain applicatins in the area f maintenance evaluatin. A discussin is given in Chapter 4 n varius characteristics f the mdels and the physical prperties n which they are based. The tw bjectives f this chapter are (1) t prvide ptential users f the mdels with backgrund infrmatin which will be beneficial in applicatins and (2) t discuss certain relatinships between human riding quality and rad-surface defrmatin which are felt t be f practical significance. The summary and cnclusins f the study are given in Chapter 5. The mdels themselves are presented in Appendix 1. Appendix 2 includes a discussin f the design f a pssible future experiment t assess the effect f

22 4 vehicle speed n riding quality. The data frm which the mdels were develped are presented in Appendix 3, and an alternate statistical apprach t the ne used in this prject is discussed in Appendix 4. The mathematical methds used t extract the meaningful infrmatin frm the large set f measured data required t describe a rad surface are discussed in cnsiderable detail in Appendix 5.

23 CHAPTER 2. LITERATURE SURVEY The purpse f this chapter is t intrduce the basic cncepts f the evaluatin f the present quality f a rad sectin and t summarize and discuss the state f the art. It is felt that this bjective is best achieved by discussing in sme detail several published papers and reprts which are bth imprtant and clsely related t this study. A number f ther publicatins are referenced in the fllwing chapters and appendices as their relevancies arise. The Serviceability Rating Ref 5. A methd fr rating the present quality f a rad is discussed in The basic term, present serviceability, is given the fllwing defini tin: "the ability f a specific sectin f pavement t serve highspeed, high-vlume, mixed (truck and autmbile) traffic in its existing cnditin." As the term itself and its definitin bth indicate, present serviceability relates t present cnditin nly and nt t the past r future cnditin f the pavement. The Present Serviceability Rating (PSR), then, is defined as the mean f the evaluatins f present serviceability made by a human panel. panel is I'intended t represent all highway users," n a scale frm 0 t 5 n the basis f the fllwing: 4-5 very gd 3-4 gd 2-3 fair 1-2 pr 0-1 very pr. The The ratings are made Serviceability Predictin: SI The PSR is a meaningful measure in that it is a direct evaluatin by rad users. The panel ratings, hwever, are time cnsuming and, therefre, are expensive. Thus, there was a need fr an alternate evaluatin methd. 5

24 6 The apprach which was intrduced in Ref 5 and which has subsequently been taken by several ther investigatrs (Refs 11, 16, 19, 20, and 21) is (1) t btain (a) panel ratings and (b) physical measurements bth fr the same set f rad sectins and then (2) t use multiple regressin t btain an equatin t predict PSR in terms f the physical measurements. An estimate f a human panel rating f a rad can be btained fr any rad sectin by first btaining the physical measurements necessary fr the predictin and then evaluating the predicting equatin. s btained is called the Serviceability Index (SI). The PSR estimate In Ref 5, equatins are given t cmpute SI in terms f a rughness term, slpe variance; the area f cracking r patching per 1000 square feet (92.90 square meters) f pavement surface area; and rut depth, which is the average surface depth in the wheelpath cmpared t "a line jining tw pints each tw feet [.6 meters] away (transversely) frm the center f the wheelpath." The rughness term, slpe variance, was cmputed as fllws: Fr each wheelpath, "a cntinuus recrd f the pavement slpe between pints 9 in [22.86 centimeters] apart" was btained. Then the slpes were sampled, "generally at l-ft [.3048 meter] intervals, ver the length f the recrd" fr each wheelpath. Then the variances f bth f the tw sets f slpes were cmputed, and the tw variances were averaged t btain the slpe variance. Separate equatins were derived fr asphaltic cncrete and fr prtland 1 cement cncrete pavements. The multiple crrelatins btained fr the asphalt and cncrete cases are.92 and.96, respectively. The standard errrs are.38 and.32, respectively. Predictin f PSR in Terms f Pwer Spectra The wrk reprted in Ref 5 shws that PSR can be predicted with high crrelatin in terms f physical measurements. Because f (1) the efficiency and accuracy with which rughness measurements can be made using mdern highspeed equipment and (2) the clse relatinship demnstrated in Ref 5 and ther 1 The terms "cncrete" and "asphalt" will, fr simplicity, be used hencefrth instead f "prtland cement cncrete" and "asphaltic cncrete," respectively.

25 7 studies between rughness and PSR, subsequent wrk in this area has invlved primarily PSR predictin in terms f rughness. Reference 21 presents a study in which an SI mdel invlving nly rughness measurements is btained. The measurements were made using the General Mtrs Surface Dynamics Prfi1meter (SDP). The measurements are in the frm f a rad prfile, i.e., a recrd f rad elevatin versus distance alng the rad, fr bth the right and left whee1paths. Discussins f the measuring system itself are given in Refs 12, 16, 18, and 22. A brief summary f the system is given at the beginning f Appendix 5. Althugh the slpe variance is a meaningful characterizing measure 2 f the rad rughness, n single number culd cntain all f the infrmatin which is inherent in a measured rad prfile. Because f the large number f pints required t describe a rad surface, hwever, sme srt f scheme fr cmputing a small set f summarizing rughness measures is necessary. In the study presented in Ref 21, pwer spectral analysis is used t calculate a set f summary measures f the rad rughness. The mathematics f pwer spectra as they are applied t rad prfile analysis is discussed in sme detail in Appendix 5. Spectral analysis can be thught f as a methd fr decmpsing the rughness in the rad prfile int cmpnents by wavelength. A rt-mean-square (r.m.s.) amp1itude 3 is cmputed fr each f a discrete set f wavelengths. This type f analysis is relevant t ride quality, since rad surface irregularities f different wavelengths induce different types f mtin in passing vehicles. A regressin mdel invlving rughness amplitudes and a dummy variable t accunt fr visual r auditry differences between cncrete and asphalt pavements was develped. The mdel includes rughness terms representing 2 It will be cnvenient t use the term "measure" t refer t any quantity which is calculated frm a set f measurements, such as a rad prfile, and which is intended t characterize a certain prperty, such as the rad rughness with wavelengths within a given band. 3The calculatin f amplitudes frm pwer spectra is discussed in Appendix 5. Als discussed are the meaning and the calculatins f crss-amplitude terms fr a discrete frequency set.

26 8 4 wavelengths frm 8.6 t 86 feet (2.6 t 26 meters); althugh they are significant, amplitudes f shrter waves had t be excluded because f highfrequency tape recrder nise. chapter. The mdel has 22 terms, a multiple crrelatin f.94, and a standard errr f.33. This pint is discussed further in the next Thus, it was shwn that it was pssible t predict PSR with high crrelatin frm rughness terms cmputed frm measured rad prfiles. Digital Filtering and Amplitude-Frequency Distributins It is pinted ut in Ref 4 that, althugh the pwer spectral apprach is effective, the calculatin f a single verall amplitude crrespnding t a given frequency is less than ideal. A mre meaningful characterizatin f the rughness wuld include infrmatin abut the amplitude variatin as well as the verall average amplitude; any number f cmbinatins f small and large bumps culd prduce the same verall r average amplitude. In Ref 4, a methd fr cmputing a prbability distributin f the rughness amplitudes is given. (1) btain a rad prfile measurement, This is achieved by the fllwing three steps: (2) filter the measured prfile s as t islate the rughness with wavelengths within a specified band, and (3) cmpute the sample distributin functin f the peaks in the filtered prfile. The term "filtering" refers t a transfrmatin, which can be perfrmed either by electrnic hardware r by digital calculatin, thrugh which an artificial prfile is cmputed frm the measured rad prfile. It is pssible t design the filter s that certain types f surface irregularities are eliminated, but ther types are essentially unaffected, by the filtering peratin. In this way, it is pssible t islate fr further study the rughness in the 4'Actually, 8.6 feet (2.6 meters) is the center f a band f wavelengths which are included. The band extends frm 7.8 t 9.5 feet (2.4 t 2.9 meters). The width f each f the discrete set f frequency intervals is.0116 cycles per ft (.0381 cycles per meter).

27 9 measured prfile with wavelengths within certain limits. This simplified descriptin f filtering is given here t facilitate the discussin f the characterizatin f rad rughness. A much mre extensive treatment f filtering techniques and their applicatin in prfile analysis is given in Appendix 5. The figures in the fllwing chapter illustrate the filtering cncept by displaying measured and filtered prfiles n the same plt. By perfrming the three steps listed abve, ne btains a detailed descriptin f the rughness with the selected range f wavelengths; the median is an "average" rughness measure, while the pints in the upper tail f the distributin characterize the mst severe rughness in the rad sectin. If the three steps are perfrmed fr a set f cntiguus wavelength bands which cvers the range f interest, then an extensive descriptin f the lngitudinal rughness is btained. The set f distributin functins is called the "Amplitude-Frequency Distributin." A technical statistical aspect f the predictin f PSR frm a set f rughness measures is discussed in the fllwing sectin. This discussin is relevant t this study and shuld be f interest t researchers wrking in the ride quality field. The reader wh is interested primarily in the results f the study, hwever, may want t skip t the beginning f Chapter 3. Assciatin f Meaning with the Regressin Cefficients in Estimating PSR A discussin is given in Ref lin certain prblems which are encuntered in the regressin f PSR n a set f highly crrelated rughness values - specifically, pwer spectra. The prblems arise because f the relatinships amng the amplitudes f the rughness f different wavelengths; a rad which has severe rughness f, say, wavelength 10 feet (3.048 meters) is likely als t have wrse-than-average rughness f wavelength 15 feet (4.572 meters). Althugh exceptins exist, the pint is that the pwer spectral values have high intercrrelatins. Thus, the presence f a given variable in the mdel can have extreme effects n the cefficients f the ther variables. It is even pssible fr a pwer value t have a psitive cefficient, which wuld seem t indicate that as a particular type f rughness became wrse, the serviceability wuld increase. The explanatin, f curse, is that ther terms, which crrelate psitively with the rughness term which has a psitive cefficient, are present with negative cefficients; thus, the net effect

28 10 f an increase in the severity f the rughness is, in all likelihd, a decrease in SI. Nevertheless, it is impssible t infer anything abut the relatinship between PSR and a particular type f rughness frm the cefficient f a term in the SI mdel. Stepwise regressin (Refs 6 and 7) is a statistical methd fr selecting the subset f a cllectin f independent variables which has the greatest cmbined value fr predicting a specific dependent variable. It is pinted ut in Ref 11 that, althugh it may appear that this type f selectin wuld slve the prblem by limiting the terms in the mdel t thse with the greatest predictive value, it des nt. Randm sampling variatins playa large rle in determining which variables enter the mdel. Frm Ref 21, mrever, we knw that psitive cefficients still appear in the mdel, even if stepwise prcedures are used. The nature f stepwise regressin is discussed briefly in Chapter 4. Thus, certain cnstraints fr cntrlling the values f the cefficients were investigated in Ref 11. First, the regressin mdel is written SI = -y+ N t ~. s. i 1 ~ ~ where the Nw, define s. are measures f the rughness at a discrete set f frequencies. ~ ~. ~ k n ~ where i 1, 2,... Nand n < N. Thus, the weight ~. assciated with the rughness at the ith frequency is ~ a plynmial functin f the frequency. The Serviceability Index, then, can bviusly be expressed as a functin which is linear in the des nt directly invlve the ~'s a's and which Multiple regressin analysis is used t determine the a's, which in turn determine the ~'s. There are several

29 11 prblems assciated with this methd, hwever, including the fact that there is n explicit cnstraint n the signs f the cefficients. Thus, the fllwing frmulatin was used: ~. ~ k i where i = 0, 1,.. N. (The cefficient f the expnential functin is apparently intended t insure zer weight assciated with the zer-frequency cmpnent). Thus, the ~'s are cnstrained t be psitive (the authrs f Ref 11 are evidently using a P5R scale which increases with decreasing ride quality; therwise, the requirement that all the cefficients be psitive des nt make sense). Nnlinear multiple regressin is used, then, t cmpute the befre, determine the ~'s a's, which, as The methd discussed in Ref 11 was shwn t be successful in that an 51 mdel was develped in which the cefficients (1) are all psitive and (2) have a single peak crrespnding t a wavelength f abut 8 feet, suggesting that humans are mst sensitive t rughness with wavelengths in this range. This result is intuitively plausible and is cnsistent with the wavelength study presented herein. Mst imprtantly, n inference whatsever abut the relative imprtance f the different wavelengths can be drawn frm ther studies in which regressin analysis is applied in a straightfrward way. Additinally, Ref 11 identified and explained a statistical fine pint which has imprtant cnsequences and, therefre, paved the way fr further research in relating P5R t individual types f rughness. Nevertheless, the fllwing pints shuld be taken int accunt. (1) When the regressin techniques select a mdel with cefficients which, when taken individually, seem nt t make sense, it is because these terms when taken as a whle give the best fit t the data. Any restrictins which are placed n the mdel are

30 12 likely t decrease the extent t which the regressin mdel fits the experimental data n which it is based. Neither the multiple crrelatin nr any ther infrmatin is given in Ref 11 which wuld allw the reader t assess the adequacy f the fit f the mdel. S If the bjective f develping the mdel were t predict PSR in the mst accurate and physically realistic way, then these cnsideratins, nt the physical meaningfulness f the individual terms, shuld be the basis f evaluatin f the mdel. The relatinships between PSR and the individual types f rughness are f interest, hwever, fr numerus practical reasns; fr example, sme srt f judgement must be made abut which types f surface dereriratin are undesirable in rder t allcate highway maintenance resurces. If these relatinships are t be examined by means f a single mdel invlving the entire set f rughness measures, then ne wuld require a reasnable agreement between the mdel and the data; therwise, there is n empirical justificatin fr use f the mdel fr any purpse. (2) The methd discussed in Ref 11 is dependent n the assciatin f a single variable, wavelength, with the rughness measures; it is n this basis that the plynmials are frmulated. If the methd were t be applied using an amplitude-frequency distributin instead f a pwer spectrum as the set f predictr variables, then the tw variables, wavelength and anther variable assciated with the distributin functins, wuld have t be included. The prbability level assciated with the varius amplitudes in the distributin, fr example, culd be the secnd variable. Furthermre, sme accunt shuld be made f the fact that the imprtance f the varius prbability levels might vary with frequency; this culd be taken care f by including crss-terms in the plynmial. The plynmial wuld becme even mre cmplex if a set f transverse rughness terms were added t the amplitude-frequency descriptin f the lngitudinal waves. The cmments abve are nt intended as criticism f the plynmial apprach as it is used in Ref 11. The pint is that the apprach might becme unwieldy when applied t very detailed rughness characterizatins. In summary, the Serviceability Index, is bth (1) a valuable measure f riding quality, since it is clsely related t human panel ratings f rideability, and (2) ecnmical t btain, since it is calculated in terms f physical measurements and des nt require a panel rating. Because f SA plt f "residuals 2 " versus K is given, but "residuals 2 " is nt defined as a Sum f squares, mean square, etc. In any case, additinal infrmatin wuld be needed t calculate the multiple crrelatin.

31 13 the clse relatinship between riding quality and rughness, research emphasis has been placed n the develpment f 81 mdels in terms f rughness. In the fllwing chapter, a methd fr extending the 81 cncept t btain a mre detailed riding-quality evaluatin is presented.

32

33 CHAPTER 3. THE DETAILED RELATIONSHIP BETWEEN SERVICEABILITY AND SURFACE ROUGHNESS While the Serviceability Index, which is discussed in the preceding chapter, is a meaningful verall measure f the riding quality f a rad, n single number can adequately characterize a cmplex pattern f surface waves. Sets f summarizing measures, such as the amplitude versus wavelength functin and the amplitude-frequency distributin, describe the physical prperties f the surface, but are hard t interpret frm the standpint f riding quality. A rise and fall f a given amplitude, say 1 inch (2.54 centimeters), causes a much mre severe sensatin if it ccurs ver a 2-ft (.6096-meter)-lng interval than if it ccurs ver 50 feet (15.24 meters) alng the radway. Thus, there is a need fr a set f characterizing measures similar t the verall Serviceability Index which has been used in the past. The cllectin f prfile measurements and accmpanying panel serviceability ratings which were used t develp an SI mdel emplying pwer spectra (Ref l6,and 21) were used t develp such a set f measures. terms, the mdel develpment was accmplished by predicting PSR frm Stated in brief (1) an verall set f rughness amplitudes characterizing the rad rughness and (2) subsets f the verall set f rughness measures which characterize, respectively, rughness with the fllwing wavelengths: 4 t 10 feet (1.219 t meters), 10 t 25 feet (3.048 t meters), 25 t 50 feet (7.620 t meters), and 50 t 100 feet (15.24 t meters). The mdel fr the 4 t 10-ft (1.219 t meter) wavelengths, fr example, is an equatin f the frm S I = f (M l, M 2, M n) where SI is the Serviceability Index crrespnding t this specific type f rughness, and the M. 1. are the characterizing measures f this class f 15

34 16 rughness. (The additinal inclusin f dummy variables is discussed belw. ) Because the prfile measurements were made in 1968 when the SDP was equipped with an lder-style tape recrder, nise prevented the analysis f rughness with wavelengths shrter than 4 feet 1 (1.219 meters). While the nise prblem has been mitigated cnsiderably by installing a newer tape recrder with better reslutin, it was necessary fr the purpses f this study t use the prfile measurements that were made at the time f the rating panel evaluatins. At this pint, we digress slightly t describe the calculatin f the characterizing rughness measures used in this study. Althugh the highly technical details are deferred t Appendix 5, it is felt that the fllwing brief discussin will aid the reader in assciating physical meaning with the rughness terms used in the SI mdels. Calculatin f Rughness Measures The rughness measures which describe the lngitudinal rughness fr either whee1path fr, say, the 4 t 10-ft (1.219 t meter) band are calculated by a three-step prcess: (1) By using digital filtering, calculate frm the measured prfile an artificial prfile cntai~ing nly the 4 t 10-ft(1.219 t meter)-lng waves. 2 (2) Calculate a mving rt-mean-square (r.m.s.) amplitude centered at each pint in the sectin, excluding the values which wuld be distrted by end effects. Each r.m.s. amplitude describes the rughness alng a 10-ft (3.048-meter) interval, which is the lngest wavelength f the rughness t be islated by filtering. lit is stated in the preceding chapter that in the riginal study reprted in Ref 21, wavelengths shrter th'an abut 8 feet (2.438 meters) were excluded. In the present study filtering was used t eliminate the nise, which was predminatly in the 2 t 3-ft (.610 t.914-meter)... wave1ength range. Visual examinatin f measured and filtered prfile plts was used t verify that the large-amplitude tape recrder nise did nt cntaminate the infrmatin in the 4 t 10-ft (1.219 t meter)-wave1ength range. 2 Actually, perfect. which was the cutff at the edges f the frequency band f interest is nt The characteristics f the sixth-rder digital Butterwrth filter used are discussed in Appendix 5.

35 17 (3) Calculate the sample distributin f the lcal rughness measures. 3 The fiftieth percentile pint, the value which is greater than r equal t exactly half the rughness amplitudes, is then an verall r average measure f the sectin rughness. The ninetieth percentile pint, the value greater than r equal t exactly 90 percent f the amplitudes, indicates the severity f the wrst rughness in the sectin. The three steps are perfrmed fr bth the right and left whee1path prfiles. Then, since the runs were nt cnsistently made in either the inside r utside lane, n meaning can be assciated with the right as ppsed t the left whee1path. Fr this reasn, the crrespnding rughness measures fr the right and left prfiles, e.g., the fiftieth percentile amplitudes, are averaged t btain a set f lngitudinal rughness measures. Nw, the transverse rad-surface irregu1atities are als f interest, since they cause passing vehicles t "rll" abut a lengthwise axis and since this type f mtin adversely affects the riding quality f the rad. T study the transverse rughness, we first btain an artificial prfile by cmputing the pintwise difference between the successive right and left prfile elevatins. Changes in this prfile, then, crrespnd t changes in elevatin f ne whee1path relative t the ther and, thus, directly relate t vehicle rll. The difference prfile is als prcessed by the steps (1), (2), and (3) listed abve t btain transverse rughness measures. The wavelengths have the same meaning as in the lngitudinal case; they are measured lngitudinally alng the rad surface. Suppse that the right whee1path gradually rises by.1 inches (.254 centimeters) relative t the left whee1path and then gradually returns t its riginal elevatin and that the sinusidal rise and fall ccurs ver a distance f 10 feet (3.048 meters) alng the radway. Then this transverse surface irregularity has a wavelength f 10 feet (3.048 meters). 3Ntice that this apprach differs frm the amplitude-frequency methd discussed in the previus chapter in that the distributin f lcal r.m.s. rughness measures is cmputed rather than the distributin f peaks in the prfile. It is shwn in Appendix 5 that the lcal r.m.s. values are less susceptible t certain filter-induced distrtin than the peaks are. Fr this reasn, the lcal r.m.s. values rather than the peaks were used in this study. This is nt a critical pint, hwever; the amplitude-frequency apprach is a valid and natural way t characterize rad rughness.

36 18 The SI mdels which are discussed in the fllwing sectins include as predictr variables bth lngitudinal and transverse rughness terms and a dummy variable. The dummy variable T is defined as fllws: T = 1 if the pavement is cntinuusly reinfrced cncrete (CRG) = 0 therwise fr cncrete pavements and T = 1 if the pavement is surface-treated (ST) = 0 therwise fr asphalt pavements. The mdels are given in detail in Appendix 1. The reasn fr including the dummy variables is t accunt fr visual r auditry differences experienced in riding ver different types f pavements. The sund f the car passing ver the jints in sme cncrete pavements (but nt CRC), fr example, might influence PSR, but might nt be adequately reflected in the rughness terms. Illustrative Test Case 1 Prfile measurements were made n a ht-mix asphalt-cncrete rad sectin n IH20 near Odessa, Texas, just befre (March, 1974) and just after (April, 1974) an verlay. The pavement which had had a surface treatment, was badly cracked and had severe shrt-wavelength rughness. A 1 1/4- inch (3.l75-centimeter) ht-mix verlay was perfrmed t alleviate this situatin. Befre and after-verlay prfile plts are shwn in Figs 3.1 and 3.2, respectively, fr the Odessa sectin. An artificial prfile cmputed by filtering is shwn as an illustratin f the islatin f that part f the rughness with a certain range f wavelengths. Additinal plts f filtered BOp unfiltered prfiles fr these and ther sectins are presented in Appendix 5. The SI values btai ed frm the mdel which is discussed bth in Ref 21 ad in the preceding chapter are 3.5 befre the verlay and 4.0 after.

37 UNFILTERED - SOLID LINE FILTERED- DOTTED LINE Left Wheelpath Frame 1 0 ~ 0 Z t-----< 0 r-j Z 0 r- f- CC O >N L1.J~.J I W O::u O(!) cr: This filtered prfile was cmputed mathematically frm the measured prfile in rder t islate fr further analysis the rughness with 4 t lo-ft wavelengths. 7 lji""1""h'~... ~~~,--"-.- -L\-~~WlJt!~Atf' n' Mftftiil ~I' &~'. "J" n!'ji"fpiiii.t" 1\, ',. _ "J. '\, ",..... I'... a. '-' 1 ~, '-~ a',f '\~ - I I CJJ ~ I T r ,. r I I I ~O!SO.OO ~ POSITIO~ RLONG THE RORD (FT. 1 1 ft =.3048 meters 1 inch = centimeters Fig 3.1. Odessa Test Sectin, March, 1974: Rughness with 4 t 10-ft wavelengths islated by filter. Befre verlay. (cntinued) t--' \0

38 UNFILTERED- SOLID LINE FILTERED.- DOTTED LINE Left Wheelpath Frame 2 "-> (!:' z..., ~. 0 c~ 20 I ITO ;> r~ llj I W \/\f\/\ j IT ~~ ~t----~,-----,--- '-'-"~ G.GO CJO :) 2 0. :)0 :; ['; 0. ':J C; 0.!J 0 POSITION RLONC THE RORQ [FT. 1 Fig 3.1. (Cntinued)

39 UNFILTERED SOLID LINE F1LTERED - DOTTED LINE Left Wheelpath Frame 3 c t--1 _,, _, /,,-1-1""1\.1'11 ~._t!. h 1\ I \. /' ""'I',....,1'1.-- ~<7"'~'~-!f N II' I, ' -~. n \ J \1. \, CI Dc 0:: 0 L-' T -,---- _n._. I -,-- mm I I O.CJO 8!JC.CJC 840.~JG O.CJO 9?O.ClCi CL% POS1TION RLONG THE ROAD fft.) Fig 3.1. (Cntinued) N...

40 a UNFILTERED- SOLID LINE FILTERED - DOTTED LINE Left Wheelpath Frame 1 N N a t a Z >-I ~ a r>j Za..., /-. 0: 0 >N w....]0 w 0: 0 O(C 0: 0 I a ' I l f I f I POSITION RORD (FT.) 1 ft =.3048 meters 1 inch = centimeters Fig 3.2. Odessa Test Sectin, April, 1974: Rughness with 4 t 10-ft wavelengths islated by filter. After verlay. (cntinued)

41 a UNFILTERED - SOLID LINE FILTERED - DOTTED LINE Left Wheelpath Frame 2 z - a (D a a r-j Za l--< f IT > c-j w~ --.J 1 W IT Or..c era I ,- J, rm I 1,! I ~O S20.QO POSIT10N RLONG THE RORe (FT.l Fig 3.2. (Cntinued) N W

42 UNFILTERED - SOLID LINE FILTERED - DOTTED LINE Left Wheelpath Frame 3 N +' ld z ~ 0 N Z t----, I-- IT ::> (-J ~~ ----.J I L1J IT OlD 0::: 0 I I I, I I I ~ GO 840.GO GO 1040.c]O 1080.% POSITION RLONG THE RORD (FT.) Fig 3.2. (Cntinued)

43 25 The new mdels were als applied t this sectin. The 81 values which were btained are given in Table 3.1. The fllwing pints shuld be nted abut this test case. (1) The verall 81 value befre the verlay is lwer than the value fr the existing 81 mdel presented in Ref 21 (3.13 versus 3.5). The lwer value is cnsidered t be mre reasnable because f the severely distressed cnditin f the pavement. (2) Fr cnditins just after the verlay, the 81 value cmputed with the new mdel is higher than the value cmputed with the existing mdel (4.0 versus 4.25); mre imprtantly, the 81 change is.5 fr the existing mdel and 1.12 fr the new mdel. The larger change is cnsidered t be mre reasnable in view f the dramatic imprvement f the shrt waves. (3) It is evident frm the rad-prfile plts that, althugh the shrt-wavelength rughness is imprved cnsiderably, there is an increasing similarity between the befre and after-prfiles as the wavelength increases. Fr the prfile befre the verlay, the 81 values are very lw and increase as the wavelength increases. The 81's are unifrmly high after the verlay. The greater imprvement f the shrt waves was expected in view f the cnstructin prcedure. A 25-ft (7.6l8-meter) skid is custmarily used t perfrm this type f verlay, which means that waves shrter than 25 feet (7.618 meters) are cntrlled, but lnger waves are nt. The abve test case is given as an illustratin f the use f the new 81 mdels. It is felt that the mre detailed evaluatin f the rad rughness gives a much clearer picture f the nature f the imprvement achieved by the verlay than a single 81 value des. imprtant area f applicatin f the SI mdels: effectiveness f the different maintenance techniques. The test case als indicates an the evaluatin f the The Odessa sectin was chsen because the nature f the imprvement achieved by the verlay is easy t see frm the befre and after-maintenance prfile plts; thus, it is pssible t verify that physically realistic and meaningful results are btained frm the 81 mdels. The very cmplex nature f a typical rad user's subjective evaluatin f the present quality f a rad sectin is, in general, nt easy t analyze. If such evaluatins were made even by the same set f highway engineers r ther raters befre and after a maintenance prgram, randm human variatins might prevent an accurate evaluatin f the benefits f the prgram. A change in the weather, fr example, might bias the results cnsiderably. If it were desired t evaluate the cntinuing effect f the maintenance by btaining ratings at say, six-mnth intervals fr several years, then the

44 26 TABLE 3.1. SI VALUES BEFORE AND AFTER OVERLAY Wavelength (feet) All Rughness 4 t t t t 100 Befre After Imprvement NOTE: 1 ft =.3048 meters

45 27 extraneus effects, which wuld prbably be exacerbated by persnnel changes, culd have seriusly deleterius effects n the evaluatin prgram. The 81 values btained frm prfile measurements, hwever, prvide a relatively cnsistent means f perfrming repeat evaluatins at successive pints in time. Furthermre, since the mdels are develped frm P8R data, the 81 values are clsely related t a human rating f ride quality. The rughness amplitudes n which the serviceability predictins are based are given in Table 3.2. Althugh the rad has a shulder, there is n curb, and the higher amplitudes fr the right wheelpath reflect the mre severe deteriratin in the utside f the lane. It is interesting that the amplitudes remain larger after the verlay fr the utside wheelpath fr the 0 t 4, 4 t 10, and 10 t 25-ft (0 t 1.219, t 3.048, and t meter) wavelengths. This is apparently a residual effect f the riginal rughness. Anther pint illustrated by this test case is that all three types f infrmatin discussed, rad prfile plts, 81 values, and rughness amplitudes, can help the highway engineer t explain the effects f an verlay n the rad-surface cnditin. In sme areas, the lng-wavelength rughness was actually made wrse by the verlay; see, fr example, psitin 280 feet (85.34 meters) in frame 1 f Figs 3.1 and 3.2. That this is a real effect, and nt a spurius measurement, is evidenced by the fact that the same rughness pattern appears in the prfile measured abut a year after the verlay (see Fig 3.3). Table 3.2, hwever, shws that the rughness amplitudes fr the right wheelpath and fr the transverse prfile are decreased by the verlay. Thus, it is reasnable that the 81 values fr all wavelengths are increased. While prcedural decisins cannt be made n the basis f a single test case, the example discussed abve suggests that pavers with sensrs and side wires shuld pssibly be used fr verlays in rder t imprve the lng, as well as the shrt, waves. Care wuld have t be exercised t prevent the intrductin f peridic waves int the prfile because f a sag in the wires between stakes. Additinal test sectins, perhaps affected by a swelling subgrade, with mre severe lng waves wuld be necessary t verify this suggestin. It is nted in the abve test case, hwever, that the prnunced lng (loo-ft r meter) wave in the prfile is virtually unaffected by the verlay. This wave is believed t have been intrduced in the grading