ME PDG Rigid Pavement Design Reliability Update. Further Calibration of the Distress Prediction Models & Reliability Effects

Transcription

1 ME PDG Rigid Pavement Design Reliability Update Further Calibration of the Distress Prediction Models & Reliability Effects

2 NCHRP 1-40B 1 & 1-40D 1 Team Applied Research Associates Michael Darter Jagannath Mallela Harold Von Quintus Gregg Larson Leslie Titus-Glover Chetana Rao Dulce Rufino Alex Gotlif U of Minnesota Lev Khazanovich Graduate students NCHRP Ed Harrigan & Panel ASU Matt Witczak Mohamed El-Basyoumy Claudia Zapata Graduate students

3 Presentation Reliability Definition M-E PDG 2004 Calibration 2005 Independent validation 2006 Recalibration Illustrations of project predictions Conclusions

4 M-E E PDG Design Reliability Definition Rigid Pavement JPCP R F = P [Fault < Critical Fault] R C = P [Crack < Critical Crack] R IRI = P [IRI < Critical IRI] CRCP R PO = P [Punchout < Critical PO] R IRI = P [IRI < Critical IRI]

5 M-E E PDG Design Reliability Output Rigid Pavement IRI, in/mile Critical IRI IRI at 95% IRI at 50% Reliability, Residual Pavement age, years

6 M-E E PDG Design Reliability Output Rigid Pavement Performance Criteria Distress Target Reliability Target Distress Predicted Reliability Predicted Acceptable Terminal IRI (in/mi) Transverse Cracking (% slabs cracked) Mean Joint Faulting (in) Fail Pass Fail

7 Design Reliability Approach IRI model: closed form variance solution: Var {IRI} = Var {Initial IRI} + Var {Distress} + Var {Site Factors} + Var {Residual} Distress models: used model residuals (predicted measured) determined from calibration results

8 Residuals from Calibration Residual = predicted measured Causes Random (replicate) variation Input selection error Distress/IRI measure error Difference between project mean and 500-ft section Model deficiency residual Relative magnitude of each? M e a n tra n sv e rs e jo in t fa u ltin g, in Section Faulting History Age, years LTPP data 1-37A model Resid

9 Residuals from Calibration Residuals represent the knowledge that exists of the accuracy of the distress prediction model. Large model deficiency residuals indicates we are not explaining the physical phenomenon well. Predicted percent slabs cracked Predicted Vs Measured Slab Crack R 2 = 0.86 SEE = 5.02 percent N = Measured percent slabs cracked

10 Derivation of Cracking Standard Deviation Function Predicted percent slabs cracked y = x R 2 = Measured percent slabs cracked

11 Predicted vs. Standard Deviation of Measured Cracking STD C = *CRACK *CRACK Standard deviations, percent cracking measured predicted Predicted cracking, percent

12 JPCP Cracking Reliability CRACK R = CRACK 50 + STD C Z R CRACK R = predicted cracking at reliability level R, percent slabs CRACK 50 = predicted cracking based on mean inputs (50% reliability), percent slabs STD C = standard deviation of cracking at predicted level of mean cracking Z R = standard normal deviate (one-tailed)

13 Residuals of Prediction Variations/Uncertainties NOT included: Future design traffic estimation error Calibration database limitations (short sections, missing some design and materials, etc.) Others???

14 2004 Sections Used 1-37A1 Projects JPCP LTPP GPS-3 & SPS-2 JPCP FHWA RPPR CRCP LTPP GPS-5 CRCP Vandalia, IL CRCP Other Rehab JPCP CPR, OL Rehab CRCP OL No. of Sections Location 23 States 9 States 22 States 20-years, US 40, IL Chicago, Illinois 14 States 10 States

15 2005 Validation: NCHRP 1-40B1 Objective was to evaluate adequacy of models developed under NCHRP 1-37A for ME PDG system. Criteria Pavement sections NOT used in original calibration under NCHRP 1-37A. Pavement sections had mostly level 1 and 2 data available for analysis.

16 2005 Test Sections NCHRP 1-40B1 Extended AASHO (Interstate 80) MnRoad Projects LTPP SPS-2 (Supplemental) LTPP SPS-2 LTPP SPS-2 LTPP GPS-5 (CRCP) I-77 (CRCP) Smart Road (CRCP) No. of Sections Location Illinois Arizona Arkansas Iowa Minnesota CT, IA, MI, TX South Carolina Virginia

17 2006 Recalibration: NCHRP 1-40D1 Objective was to recalibrate models included in the original M-E PDG. DATA: Major increase in quantity & quality of performance data. Original sections: Updated observed distress/iri from 2000 to Added new JPCP, CRCP, and Rehab sections. Combined database: 2004, 2005, & 2006

18 LTPP GPS-3 JPCP Sections LTPP SPS-2, MnROAD, & AASHO JPCP Sections

19 LTPP GPS-5 CRCP Sections Illinois, South Carolina, & Virginia CRCP Sections

20 LTPP, NCHRP 10-41, ACPA Diamond Grinding Sections SPS-6 Sections

21 2006 Recalibration: JPCP Projects No. of Sections Location LTPP GPS-3, SPS States Extended AASHO (AASHO + I-80) MN Road 25 9 Illinois Minnesota Note: Most sections have multiple observations over time

22 2006 Recalibration: CRCP Projects LTPP GPS-5 (CRCP) I-77 (CRCP) Smart Road (CRCP) Illinois Interstate Vandalia, IL No. of Sections (crack spacing) 10 (heavy traffic) 6 Location 22 States South Carolina Virginia Chicago, Illinois US 40 Vandalia, IL Note: Most sections have multiple observations over time

23 2006 Recalibration: Rehab Projects JPCP CPR and OL (SPS-6, GPS-9, NCHRP 10-41, Others) CRCP Overlays (GPS-9, NCHRP 10-40, others) LTPP SPS-6 (HMA overlays) No. of Data Points To be completed Location 15 States 10 states Note: Most sections have multiple observations over time

24 2006 Recalibration: NCHRP 1-40D1 Calibration process Preparation of expanded database (huge effort) Improvement of models and algorithms & software modification Calibration process, models Preparation of final calibrated software

25 2006 Recalibration: NCHRP 1-40D1 Calibration process documentation NCHRP Research Results Digest 308: >300 software and engineering improvements Version Many additional improvements for Version (Feb 2007)

26 JPCP Cracking Model Predicted percent slabs cracked R 2 = 0.86 SEE = 5.02 percent N = Measured percent slabs cracked 1 Cracking(%) = C 1+ C FD 1 2 Model Coefficients 1-37A: C1 = 1.0 ; C2 = D: C1 = 1.0; C2 = -2.00

27 JPCP Unbonded Overlay Cracking Model 2006 Predicted percent slabs cracked R 2 = 0.72 SEE = 3.95 percent N = Measured percent slabs cracked 1 Cracking(%) = C 1+ C FD 1 2 Model Coefficients 1-37A: C1 = 1.0 ; C2 = D: C1 = 1.0; C2 = -2.00

28 JPCP CPR Cracking Model 2006 Predicted percent slabs cracked R 2 = 0.90 N = 94 SEE = 6.5 percent Measured percent slabs cracked 1 Cracking(%) = C 1+ C FD 1 2 Model Coefficients 1-37A: C1 = 1.0 ; C2 = D: C1 = 1.0; C2 = -2.00

29 JPCP Joint Faulting Model R 2 = 0.62 SEE = in N = 1260 Predicted mean transverse joint faulting, in Measured mean transverse joint faulting, in

30 JPCP CPR Faulting Model Predicted transverse joint faulting, in R 2 = 0.61 N = 40 SEE = 0.02 in Measured transverse joint faulting, in

31 CRCP Punchout Model 2006 Measured punchouts, #/mile y = x R 2 = 0.74 R SEE= 3.6 N = Predicted punchouts, #/mile PO = C1 1+ C FD 2 C 3 Model Coefficients 1-37A: C1 = 105 ; C2 = 4; C3 = D: C1 = 196; C2 = 20; C3 = -0.50

32 Comparison of Measured and Predicted JPCP IRI Predicted IRI, in /m ile N = 1148 R 2 = 0.64 SEE = 13.7 in/mi Measured IRI, in/mile

33 Percent Slabs Cracked LTPP , 6 CA Pavement Age, years Measured (12-ft) Measured (19-ft) Predicted (12-ft) Predicted (19-ft)

34 LTPP , FL Percent Slabs Cracked Pavement Age, years Measured Predicted

35 LTPP , 5 AR Mean Joint Faulting, in Pavement Age, years Measured Predicted

36 LTPP , 3013, WA Mean Joint Faulting, in Pavement Age, years Measured Predicted

37 Crack Spacing in VA Smart Road Crack Spacing, inch M-E PDG prediction - friction limit and set temperature Four year field data ATB 12 CTB 14 ATB 12 CTB 14

38 25 20 Crack Width at Steel Depth VA Smart Road ME PDG - at friction limit, set temperature ME PDG - field crack spacing Measured at surface Crack Width, mils ATB 12 CTB 14 ATB 12 CTB 14

39 Punchout Prediction for LTPP Texas 48_ Punchout per mile Measured punchout Predicted punchout Pavement age, years

40 Punchout Prediction for LTPP Connecticut 9_ Punchout per mile Measured punchout Predicted punchout Pavement age, years

41 IRI Prediction for LTPP CRCP Iowa 19_ Predicted IRI Measured IRI 175 IRI, in/mi Pavement age, years

42 Comparison 1-37A 1 with 1-40D1 Distress Cracking Faulting Punchouts 1-37A N, R D

43 Local Calibration Potential All models can be adjusted (Tools, Calibration, Coefs.) Key effect: Eliminate bias of prediction (significant over prediction or under prediction of distress). Possible effect: Reduce residual of prediction (depends on quality of data). Predicted mean transverse joint faulting, in R 2 = 0.74 SEE = in N = Measured mean transverse joint faulting, in

44 Conclusions The design reliability methodology used in the ME-PDG is radically different than that used in the version. Overcomes the major problem that exists with the old AASHTO version for heavy traffic where the AASHTO model must be entered with extremely large ESALs which extrapolates far, far, far beyond the accuracy of the model. The result is extreme conservatism in thickness which results in large over design (and excessive thickness often doesn t extend life).

45 Conclusions Reliability design methodology in ME-PDG is practical to use (not complicated). But, requires consideration of more than thickness. JPCP: thickness, dowel diameter, & joint spacing must be considered together. Other factors may be varied as well. CRCP: thickness and reinforcement content must be considered together. Other factors may be varied as well. Methodology appears to produce reasonable results over range of reliability

46 Reliability Level Impact for JPCP Project Design Reliability Vs Slab Thickness Slab Thickness, in Design Reliability

47 Reliability Level Impact for JPCP Project Design Reliability Vs Dowel Diameter 1.9 Dowel Dia., in Design Reliability

48 Conclusions Concepts applied uniformly between pavement types and overlays and thus should produce equitable designs. The value of Reliability (e.g., 90% or 95%) should not be considered as exact probability, but relative probability only.

49 Conclusions Selection of Design Reliability values for designs of low to heavy traffic requires local consideration and should be a policy issue. It should be based on consequences of success or failure of the pavement to perform as designed & on cost implications. Do NOT use the old AASHTO recommendations for the new ME-PDG reliability.

50 Conclusions At the end of the day, the prediction models gain validity for: predicting in situ pavement performance, & showing reasonable practical sensitivity (see CA, TX, AR, IA, MN, & KS studies). This validation activity is far larger and more comprehensive than any other ever conducted to validate predictive engineering models. This widespread validation adds to improvement in design reliability.

51 Thank You! Any Questions?