1. 1 Structural Responses – Preliminary Results and Observations PEER GMSM Program Workshop, Richmond CA, October 29, 2007 Curt B. Haselton, PhD, PE Assistant Professor California State University, Chico And other members of PEER GMSM Working Group

2. 2 Our Purpose for the Next Two Hours  The purpose of convening everyone to review the preliminary results is:  To ensure that we correctly interpret the various methods.  To obtain feedback and insights from the group.  Our approach to interpreting the results  Our specific observations  Other things that we may need to think about or address  To begin the process of coming to a collective consensus. The product of this study will be more useful to the ground motion community if we can all establish a group agreement on the results.  Note: Our dialogue will be ongoing, but this is our only formal meeting before the preparation of the first PEER report.

3. 3 Outline  Introduction to the results figures and the available results files  Brief overview of preliminary results (15 minute overview before starting detailed discussion)  Detailed review of preliminary results and observations (1.5 hours of review and group discussion)

4. 4 Introduction to Results Figures The point of comparison (POC) is used to compare predictions to, as discussed previously. The counted median response includes collapsed records (assuming they have a very large MIDR). A blue point at the top of the figure indicates collapsed record(s). The number indicates the ratio of collapsed records to total records in the set. The method tags correspond to the numbers in the index Excel file. There is one entry per ground motion set (typically four per method) Each blue data point is the response from a single scaled ground motion record. Numbers next to the “+” indicate that the sets are ranked, with #1 being better than #4.

5. 5  To facilitate your detailed review, the following files are available (on GMSM website, and available USB drives):  Index of methods (provides method names/numbers)  Acceleration spectra for each method and ground motion set Materials Available to Facilitate Detailed Review

6. 6  To facilitate your detailed review, the following files are available (on GMSM website, and available USB drives):  Index of methods (provides method names/numbers)  Acceleration spectra for each method and ground motion set  Results for Buildings B, C, and D  PowerPoint overview of results (should not need today) Materials Available to Facilitate Detailed Review

7. 7  To facilitate your detailed review, the following files are available (on GMSM website, and available USB drives):  Index of methods (provides method names/numbers)  Acceleration spectra for each method and ground motion set  Results for Buildings B, C, and D  PowerPoint overview of results (should not need today)  Excel file giving median predictions and comparison to POC  Excel files giving detailed results for each method Materials Available to Facilitate Detailed Review

8. 8 Overview of Structural Response Results  The overall study includes:  Four buildings (4-story frame, 12-story frame and wall, 20-story frame)  Four objectives (response conditioned on M, R and response conditioned on M, R, and a +2 σ Sa value)  We will focus on the results for:  Building C (20-story RC frame), with comparisons to Building D (12-story RC wall) and Building A (4-story RC frame)  Objective 4 (predicting median MIDR for a +2 σ motion).  Review results by classification:  Sa(T 1 ) methods  Methods based on matching the uniform hazard spectrum (UHS)  Methods based on matching the conditional mean spectrum (CMS)  Methods using a proxy for spectral shape (e.g. ε)  Inelastic response based methods

9. 9 Overview: Sa(T 1 ) Scaling Median = 0.028 POC = 0.019 High This is the coefficient of variation of the median predictions.

10. 10 Overview: Matching the UHS Figure 1 of 3 Median = 0.024 POC = 0.019

11. 11 Overview: Matching the UHS Figure 2 of 3 Median = 0.024 POC = 0.019

12. 12 Overview: Matching the UHS Figure 3 of 3 Median = 0.024 POC = 0.019

13. 13 Overview: Matching Conditional Mean Spectrum Median = 0.019 POC = 0.019

14. 14 Overview: Matching Conditional Mean Spectrum Median = 0.019 POC = 0.019 Matching UHS Matching CMS Matching UHS Matching CMS

15. 15 Overview: Proxy for Spectral Shape (e.g. ε) Median = 0.026 POC = 0.019 With 57-58 removedAll methodsCMS-matching This 15% higher prediction is also associate with a higher collapse rate. CMS (and POC): 0/28 Proxy: 6/28

16. 16 Overview: Inelastic-Based Methods AAAaAAAa Median = 0.022 Higher prediction again associated with higher collapse rate. CMS methods (and POC): 0/28 Inelastic methods: 3/28 POC = 0.019

17. 17 Overview of Structural Response Results  Results summary by method classification for Building C (median MIDR/POC):  Sa(T 1 ) and UHS methods results in highly variable predictions which tend to be larger than the POC.  Methods that match the CMS agree well with the POC.  CMS Proxy methods and Inelastic methods:  On average, results in response 15% higher than the POC.  This seems to also come with a higher collapse rate.  **These observations are an average or all methods in the class; some of the individual methods have predictions nearly the same as the POC.

18. 18 Overview of Structural Response Results  Comparison of results for Buildings A and D  Sa(T 1 ) methods: Tend to over-predict response, predictions depend on spectra shape and contributions of inelastic and higher-mode responses.  Matching CMS: Predictions close to POC in all cases.  Proxy methods: Consistent at 10% above the POC.  Inelastic methods: 15% above the POC for frame buildings (C and A), and right at POC for shear wall (Building D).

19. 19 Outline  Introduction to the results figures and the available results files  Brief overview of preliminary results (15 minute overview before starting detailed discussion)  Detailed review of preliminary results and observations (1.5 hours of review and group discussion)

20. 20 Detailed Review: Sa(T 1 ) Scaling Overall Median = 0.028 POC = 0.019 Sa(T 1 ) scaling (#4, 53) Median = 0.037 Median = 0.022 ATC-58 30% draft Method (#67)

21. 21 Detailed Review: Sa(T 1 ) Scaling (#4, 53) Median = 0.037 POC = 0.019 Median MIDR / POC: - Building C (T 1 = 2.63s): 1.94 Sa(T 1 ) scaling (#4, 53) Bin: [0 < R < 20km, 6.75 < M < 7.25] Records in bin tend to have an average far-field spectral shape (higher frequency content at shorter periods).

22. 22 Detailed Review: Sa(T 1 ) Scaling (#4, 53) Median MIDR / POC: - Building C (T 1 = 2.63s): 1.94 Possible higher modes

23. 23 Detailed Review: Sa(T 1 ) Scaling (#4, 53) Median MIDR / POC: - Building C (T 1 = 2.63s): 1.94 - Building D (T 1 = 1.2s): 1.13 Spectrum lowered due to scaling at 1.2s rather than 2.63s. Spectrum matches UHS at T 1, but overall spectrum is 40-45% lower.

24. 24 Detailed Review: Sa(T 1 ) Scaling (#4, 53)  Observations/Comments:  Building C response is over-predicted 70% more than Building D, and Building A over-prediction is 13% higher than D.  Comparing C (20-story frame; T 1 = 2.63s) and D (12-story wall ; T 1 = 1.2s):  Scaling period and spectral shape lead to a 40-45% difference in the overall position of the spectrum.  For Building C, the spectrum falls off a bit more slowly at T > T 1, causing higher inelastic response.  Higher mode effects may also increase Building C response.  Comparing D and A:  Similar fundamental periods, but Building D has less structural damage, so period elongation is less drastic. Median MIDR / POC: - Building C (T 1 = 2.63s): 1.94 - Building D (T 1 = 1.2s): 1.13 - Building A (T 1 = 1.0s): 1.28

25. 25 Detailed Review: Sa(T 1 ) Scaling (#67) Median = 0.022 POC = 0.019 ATC-58 30% draft Method (#67) [0 6.5] Spectral shape differs from the previous methods – pulsy motions with more spectral content from 0.5-3.5 sec Median MIDR / POC: - Building C (T 1 = 2.63s): 1.17

26. 26 Detailed Review: Sa(T 1 ) Scaling (#67) Median MIDR / POC: - Building C (T 1 = 2.63s): 1.17

27. 27 Detailed Review: Sa(T 1 ) Scaling (#67) Median MIDR / POC: - Building C (T 1 = 2.63s): 1.17 - Building D (T 1 = 1.2s): 1.39 - Building A: (T 1 = 1.0s): 1.48 Spectral shape is nearly the same as the UHS (for 0.5-3.5s). Therefore spectral values are nearly unchanged when scaled to either Sa(1.2s) or Sa(2.63s). However, spectral values at T > T 1 are much greater for T 1 = 1.0-1.2s. Building A has higher response than D due to higher levels of damage.

28. 28 Detailed Review: Sa(T 1 ) Scaling (#67)  Summary for Sa(T 1 ):  MIDR/POC ranges from 1.11 to 1.94 for the various methods and Buildings.  With Sa(T 1 ) scaling, we scale all of the spectra to the UHS at a given period. This causes predictions to be highly variable and depend on:  The spectral shape of the record bin, as compared to the UHS.  The period used for scaling.  These items affect the shape of the spectrum away from T 1, which affect inelastic response.

29. 29 A Detailed Look at Results of Building Code-Based Record Selection PEER GMSM Program Workshop, Richmond CA, October 29, 2007 Jack W. Baker, PhD Assistant Professor Stanford University And other members of PEER GMSM Working Group

30. 30 IBC (ASCE 7-05) building code requirements From ASCE 7-05, section 16.1.3.1: “Each ground motion shall consist of a horizontal acceleration history, selected from an actual recorded event” “obtained from records of events having magnitudes, fault distance, and source mechanisms that are consistent with those that control the maximum considered earthquake.” “The ground motions shall be scaled such that the average value of the 5 percent damped response spectra for the suite of motions is not less than the design response spectrum for the site for periods ranging from 0.2T to 1.5T “

31. 31 Number of ground motions From ASCE 7-05, section 16.1.4: “If at least seven ground motions are analyzed, the design member forces … and the design story drift … is permitted to be taken respectively as the average of the … values determined from the analyses”

32. 32 Site-specific design spectrum From ASCE 7-05, section 21.2.1:  “The probabilistic MCE spectral response accelerations shall be taken as the spectral response accelerations represented by a 5 percent damped acceleration response spectrum having a 2 percent probability of exceedance within a 50-yr. period.”  Here we use the +2 σ response spectrum at all periods, to facilitate comparison and since that has a ~2% probability of exceedance, given the scenario magnitude and distance.  The 150%-of-median deterministic cap is ignored here to allow comparison with other results.

33. 33 Selection approach  Start with the NGA ground motion library (7038 horizontal components)  Eliminate records not meeting specified criteria (e.g., magnitude and distance ranges)  Select the 28 records that most closely matched the target spectra after scaling  If the average of the 28 spectra fell significantly below the target spectrum, perform some additional minor scaling

34. 34 Case 1 (Method Tag 9980)  No M/R/Mech. Restrictions  No filter frequency restriction  7038 records available 0.2 T 1 1.5 T 1

35. 35 Case 2 (Method Tag 9981)  No M/R/Mech. Restrictions  Restricted filter frequencies  3454 records available (50%)

36. 36 Case 3 (Method Tag 9982)  6.5 < M < 7.6  No Dist./Mech. Restrictions  1122 records available (16%) Set #4

37. 37 Case 4 (Method Tag 9983)  0 < R < 30 km  No Mag./Mech. Restrictions  856 records available (12%)

38. 38 Case 5 (Method Tag 9984)  Strike slip events only  No Mag./Dist. Restrictions  978 records available (14%)

39. 39 Case 6 (Method Tag 9985)  6.5 < M < 7.6,  0 < R < 30 km  Strike slip events only  Target spectrum not always exceeded  132 records available (2%)

40. 40 Case 7 (Method Tag 9986)  6.5 < M < 7.6  0 < R < 30 km  Strike slip events only  Target spectrum exceeded  132 records available (2%)

41. 41 Case 8 (Method Tag 9975)  6.5 < M < 7.6  0 < R < 30 km  Strike slip events only  Max scale factor = 4  132 records available (2%)

42. 42 Case 9 (Method Tag 9976) Set #4  6.5 < M < 7.6  0 < R < 30 km  Strike slip events only  Max scale factor = 2  132 records available (2%)

43. 43 Case 10 (Method Tag 9989)  6.5 < M < 7.6  0 < R < 30 km  Strike slip events only  Max one record per event  9 records available (0.1%)

44. 44 Observations  Are the magnitude/distance/mechanism restrictions needed?  We know they affect spectral shape, but we are already specifying a target spectral shape Median response spectra from events with differing magnitudes and distances

45. 45 Observations  Are scale factor restrictions needed?  They don’t seem to have an effect. Again, this may result from the target spectrum requirement  Note that no such restriction is given in the code  Is the one-record-per-event restriction needed?  It doesn’t seem to have an effect, and severely limits the available number of records  Note that no such restriction is given in the code

46. 46 Observations  Is the filter frequency limitation needed?  Presumably over-filtered motions will not match the design spectrum, so the target spectrum should ensure we have records with proper filtering (assuming that 0.2T to 1.5T are the only periods we need to worry about)  Note that no such restriction is given in the code

47. 47 Conclusions  Responses seem to be controlled by the target spectral shape, rather than the other selection criteria  This suggests that the choice of the target spectrum is a more important consideration than the choice of additional criteria  Benefit of the additional criteria: more “insurance” that you have appropriate record properties  Disadvantage of the additional criteria: a reduced number of records to chose from, meaning that you will not be able to match the target spectrum as closely

48. 48 Structural Responses – Preliminary Results and Observations PEER GMSM Program Workshop, Richmond CA, October 29, 2007 Curt B. Haselton, PhD, PE Assistant Professor California State University, Chico And other members of PEER GMSM Working Group

49. 49 Detailed Review: Matching CMS Median = 0.019 POC = 0.019

50. 50 Detailed Review: Matching CMS Median = 0.019 POC = 0.019 All methods aim to match the CMS one algorithm or another. The predictions are all close to the POC, on average. Comparison to Buildings A and D: Trends are similar.

51. 51 Detailed Review: Matching CMS Overall Median = 0.019 POC = 0.019 10: CMS with Scaling [Baker] Median = 0.0189 15: Genetic Algorithm [Alimoradi, Naeim] Median = 0.0192 24: Semi-Automated Selection and Scaling [Rathje, Kottke] Median = 0.0197 45: Design Ground Motion Library [Wang, Power, Youngs] Median = 0.0199

52. 52 Detailed Review: Matching CMS (#10, 15, 24) POC = 0.019 Methods 10, 15, and 24: Predictions agree with the POC consistently.

53. 53 Detailed Review: Matching CMS (#45) POC = 0.019 Average predictions agree well with the POC [median = 0.0199]. However, there is more scatter in the prediction. Let’s looks at the spectra to see if that explains it.

54. 54 Detailed Review: Matching CMS (#45) Record Set: 1 Median MIDR = 0.014

55. 55 Detailed Review: Matching CMS (#45) Record Set: 2 Median MIDR = 0.029

56. 56 Detailed Review: Matching CMS (#45) Record Set: 3 Median MIDR = 0.021

57. 57 Detailed Review: Matching CMS (#45) Record Set: 4 Median MIDR = 0.015 Small differences in spectra at T > T 1 seem to correlate with predictions. However, similar variations are present in spectra for method 10, and the predictions do not vary as much. Is there another reason for the differences in the Method 45 predictions, which is not shown in the acceleration spectra?

58. 58 Detailed Review: Matching CMS  Summary of Observations (same for Buildings C, A, and D):  All methods give predictions close to POC.  Methods 10, 15, and 24:  Predictions agree with POC consistently.  Method 45:  Average prediction agrees with POC.  There is more prediction variability.  Variability may be partially explained by differences in the spectra, but it looks like spectra do not full explain the differences.  Explanations?

59. 59 Detailed Review: Proxy for Spectral Shape Overall Median = 0.022 (w/o 57-58) POC = 0.019 With 57-58 removed 57-58: Assume some error in our analyses, so excluded from discussion for now.

60. 60 Detailed Review: Proxy for Spectral Shape POC = 0.019 57-58: Assume some error in our analyses, so excluded from discussion for now. 30-31: ε Selection with Sde(T 1 ) Scaling [Tothong and Luco] Median MIDR = 0.022 20: Selection based on M, R, ε, and Site Class [Skyers, Stewart, Goulet] Median MIDR = 0.021 Overall Median = 0.022 (w/o 57-58) Comparison to Buildings A and D: Trends are similar.

61. 61 Detailed Review: Proxy for Spect. Shape (#31) POC = 0.019 30-31: ε Selection with Sde(T 1 ) Scaling [Tothong and Luco] Median MIDR = 0.022 Look at spectra for method 31 (objective 4). Try to explain variability in predictions.

62. 62 Detailed Review: Proxy for Spect. Shape (#31) All 4x7 records. The spectra match the CMS well on average. Let’s look at spectra of individual sets….

63. 63 Detailed Review: Proxy for Spect. Shape (#31) Set 1

64. 64 Detailed Review: Proxy for Spect. Shape (#31) Set 2

65. 65 Detailed Review: Proxy for Spect. Shape (#31) Set 3

66. 66 Detailed Review: Proxy for Spect. Shape (#31) Set 4 Observation: All 28 records match CMS well, but individual sets of 7 have scatter. This explains scatter in predicted response.

67. 67 Detailed Review: Proxy for Spect. Shape (#20) POC = 0.019 20: Selection based on M, R, ε, and Site Class [Skyers, Stewart, Goulet] Median MIDR = 0.021 Look at spectra…

68. 68 Detailed Review: Proxy for Spect. Shape (#20) All 4x7 records. Individual sets of 7 look similar. The spectra do not match the CMS. They are higher at T T 1. Even so, the predictions agree with the POC. This is the case for this building, and also Buildings A and D.

69. 69 Detailed Review: Proxy for Spectral Shape  Summary of Observations:  Predictions are typically 10-15% higher than POC on average.  Methods 30-31 [prediction is 15% above POC]:  Spectra match CMS on average, though each set of 7 has variability.  Variability in spectra leads to variability in MIDR prediction.  Method 20 [prediction is 10% above POC]:  Spectra consistently do not match the CMS.  Even so, this gives consistent predictions that are close to the POC. This also is true for Buildings D and A.  Why good/consistent predictions when the spectral shape is distinctly different from CMS?

70. 70 Detailed Review: Inelastic-Based Methods AAAaAAAa Overall Median = 0.022 POC = 0.019 26-27: Sdi [Tothong and Luco] Median MIDR = 0.024 6: Vector of Record Properties Identified by Proxy [Watson-Lamprey] Median MIDR = 0.019 35-35: IM1I&2E [Luco and Tothong] Median MIDR = 0.022 11: Inelastic Response Surface Scaling [Shantz] Median MIDR = 0.023

71. 71 Detailed Review: Inelastic Methods (#26-27) AAAaAAAa Overall Median = 0.022 POC = 0.019 26-27: Sdi [Tothong and Luco] Median MIDR = 0.024

72. 72 Detailed Review: Inelastic Methods (#6) AAAaAAAa Overall Median = 0.022 POC = 0.019 6: Vector of Record Properties Identified by Proxy [Watson-Lamprey] Median MIDR = 0.019

73. 73 Detailed Review: Inelastic Methods (#34-35) AAAaAAAa Overall Median = 0.022 POC = 0.019 35-35: IM1I&2E [Luco and Tothong] Median MIDR = 0.022

74. 74 Detailed Review: Inelastic Methods (#11) AAAaAAAa Overall Median = 0.022 POC = 0.019 11: Inelastic Response Surface Scaling [Shantz] Median MIDR = 0.023 Sets 3-4: Includes consideration of 2 nd mode

75. 75 Detailed Review: Inelastic Methods (#11) AAAaAAAa Overall Median = 0.022 POC = 0.019 11: Inelastic Response Surface Scaling [Shantz] Median MIDR = 0.023 Sets 1-2: No consideration of 2 nd mode

76. 76 Detailed Review: Inelastic Methods  Summary of Observations:  Predictions are typically 15% higher than the POC, on average.  Not all spectra pass through Sa(T 1 ) target.  Consideration of higher mode seems to help prediction.

77. 77 Summary of Results  Comparison of results for Buildings A and D

78. 78 Discussion  To reiterate the purpose of this time:  To obtain feedback and insights from the group (preliminary results, approach, etc.).  To ensure that we correctly interpret the various methods.  To begin the process of coming to a collective consensus.  Discussion/Comments/Questions/Suggestions:  What are your thoughts on the results and possible explanations for what we have seen?  What are we missing, or what have we not bee considering that we should?  What do we need to address as we to refine and write-up the findings of this study?  Other comments, suggestions, questions, discussion?