Caterpillar C13 Matrix Data Analysis Elisa Santos Elisa.Santos@Infineum.com Caterpillar C13 Matrix Data Analysis Discussed at meeting on October 20th, 2005 Participants: Jim Rutherford, Elisa Santos, Phil Scinto and John Zalar Participants in part: Jeff Clark and Todd Dvorak “The industry statisticians reached consensus on analyses of the PC-10 Precision Matrices. We agreed that we have more work to do, more details to examine, more questions to address, etc. However, we don't expect the basic analyses to change substantially from what we have today and we are ready to share with the industry.”

Outline Summary Data Source Modeling Analysis by parameter Delta Oil Consumption: Pages 8 - 13 Deposits Pages 15 - 34 Correlations Page 35 - 42 Precision Page 43 Appendices Plots by parameter Pages 44 to 56 Precision for the 1P … Precision by cylinder for the C13 Summary statistics for 32 tests Summary statistics for 24 tests Follow up to the BOI meeting

Statistical evidence that Lab F is severe on Delta OC Summary (1) Statistical evidence that Lab F is severe on Delta OC Analysis with 32 tests shows that Lab A is mild for Delta OC Lab B is severe for TLC and TLHC Additional Lab differences UWD: Lab A & Lab B; Lab A & Lab G; Some indication of Lab B severity TGC: Lab A & Lab G TGF: Lab A & Lab F ; Lab A & Lab G

Summary (2) Impact of Base Oil on Delta OC seems to vary with Technology Delta OC increases with Base Oil (1,2,3) for Technology B And there are no significant differences among Base Oils for Technology A In general, Deposits for Base Oil 3 are higher compared to Base Oil 2 and Base Oil 1 Correlation of Delta OC with Deposits is very weak: ~ 0.4 or lower, most of them not significantly different from zero Precision: E p is greater than 1 for TLC and TLHC ~ 0.85 for TGC ~ 0.65 for Delta OC and TGF

Data Source TMC file with 32 tests; 26 valid matrix tests; 5 valid mini matrix tests; 1 extra test 24 PC10 oils + 2 Oil A + 3 Oil D + 3 PG10 G Test 55017 was eliminated from the analysis because was operationally invalid. Test 55739 was aborted. Observation about Test 55740: the intermediate (or second) ring was stuck 24 tests for PC10 oils

Modeling 24 PC10 Oils: The model used for the analysis includes Lab, Stand(Lab), Technology Type and Basel Oil Type PC10 Oils + mini matrix + PC10G: The model used for the analysis includes Lab, Stand(Lab) and Oil Type Transformations were used when deemed necessary to satisfy the assumptions of the model and to allow for performing valid tests of hypothesis. The tests are corrected for multiple comparisons With respect to the plots with confidence intervals: if the confidence intervals overlap then there are no significant differences between Labs (or Oils). Precision is the residual standard error of the final model for each parameter. The estimates are given in their original scale.

Delta Oil Consumption

Delta OC: 24 tests Lab differences: Statistical evidence that Lab F is severe relative to other labs in the matrix No differences between Stands inside Labs Final Model: Lab, Technology, Base Oil and interaction of Technology & Base Oil Observations about the impact of Base Oil on Delta OC Delta OC increases when moving from Base Oil 1 to Base Oil 2 to Base Oil 3 for Technology B.  There is statistical evidence that Delta OC for Base Oil 3 is larger than the other Base Oils. Delta OC decreases when moving from Base Oil 1 to Base Oil 2 to Base Oil 3 for Technology A.  Note, however, that there is not enough statistical evidence to conclude that the Base Oils are different. For instance, the difference between A1 and A3 is ~20, and the 95% conf interval for A1 - A3 is [ -1.68; 41.61]. For the difference to be statistically significant this interval must not include zero.

Details for Delta OC: 24 tests The impact of Base Oil on Delta OC seems to vary with Technology Levels not connected by same letter are significantly different

Details for Delta OC: 24 tests No transformation used Rsquare adj: 81% Precision: 6.5 Statistical evidence that Lab F is severe relative to the other labs Levels not connected by same letter are significantly different

LN Delta OC: 32 tests Ln transformation was used Final model: Lab and Oil Type Rsquare adj: 79% Precision: 6.82 Statistical evidence that Lab F is severe Statistical evidence that Lab A is milder than Lab G Evidence that Lab A is milder than the other labs Levels not connected by same letter are significantly different

Details for LN Delta OC: 32 tests Oil discrimination at 5% level The conclusions are equivalent to the previous analysis based on 24 tests, Technology and Base Oil

Details for LN Delta OC: 32 tests 95% conf Details for LN Delta OC: 32 tests 95% conf. limits to help visualize Oil differences

Deposits

Differences among cylinders by parameter

Outlier screened UWD (OUWD): 24 tests taking into account differences among cylinders Similar results are obtained for the analysis before screening for outliers No transformation used Final Model: Lab and Base Oil; Rsquare adj: 66% Precision: 8.02 No significant differences among Stands or Technologies Lab discrimination: Lab A & Lab B; Lab A & Lab G Some indication that Lab B may be severe compared to the other labs. Differences not significant at 5% Levels not connected by same letter are significantly different

Outlier screened UWD (OUWD): 24 tests taking into account differences among cylinders Impact of Base Oil on OUWD Higher values of OUWD correspond to Base Oil 3 Levels not connected by same letter are significantly different

Similar results are obtained for the analysis before screening for outliers Outlier screened UWD: 32 tests taking into account differences among cylinders No transformation used Final model: Lab and Oil type; Rsquare adj: 56% Precision: 8.38 No significant differences among Stands inside Labs Lab discrimination: Lab A & Lab B; Lab A & Lab G; Lab B & Lab D Levels not connected by same letter are significantly different

OUWD: 32 tests taking into account differences among cylinders Levels not connected by same letter are significantly different

Outlier screened TGC (OTGC): 24 tests taking into account differences among cylinders No transformation used Final model: Lab and Base Oil; Rsquare adj: 56% Precision: 5.57 No significant differences among Stands or Technologies Lab discrimination: Lab A & Lab G Similar results are obtained for the analysis before screening for outliers Levels not connected by same letter are significantly different

Outlier screened TGC (OTGC): 24 tests taking into account differences among cylinders Impact of Base Oil on OTGC Higher values of OTGC correspond to Base Oil 3; Difference between Base oil 1 and Base oil 3 is borderline significant

Outlier screened TGC (OTGC): 32 tests taking into account differences among cylinders No transformation used Final Model: Lab and Oil Type; Rsquare adj: 55% Precision: 5.46 No significant differences among Stands inside Labs Lab discrimination: Lab A & Lab G Similar results are obtained for the analysis before screening for outliers Levels not connected by same letter are significantly different

Outlier screened TGC (OTGC):32 tests taking into account differences among cylinders No Oil differences; borderline discrimination Oil D & PC10 F

Outlier screened TLC (scrnd TLC): 24 tests No transformation used Final Model: Lab, Technology, Base Oil and interaction of Technology & Base Oil Rsquare adj: 63% Precision: 4.02 No significant differences among Stands Statistical evidence that Lab B is severe Similar results are obtained for the analysis before screening for outliers

Outlier screened TLC (scrnd TLC): 24 tests Impact of Base Oil on scrnd TLC depends upon Technology Technology B/Base oil 3 seems significantly higher than the other Technology/ Base Oil combinations, except for Technology A/Base oil 3 Levels not connected by same letter are significantly different

Outlier screened TLC (scrnd TLC): 32 tests No transformation used Final model: Lab and Oil Type Rsquare adj: 52% Precision: 4.25 No significant differences among Stands inside Labs Statistical evidence that Lab B is severe Similar results are obtained for the analysis before screening for outliers

Outlier screened TLC (scrnd TLC): 32 tests Oil discrimination: PC10 F & all the other oils, except for oil A & PC10 C Oil A & PC10 A

Outlier screened TGF (OTGF): 24 tests taking into account differences among cylinders No transformation used Final Model: Lab; Rsquare adj: 47% Precision: 7.22 No significant differences among Stands, Technologies or Base Oil Lab discrimination: Lab A & Lab F; Lab A & Lab G; Similar results are obtained for the analysis before screening for outliers Levels not connected by same letter are significantly different

Outlier screened TGF (OTGF): 32 tests taking into account differences among cylinders Similar results are obtained for the analysis before screening for outliers No transformation used Final model: Lab and Oil Type; Rsquare adj: 47.5% Precision: 6.96 No significant differences among Stands or Oil types Lab discrimination: Lab A & Lab F; Lab A & Lab G

Outlier screened TGF (OTGF): 32 tests taking into account differences among cylinders No Oil Discrimination

Outlier screened TLHC (scrnd TLHC): 24 tests No transformation used Final Model: Lab, Technology, Base Oil and interaction of Technology & Base Oil Rsquare adj: 80% Precision: 3.05 No significant differences among Stands Statistical evidence that Lab B is severe Similar results are obtained for the analysis before screening for outliers

Outlier screened TLHC (scrnd TLHC):24 tests Impact of Base Oil on scrnd TLHC depends upon Technology Technology B/Base oil 3 seems significantly higher than the other Technology/ Base Oil combinations Levels not connected by same letter are significantly different

Outlier screened TLHC (scrnd TLHC):32 tests No transformation used Final Model: Lab, Oil Type; Rsquare adj: 69% Precision: 3.45 No significant differences among Stands Some indication that Lab B is severe

scrnd TLHC:32 tests Oil discrimination:

Correlations: 24 tests taking into account the final model for each parameter

Pairwise Correlations 24 tests: taking into account the final model for each parameter

Correlations 24 tests: raw data

Pairwise Correlations 24 tests: raw data

Correlations 32 tests: taking into account Lab and Oil Type

Pairwise Correlations 32 tests: taking into account Lab and Oil Type

Correlations 32 tests: raw data

Pairwise Correlations 32 tests: raw data

Precision based on the model Desirable values for E p are greater than 1 E p is greater than 1 for TLC and TLHC   Precision based on the model Median of MAD survey E p1 E p2 Parameter 24 tests 32 tests Delta OC 6.5 6.82 4.5 0.6923 0.6598 OUWD 8.15 8.5 OTGC 5.85 5.74 5 0.8547 0.8711 OTGF 7.22 6.96 0.6233 0.6466 scrnd TLC 4.02 4.25 1.1194 1.0588 scrnd TLHC 3.05 3.45 4 1.3115 1.1594 MAD survey indicates the maximum acceptable difference between two test results on the same formulation

Appendix 1 Plots of the performance measures by Oil

Appendix 2: Precision for 1P ASTM TMC requirements for Engine Test Stand/Lab Calibration (Page 12-1) Calculated from all data: 103 tests: Chart = Yes WD: There seems to be Lab differences for WD OC-g/h: There seems to be Lab differences and Oil differences

Appendix 3: C13 Precision by cylinder based on 24 tests The numbers below may be compared to the precision for the 1P TGC: TGC1=10.36 TGC2= 9.04 TGC3= 9.95 TGC4= 7.73 TGC5= 9.21 TGC6= 7.7 TLC: TLC1=10.24 TLC2=9.23 TLC3= 7.47 TLC4= 8.9 TLC5=8.71 TLC6= 9.16

Appendix 4: Summary statistics - 32 tests

Appendix 5: Summary statistics - 24 tests Area under construction Appendix 5: Summary statistics - 24 tests

Appendix 6: Additional plots Follow up to the BOI meeting

Delta OC versus Base Oil

Parameter versus Tech/Base Oil Combination

OTGC versus Base Oil

scrnd TLC versus Base Oil

OTGF versus Base Oil

OUWD versus Base Oil

Base Oil Effect Summary from the BOI presentation (10/21/05) Prmter Tchnlgy Base Oil Effect Observed Statistically Significant? OC A Higher Sats/BOVI=Lower OC No B Higher Sats/BOVI=Higher OC Group III UWD A & B Group III=Higher UWD Yes TLC Higher Sats/BOVI=Higher TLC Group III=Higher TLC TLHC Higher Sats/BOVI=Higher TLHC Group III=Higher TLHC TGF NONE NA TGC Group III=Higher TGC