160 likes | 303 Views
C13 LTMS discussion Follow-up to San Antonio O&H meeting. Elisa Santos November 29 th , 2005. Objective. Reevaluate the decision made during the O&H meeting The group decided to use targets and standard deviations based on 29 tests (matrix plus concurrent tests) without taking into account
E N D
C13 LTMS discussion Follow-up to San Antonio O&H meeting Elisa Santos November 29th, 2005
Objective • Reevaluate the decision made during the O&H meeting • The group decided to use targets and standard deviations based on 29 tests (matrix plus concurrent tests) without taking into account • that Lab F is different from all the other labs in Delta OC • and that Lab B is different from the other labs in TLC • The immediate consequence of using inflated standard deviations is that all stands will calibrate • Also, the control chart bounds generated for the chosen reference oil will be affected by this decision, making it harder to detect Lab shifts
Data source • 29 tests from the ltms file • 23 matrix tests for PC10 oils • 3 tests for PC10 G • 3 concurrent tests • Critical parameters: • Delta OC • TGC • TLC • R2TCA (not included)
Delta OC Excluding Lab F • Transformation: square root • Oil discrimination • Labs are similar • Final model: Lab & Oil Type • RMSE = 0.59 Including Lab F • Transformation: square root • Oil discrimination • Lab F is different from all Labs • Final model: Lab & Oil Type • RMSE = 0.57
Means, Standard Deviations, LSMEANS & Oil discriminationBefore and After removing Lab F Before RMSE = 0.57 After RMSE = 0.59
Options for Delta OC • Summary for potential reference oils • PC10 B: • Original RMSE: 0.57 • Original STDEV: 0.73; Eliminating Lab F: 0.32 • PC10 E: • Original RMSE: 0.57 • Original STDEV: 1.43; Eliminating Lab F: 0.76 • The other oils have only three or two tests associated to them • Option1: Use LSMEANS and Unique RMSE for Delta (using data from all labs) • Option2: Use Means and Standard deviations after removing Lab F from the calculations • Option 3: Use Means and Standard deviations (using data from all labs)
Outlier screened TGC • Final model: Lab & Oil • RMSE = 6.47 • No Oil discrimination • Lab G different from Lab A LSMEANS for Oil LSMEANS for Lab Cylinder Effect
OTGC RMSE = 6.47 Lab G different from Lab A
Options for OTGC • Option1: Use LSMEANS and Unique RMSE for OTGC (using data from all labs) • Option 3: Use Means and Standard deviations (using data from all labs)
scrnd TLC • Final model: Lab & Oil Type • RMSE= 4.69 • PC10 F is different from all oils except for PC10 C & PC10 G • Lab B different from Lab A and Lab G. • The severity of Lab B is more clear with the Tech & Base Oil analysis. LSMEANS for Oil LSMEANS for Lab
scrnd TLC RMSE = 4.69
Means and Standard Deviations before and after removing Lab B Before RMSE = 4.69 After RMSE = 4.5
Options for scrnd TLC • Option1: Use LSMEANS and Unique RMSE for scrnd TLC (using data from all labs) • Option2: Use Means and Standard deviations after removing Lab B from the calculations • Option 3: Use Means and Standard deviations (using data from all labs)
Impact on calibration • Options considered for the three parameters: • Shewhart calibration 111 (Option 1 for Delta/ Option 1 for OTGC/ Option 1 for scrnd TLC) • Shewhart calibration 212 (Option 2 for Delta/ Option 1 for OTGC/ Option 2 for scrnd TLC) • Shewhart calibration 333 (Option 3 for Delta/ Option 3 for OTGC/ Option 3 for scrnd TLC) • … there are other combinations, but these are the ones I generated • Look at EXCEL spreadsheet ..\Excel files\Discussion about Targets and precision.xls • Shewhart calculations by parameter • EWMA for severity adjustments (?)
Summary of Impact on Lab/ Stand calibrationRow Color represents oil type A Shewhart Cal row different from zero indicates that that particular Yi is falling out of the K bounds for at least one parameter