NSWC Corona-MS Interval DJ June 2002

1
Dr. Dennis Jackson 909-273-4492 DSN
933-4492 JacksonDH_at_Corona.Navy.Mil
1
NSWC Corona-MS Interval DJ June 2002
2
CALIBRATION INTERVAL ANALYSIS CURRENT AND FUTURE
Dr. Dennis Jackson MS30A1 June 2002
3
Overview

• Current Calibration Interval Methods
• Interval Analysis Results
• New Approaches to Calibration Interval Estimation

4
Current MethodsWhat Is a Calibration?

• Compare the measurement values from a UUT with
  the measurement values from a calibrator.
• Deviation UUT Measurement Calibrator
  Measurement
• A UUT is considered in tolerance if
• Lower Tolerance lt Deviation lt Upper Tolerance
• Measurement Reliability is the probability of
  being in tolerance.
• A Calibration Interval is the amount of time
  between calibrations that will meet a measurement
  reliability target (keeps the UUT in tolerance).

5
Current MethodsCalibration Interval
Determination
72 EOP Reliability for GPTE 85 EOP Reliability
for Safety-of-Flight and Mission Critical
6
Current Methods Stages of the Calibration
Interval Process
7
Interval Analysis ResultsNAVSEA Interval Changes
INTERVAL ACTION COUNT
IN PROCESS 148
INITIAL INTERVALS 332
EXTENSIONS 113
DECREASES 24
NO CHANGE 361
TOTAL 978
(FY 2002 through April 2002)
8
Interval Analysis ResultsAnnual Calibration
Cost Avoidance
NAVSEA NAVY
EXTENSIONS 153K 1918 (M/H) 372K 4644 (M/H)
DECREASES -40K -495 (M/H) -60K -749 (M/H)
COST AVOIDANCE 113K 1423 (M/H) 312K 3895 (M/H)
(Based on changes made in FY 2002 Through April
2002)
9
New Approaches to Calibration Interval Estimation

• Near Term - Binomial Calibration Interval
  Estimation Methods
• More accurate interval estimates
• Alternative reliability models
• Visual analysis methods
• Long Term - Variables Data Calibration Interval
  Estimation Methods
• Fixes data problems
• More information on measurement characteristics
• Less data required
• MEASURE 2 capability with automated data

10
Traditional Reliability Methods
Assumptions You know when the failure
occurs. R 1.0 at time 0. Data Failure
Times.
Exponential Model R exp(-?t)
11
Tolerance Testing Data

• Characteristics
• The failure occurs during an interval.
• R lt 1.0 at time 0.

Note The points on this graph are observed in
tolerance proportions.
12
Using Traditional Methods On Tolerance Testing
Data
Problem The estimates dont match the data
because the intercept must go through 1.0.
13
Reliability Methods For Tolerance Testing Data
Assumptions The failure occurs during an
interval. R lt 1.0 at time 0. Data Success/F
ailure (Binomial)
Intercept Exponential Model R Ro exp(-?t)
exp(?0 ?1t)
14
Current Status of Near Term Efforts

• 2002 MSC Paper Calibration Intervals New
  Models and Techniques
• Binomial Analysis, New Models, Reliability
  Intercepts, Initial Variables Methods
• Binomial Calibration Interval Analysis System

15
Benefits of Binomial Calibration Interval
Estimation Methods

• The use of Binomial estimation methods provides
  more accurate calibration interval estimates
  based on current statistical estimation theory.
• Binomial estimation methods allow for alternative
  measurement reliability models, including
  intercept and multivariable models.
• Better graphical tools provide more understanding
  of test equipment behavior.

16
Long Term Approach Variables Calibration Data
17
Calibration Intervals Based on Variables Data

• Compute a Drift Trend.
• Compute a Variability Trend using residuals from
  the drift trend.
• Obtain a Reliability Curve using the drift and
  variability trends.
• Determine the Calibration Interval from the
  reliability curve.
• Predict the Measurement Uncertainty using the
  drift and variability trends.

18
Drift Trend Analysis

• E(d) B0 B1 t (Weighted Linear Regression on d)

19
Variability Trend Analysis

• E(res2) C0 C1 t (Linear Regression on res2)

20
A Basis for Increasing Variability
Generally, a single serial number does not show
increasing variability
21
A Basis for Increasing Variability
However, several serial numbers could have
slightly different slopes and intercepts
22
A Basis for Increasing Variability
The overall effect is one of increasing
variability for the population
23
Reliability Curve Analysis
24
Determining Calibration Intervals From Variables
Data
25
Current Statusof Long Term Efforts

• 2002 MSC Paper Calibration Intervals New
  Models and Techniques
• Binomial Analysis, New Models, Reliability
  Intercepts, Initial Variables Methods
• 2003 MSC Paper Calibration Intervals and
  Measurement Uncertainty Based on Variables Data
• NPSL, SCE
• Variables Analysis Excel Tool
• Estimates Trends, Calibration Intervals,
  Measurement Uncertainty
• MEASURE 2
• Automated/Electronic data

26
Benefits of UsingVariables Data

• MEASURE data is often suspect
• In-Tolerance data is difficult to verify
  (success/failure)
• Engineering review required for nearly all
  calibration interval determinations
• Variables data is more trustworthy
• This could significantly increase the number of
  interval analyses
• Variables data provides much more information
• Requires fewer calibrations to accurately
  determine a calibration interval than
  In-Tolerance data
• Development of automated/electronic data
  recording could reduce calibration time.

27
Summary

• Calibration intervals minimize the amount of
  calibration effort required to keep test
  equipment adequately in tolerance.
• Recent adjustments to calibration intervals will
  result in significant cost avoidance.
• Near-term improvements using Binomial methods
  will provide better visual analysis and more
  accurate estimation techniques.
• Long-term improvements using variables data
  methods will
• Fix data problems
• Provide faster analyses with less data
• Possibly reduce administrative part of
  calibration time