GAUGE REPEATABILITY AND REPRODUCIBLILITY: • Linearity – assess with equipment manufacturers data and A TTRIBUTE GR&R D EFINITION :

APQP/PPAP EVIDENCE WORKBOOK STATISTICAL calibration.

RATIONALE AND USER REFERENCE GUIDE VARIABLE GAUGE R&R: Ford Customer Specific requirements for PPAP require attribute
The following image summarises both repeatability and
GR&R studies to have a minimum of 50 parts, 3 operators and 3
INTENT: As per the Ford Customer Specific Requirements for PPAP released reproducibility. trials. Deviation from this requirement needs STA approval.
in 2023, the ANOVA method for analysis of gauge R&R is now
The intent of this reference guide is to provide an overview of the The parts used for the GR&R study should be variably measured
mandatory. The ANOVA method is superior to the Average and %
statistical tools used within the APQP/PPAP Evidence Workbook using an appropriate measurement device (e.g. CMM) prior to
and the statistical methodology underpinning the sequence of S TUDY , conducting
these tools. The instructions of how to populate the APQP/PPAP % the attribute
Evidence Workbook can be found within the APQP/PPAP Evidence GR&R. The
Workbook itself. There is an instructions sheet for each of the parts to be
APQP/PPAP Evidence Workbook sheets requiring population. used for the
Additional instructions can be found in the APQP/PPAP Evidence Range Method as it includes part to operator interaction, whereas T OLERANCE , NDC D EFINITION , +/-2 T OTAL STD DEV : attribute
Workbook training. The APQP/PPAP Evidence Workbook and the Average and Range Method does not. Many statistical packages study should be distributed as per the following graphic. This will
associated training can be accessed on the Ford Supplier Portal on are available to evaluate gauge R&R using the ANOVA method, in Ford Customer Specific Requirements for PPAP released in 2023 test the gauge throughout the tolerance range and at the
the APQP/PPAP Requirements for GPDS page (APQP/PPAP addition Ford has developed an Excel based ANOVA template to aid requires both % Tolerance and % Study to be reported but the specification limits addressing the key risk of using an attribute
Requirements for GPDS (ford.com)). suppliers who do not have a package available. This can be primary metric used for acceptance of GR&R analysis is % gauge.
accessed through the Ford Supplier Portal under Miscellaneous Tolerance. Reporting of +/- 2 Total Gauge R&R Standard
GAUGE DISCRIMINATION: Deviations is also required to understand the 95% confidence The criteria
Tools (Quality System and Supplier Documents (ford.com)).
(uncertainty) of any one measurement. This value can be used in for attribute
The first piece of information that needs to Per the Ford Customer Specific Requirements for PPAP; variable
conjunction with engineering judgment to help assess the distance GR&R
be entered into the Capability Evidence GR&R studies require: 10 parts, 3 operators and 3 repeats.
between the edge(s) of the process distribution and the acceptance
sheet within the APQP/PPAP Evidence Deviation from this requirement needs STA approval.
specification limit(s). % Tolerance is mandated over % Study are:
Workbook is the gauge resolution. Note:
A CCURACY D EFINITION : Variation, as % Tolerance is • Gauge
All columns that require user input have a
light blue column heading. relatively unaffected by the parts must
Accuracy is the closeness of agreement between an observed
selected for the GR&R study, reject all parts that are outside of the specification limits.
value and an accepted reference value or standard. Low accuracy
G AUGE RESOLUTION : whereas % Study Variation and • All Kappa values should be greater than 0.75.
reflects a systemic issue in the measurement such as a gauge out Number of Distinct Categories
Gauge Resolution which can be also referred to as the least of calibration, worn etc. Accuracy is measured as the amount of (NDC) is significantly affected. For additional information refer to the Ford Customer Specific
significant digit - LSD is the increments of measure provided by the Requirements for PPAP.
error in a measurement % Study and NDC is only meaningful if the parts used for the GR&R
gauge. Per the AIAG MSA guide, a measurement instrument’s study are representative of a production run with all known sources GAUGE R&R IMPORTANT POINTS:
discrimination should be at least one-tenth of the range to be in proportion to the total
of variation as it uses the gauge variation as a ratio of the spread of
measured (tolerance). This is also known as the 10-bucket rule. For size of the measurement. 1. Gauge resolution must be acceptable before proceeding with
parts selected. Both % Study and NDC are correlated to total
example, if a length of a part is 10mm with a tolerance of +/- 0.1mm One measurement is GR&R analysis.
variation. % Tolerance is not correlated to total variation.
providing a total tolerance of 0.2mm then multiplying 0.2mm by more accurate than If the gauge does not pass % Tolerance, it cannot perform the job 2. Do not proceed past the GR&R columns in the APQP/PPAP
10% is 0.02. This means that the gauge needs to be able to measure another if it has a smaller of determining if the parts are within the specification. % Evidence Workbook if gauges have not been shown to meet the
a difference of at least 0.02mm. relative error. Tolerance, % Study and NDC, need to be reported in the acceptance criteria for either variable or attribute GR&R.
The image of the glasses provides an example of gauge resolution. APQP/PPAP Evidence Workbook. % Tolerance must meet the 3. Any measurement analysis (i.e., capability analysis) conducted
The scale on the left has a resolution of 1cm. Using this scale, we P RECISION (R EPEATABILITY ) criteria to be able to proceed through the APQP/PPAP Evidence using a gauge that has poor Gauge R&R will be incorrect.
measure the height of the glasses to be 2cm. The scale on D EFINITION : Workbook. GR&R as a percent of tolerance criteria:
the right has a resolution of STATISTICAL PROCESS CONTROL (SPC):
• <10% is acceptable
0.1cm and therefore Is the closeness of agreement between individual measurements • ≥10% and ≤30%, contact the STA site engineer to determine if Statistical process control (SPC) is a
we can measure the or results. Precision relates to the variance of repeated the GR&R is acceptable. methodology for monitoring a process to
glasses to be 1.7cm. measurements. A measuring device with a low variance is more identify special causes of variation and signal
• >30%, it is unacceptable, and the organisation shall
Better resolution precise than another having a higher variance. the need to take corrective action when it is
implement containment actions and a corrective action plan
A measurement system may be precise but not accurate at the appropriate. When special causes are
to improve measurement capability until the GR&R
same time. The image of the targets illustrates accuracy and present, the process is deemed to be out of
requirements are met.
precision. control. If the variation in the process is due
ONE - SIDED TOLERANCE : to common cause alone, the process is said
R EPRODUCIBILITY D EFINITION : to be in statistical control. SPC is a proven
The % Tolerance calculations for one-sided tolerances are technique for improving quality and productivity.
Is the variation in the same measuring device when it is used by
different operators to measure the same parts. Causes of poor different to the above double-sided tolerance calculation but the A process must be stable and in-control before process capability
reproducibility include poor operator training. acceptance percentages are as above. can be calculated. Instability affects the process in unpredictable
Upper specification limit with no lower boundary: In these cases, ways. Capability which is calculated on a process that is not stable
V ARIABLE GR&R D EFINITION : calculate % tolerance by dividing 3 gauge R&R standard deviation is not valid, because the calculations are not based on a
gives you more accurate measurements. It is critically important predictable process. Therefore, future results of the process
Gauge R&R (GR&R) studies determine how much of the observed by the
that the measurement device selected for the measurement of a cannot be predicted.
process variation is due to measurement system variation due to: difference The APQP/PPAP Evidence Workbook requires that you select the
special characteristic is fit for purpose. Often devices are selected
to measure the magnitude of characteristics and the gauge either • Measurement System Repeatability between the upper specification limit and the mean of the data. control chart that has been used to demonstrate process stability
does not have enough resolution or the gauge is difficult to use for • Measurement System Reproducibility Lower specification limit with no boundary: In these cases and control. Moreover, identify if any points were not in statistical
the specific characteristic and generates reproducibility errors GR&R does not assess: calculate the % Tolerance by dividing 3 gauge R&R standard control. Note: if any points were not in statistical control, then you
when handled by different operators. The previous graphic deviation by will not be able to proceed through the remainder of the sheet.
• Accuracy – assess with calibration Actions will need to be taken to address the cause of the out-of-
illustrates these points. the difference
• Bias – assess with calibration control conditions. Please consult your STA engineer for assistance.
• Stability – assess with equipment manufacturers data between the mean of the data and the lower specification limit.

Version 1.0 – 6 April 2023 Page 1 of 2

T YPES OF VARIATION : general rules are applicable to both variable and attribute control Control charts (SPC) require rational sampling to effectively meets a Ppk of 1.67 but the Weibull distribution shows a Ppk of 1.27
charts. differentiate between Special Cause and Common Cause variation. which is not acceptable.
Common Cause: variations are natural variations that are a part
Test 1: Point outside of the control limit: This is the most RATIONAL SUBGROUP STRATEGIES:
of the process. The type of distribution that a process follows is governed by the
recognized rule. The cause of a point outside of the 3σ control limit
Special Cause: variations are not part of the normal process At pre-production (i.e., TT build etc) an initial baseline for each
indicates a process shift. The special cause in this instance must be
variation and are due to unusual, or out-of-the-ordinary events. special characteristic needs to be established. This can be achieved
understood and addressed prior to proceeding with any further
CONTROL CHART STRUCTURE : using 25 subgroups of 5. Typically, 300 part Phase 0 PPAP run is
analysis and PPAP. This is why it is included in the APQP/PPAP
required and thus a rational sampling plan is to measure 5 parts
Control charts graphically represent process variation and are Evidence Workbook.
and skip the next 7 parts which would total 25 subgroups:
ordered by time. The Average Chart tells us about the process In a 3-sigma process that is perfectly centred, 2,700 parts per
million are expected to exceed the 3 sigma limits. This means that • Measure parts 1 to 5;
location. The Range Chart tell us about the process stability. The • Skip parts 6 to 12;
positions of the upper control limit and the lower control limit lines 1 subgroup out of 370 subgroups will exceed the 3-sigma limit
• Measure parts 13 to 17,
without an identifiable special cause. This means the chance of any
are determined by the data. The lines are calculated at +/-3 • Skip parts 18 to 24 and so on.
standard deviations away from the centre line. The control limits points outside of the control limits before launch due to common Rational sampling requirements post launch are dependent on the
are determined by the natural variation of the process. The control cause are extremely small and any such case must be assumed to process, the characteristic and historical performance. Please
limits show the range that the process can randomly fall within. be a special cause and be investigated and understood. discuss the specifics of your process with the STA Site engineer.
The average of different subgroups made in production order will The remaining general rules are useful to help problem solving physics of the process. Leveraging historical data or process
activities. DISTRIBUTION ANALYSIS: experts to determine the expected distribution is very helpful if the
have similar subgroup averages. They will not be identical as they
are averages from different parts, but there is a limit to the Test 2: Nine points in a row on the same side of the centre line. data is not following a normal distribution. Blindly fitting a
The APQP/PPAP Evidence Workbook requires
Test 3: Six points in a row either increasing or decreasing. distribution is not acceptable as the cause of non-normal data
magnitude of difference due to random chance. This is how the that the process distribution type is identified.
Test 4: Fourteen points in a row alternating up and down. maybe due to special cause. Most processes follow a normal
control limits are determined. If the process is stable, then it is The process distribution type needs to be
Test 5: Two out of three points more than 2σ from the centre line
highly probable that the points will fall within the control limits. If distribution. Below is a table that illustrates different processes
(on the same side). identified prior to the calculation of capability
the points fall outside of the control limits, then there is a very good Test 6: Four out of five points more than 1σ from the centre line and the expected process distributions.
so that the correct distribution is used for
chance that there is a special cause present in the process. (on the same side). process capability calculations. Failure to use DISTRIBUTION ANALYSIS IMPORTANT POINTS:
Test 7: Fifteen points in a row within 1σ from the centre line (either the correct distribution type will significantly
side).
affect the accuracy and validity of process
Test 8: Eight points in a row more than 1σ from the centre line
(either side). capability. Standard process capability
equations (Ppk) assume normal distributions (Bell shaped curve)
SPC IMPORTANT POINTS:
calculations may be significantly inaccurate if data is not normal.
1. A process cannot be said to be stable and in-control if both Normality tests such as p-values and probability plots are the most
charts have points outside of control limits. 1. A distribution type needs to be determined prior to proceeding
common methods used to determine what type of distribution data with any process capability calculations.
Control limits do not equal Specification limits. Specification limits 2. Parts plotted in control charts must be in production order.
3. Control limits do not equal specification limits. follows. These tests compare the shape of the calculated curves to 2. Distribution types are determined by the physics of the process.
should be based on the voice of the customer and established Consulting with process experts and using historical process
4. A process must be stable and in-control before process the actual distribution of process data points and provide a value
during the design process. These limits define the range of data is helpful to confirm the distribution type.
capability can be calculated. If the process is not statistically indicating how well the distribution fits the curve. Consider the
customer satisfaction. Control limits are based on the voice of the stable process capability estimates have no predictive value as 3. Fitting a distribution that does not match the expected
below data, both the probability plot and the p-values show that
process and reflects how the process is running. the process parameters can change, therefore, the process distribution indicates that special causes are present that need
the data does not follow a normal distribution and that in fact it to be identified.
OUT OF CONTROL INDICATIONS : capability estimates are estimates of past capability and cannot follows a Weibull distribution. A p-value that is less than 0.05
predict future capability (the primary goal). 4. Utilise Ford STA or Master Black Belt (MBB) to help when fitting
Imagine that you are driving on a highway with several lanes. This indicates that the data does not follow that specific distribution. non-normal distributions.
is your road to establishing stability for your process. The guardrails STUDY PARAMETERS:
CAPABILITY RESULTS:
on each side represent the +/-3σ control limits. The lanes allow you The Ford Customer Requirements for PPAP
Capability refers to the ability of the
to drive at different speeds. These lanes (different sigma intervals) require a minimum of 125 measurements to
process to consistently produce a
represent the slight differences in parts within your production establish process capability. The 125
product meeting customer
process caused by common cause variation. If there was an measurements are to be obtained using
expectations. Capability is based on a
obstacle on the road this would cause you to deviate away from rational sampling methods from a
set of data taken over a period and measures all the effects of the
your average trajectory into another lane and potentially outside of population of 300 parts. Note: Any deviation from this
process. Process capability calculations make no sense if:
the +/-3σ control limits (guardrails). The same could happen in a requirement requires STA management approval.
• the process is not in stable and in control.
process, if the settings on a machine change or a material type is RATIONAL S AMPLING : The probability plot shows how well the data points fall within a • the incorrect distribution type has been used.
changed etc. Special causes are unpredictable, but they are distribution; in this case most of the points from the data set fall Process capability is a measure of the actual process variation as a
A rational sample is a collection of process output measurements
assignable causes or patterns which need to be understood and within the red lines showing that the data fits the Weibull ratio of the specification (customer requirements). Ford requires
(a subgroup) with only common cause variation affecting the
reacted to; otherwise, a process can drift further out of control and distribution best. If the distribution type was not verified before suppliers to use the Ppk index, which must be greater or equal to
samples. To ensure this, a small number of consecutively produced
in this example the car would run off the road. The following proceeding with calculating process capability the team would
parts are measured. These parts are produced under the same 1.67 at PPAP.
process conditions (i.e., not interrupted by tool changes, operator mistakenly think that the process is capable, but when analysis If a Ppk 1.67 cannot be achieved, then 100% measuring or mistake
changes, material changes etc), as these may introduce special using the correct distribution would show the process is not proofing is required until process stability and acceptable
cause variation. capable. Below the normal distribution shows that the process capability are achieved.

Version 1.0 – 6 April 2023 Page 2 of 2