Introduction
Today, many SQEs are confronted with geographically
disperse suppliers - often overseas and at times,
in locations where resources that can manage
quality systems are in short supply. Electronic
Data Interchange (EDI), ftp exchange of files
and even faxes of hardcopy quality data are
common forms of a data exchange across the supplier/customer
interface (1). Simplicity, cost and ease of
implementation at the supplier usually leads
to optimal results in defining and maintaining
underlying quality systems infrastructure. A
minimalist approach that has been used successfully
is the publishing of a secure and restricted
data source on a web-enabled server at the supplier
site.* The data source gets updated routinely
and is under the administration of the supplier.
The appropriate statistics that have been agreed
to are posted at the data source. In this example
(see Table 1), that includes lot attributes
such as Date, Supplier Site, etc. and a key
supplier performance indicator, PPM-D, or Parts-per-Million
Defective, which is derived from the Qty_Supplier_Rejected
column shown in the appendix.
Table 1. Simple text delimited
table securely published on remote system at
supplier site visible via network mount to Statit
e-QC™. Only a few rows of approximately
8,000 records shown.
Analysis
One can chart and disposition an indicator
like Qty_Supplier_Rejected and use historical
performance alone to monitor a supplier's performance.
However, there is additional analysis that can
reveal shifts in the variation of the data that,
when viewed over a sufficient time horizon,
can indicate changes to the supplier process
that might not be known or anticipated (2).
We always want our suppliers to improve, but
in a controlled manner that does not result
in unexpected consequences. Figures 1 and 2
illustrate typical supplier quality process
control charts that, when automated using Statit
e-QC, provide automated reporting that gets
delivered directly to email (3). Note that cpk
of the process of the time horizon shows stability
although this supplier is clearly challenged
to produce lots within defectivity specifications.
Figure 1. cpk of process where
defectivity (rejects) is used as a global Quality
Metric for conformance to specification
Figure 2. Trending cpk over time
for baseline control
Figure 3 presents the identical dataset analyzed
using variance of the metric using a Cum probability
plot. Now, one can see there is underlying structure
within the distribution of the Quality Metric.
Is that structure sufficient to declare a shipment
in violation of spec? Most likely, not. However,
the inquisitive SQE would like to know the source
of the variation, if possible, and potentially
query a supplier on possible changes to process,
their supply-chain dependencies, etc., as a
preventative measure to ensure unexpected or
heretofore uncontrolled product attributes are
within acceptable limits.
Figure 3. Cum plot of distribution
where continuous tails (no disjoints) indicate
natural variation and discontinuous jumps reveal
mechanisms or systematic factors affecting the
Quality Metric
Figures 4 and 5 show the drilldown, or segmentation,
of the supplier's quality data based on an attribute
the supplier has provided in the usual course
of data exchange. Note the even stronger signal
associated with Vendor Site indicating that
Vendor Site #3 is broadening the overall variation
in the data to a greater extent than the other
sites.
Figure 4
Figures 4, 5. Drilldown into segmented dataset
by splitting on "VendorSite" illustrates
Site #3 as showing more variability than Sites'
1, 2 and 4.
Also of note, the variation at Vendor Site
#3 has a marked slope change. This indicates
a change in mechanism, a process change and/or
possibly a measurement change. The chronological
signature is evidenced as shown in Figure 6.
These changes in variation get masked in routine
quality data acquisition/analysis and do not
always appear as deflections with sufficient
signal strength in cpk charts to warrant an
alarm. For example, both a shift in baseline
and shift in the width of the distribution would
have the net effect of leaving the cpk metric
unchanged.
Figures 6, 7 and 8. Running
average of DPPM does not indicate any control
violations based on att rules set when using
internal control limits. However, if Site #3
were to implement UCL from Sites 1,2 and 4,
essentially all points are in violation.
Figure 6
Figure7
Figure 8
Conclusion
When using Statit
e-QC as the centralized core of a web-enabled
data analysis infrastructure that acquires data
from distributed sources, the SQE can rapidly
distribute and communicate supplier analysis
results back to the source of origin. A common
knowledge base enables business discussion pertinent
to addressing and highlighting of both real
excursions and non-conformance, as well as more
subtle shifts in data (e.g. the dataset 'fingerprint').
The drilldown analysis, automated over a set
of indicators, can also be manipulated by a
wide range of literacy/skill levels. Trivial
data export on concise datasets (per chart level)
allows users to step back into their own tools
and native languages/compute platforms, for
further presentation if needed. More time spent
on control & improvements, less time spent
rearranging and reformatting results, managing
permissions and chasing after spreadsheet files.
References
1. Christoph Schroth, Till Janner, Alexander
Schmidt & Gunther Stuhec, From EDI to
UN/CEFACT: An Evolutionary Path Towards a Next
Generation e-Business Framework,The 5th
International Conference on e-Business 2006
(NCEB 2006) (Bangkok, Thailand)
2. EugeneGrant, Richard S. Leavenworth, Statistical
Quality Control, 7th ed., pp. 76-82
3. Statit Software, Inc. Statit e-QC Product
Guide. Please refer to http://www.statit.com/statiteqc/Statit_e-QC.pdf
*Two types of published/mountable data sources
are Linux Samba smb daemon running on Fedora
7 and Microsoft share. Both implemented under
restricted/secure login. Additional details
available upon request.
