|
It's complex. It's expensive.
And you can't live without it: Airplanes wouldn't fly, bridges
wouldn't stand and pharmaceuticals wouldn't heal. And, most
likely, profits wouldn't be made. Statistical process control
software is one of the most important investments you can
make in your quality program.
It's also difficult to select. SPC software is used to
analyze data and make intelligent decisions about products
and processes. But, how do you make intelligent decisions
about which SPC software package to buy?
Reviewing and selecting an SPC software product that's
right for your organization requires time and money, certainly.
However, the right choice can be made with confidence if
you know what to look for before buying.
Not all SPC software competes for the same market. By
knowing your needs from the start, you'll save a tremendous
amount of research time. This article provides a list of
topics that go beyond the obvious (i.e., SPC software that
creates control charts, histograms, statistics and capability
indices while supporting some form of data entry).
There are two categories of software that perform SPC
functions: post-analysis and real-time. Terms such as high-end
analysis, spreadsheets, data files and ODBC links describe
post-analysis software. These applications either stand
alone as spreadsheets or link to external databases like
LIMS and ERP systems. Such solutions are popular in academic
settings, laboratories and research environments.
Post-analysis statistical software is designed to "pull
gold nuggets" from pre-existing data sets. With this
software, data gets collected, and then someone else performs
the analysis. Popular products that fall into this category
include STATISTICA, MINITAB, STATGRAPHICS, SPSS, SAS, JMP,
GenStat, Microsoft Excel and various Excel add-ins.
In contrast, real-time SPC products are accessible across
corporate networks and specifically deployed on factory
floors--where machine settings, product runs, measurement
devices and people are always changing. In the real-time
world, each new data point offers an opportunity to either
make a process change or do nothing. Reliable real-time
SPC software should provide effective alerts when process
changes are necessary.
Because many organizations deal with these parameters,
this article will focus on selecting a real-time SPC software
product.
Data must be stored somewhere, either into data files
or a database. Products that use data files typically maintain
a separate file for each part. For example, 50 files are
required to store data from 50 parts. If your company makes
thousands of parts, you'll end up with thousands of data
files. Setting up a single file is usually simple, but managing
data files from thousands of parts can become overwhelming.
However, organizations with limited SPC requirements get
along quite well with a file-based system. If all you need
are basic printouts of control charts and histograms, you
might want to consider an Excel add-in, such as SPC XL.
There's no point in spending thousands of dollars when an
inexpensive solution will do the job.
In contrast, "databased" products are best suited
in corporatewide SPC deployments where the same data must
be accessed from multiple locations and workstations. Because
all the data reside in a database, these products are excellent
for comparative analysis across multiple parts, processes
and test characteristics. Comparative analysis, however,
requires a relational database organization. That is, all
the raw data reside in one table, and the items tagged to
the data--such as the part, process, test name, employee,
specification limits and so forth--are referenced from other
tables. A true relational database design employs "referential
constraints" in which a database item, such as a part
number, is defined only once in the database but can be
referenced several times in the database. This type of design
is very efficient and provides unlimited data analysis,
sorting and comparison functionality. One warning, however:
Simply because the data are stored to a relational database
doesn't necessarily mean that the relational features of
the database are used in the application. Finally, database
tables should be accessible by third-party applications,
such as Crystal Reports.
Obviously, the software you select must comply with any
regulatory requirements under which your organization operates.
The most common is the FDA's 21 CFR Part 11: Electronic
Signatures and Records. This regulation is specific to the
pharmaceutical and medical device community but will eventually
become common practice across industries producing products
that enter the human body. As a rule, applications that
store quality records to data files can't become compliant.
Any application that stores data to a local PC and then
forwards the data to a master database is managing a system
with two copies of the same quality record. This approach
also violates 21 CFR Part 11.
When considering a specific SPC software product, ask
for the validation document showing that the calculations
are accurate. Also ask about the calculations' precision
(i.e., How many significant digits are reliable?).
If the software will be deployed on the factory floor,
a representative from the floor must contribute to the decision
process. The best way to test shop-floor acceptability is
to pilot the software on the floor for a couple of weeks.
This takes a lot of time and resources, so this step should
only be done once you've narrowed down potential selections
to no more than three products.
Features to look for include easy and intuitive navigation
with very little button clicking. The interface should be
clean and simple. Because the shop floor user may have limited
computer and/or SPC knowledge, all unnecessary buttons and
menus should be inaccessible unless the user has proper
security privileges. Charts should be full-screen and/or
locked in place, and individual windows shouldn't move or
close. Big buttons, big message boxes and color designations
that easily distinguish different screens and results are
also important. The data collection interface must be simple
and flow automatically through the collection steps in a
logical manner.
Put the company's support service to the test. Call during
your evaluation period and make sure that a representative
can answer your questions. Ideally, the support person should
be able to access your desktop using a safe collaboration
Web tool such as WebEx. Ask to speak to the in-house statistician.
He or she should be able to help you with recommended statistical
tools for your needs. How the support service responds to
these statistical requests will help determine the software's
usefulness when an issue goes beyond simply making the software
work.
Regardless of the software, setup involves at least two
required phases: configuring a data collection plan and
creating charts. Within the data collection plan, you specify
the part, process, test characteristics, sample size, additional
tag fields, data-entry method and collection sequence. Once
this is set up and data have been collected, many software
products don't allow any changes to the collection plan.
But in reality, collection plans will probably need periodic
modifications. For example, someone may want to add a test
characteristic, change the sample size or change the collection
sequence. To avoid future headaches, make sure all options
are changeable on pre-existing data collection plans. Charts
should be treated in the same way: They must be configurable
and flexible.
For organizations that intend to make business decisions
with their SPC data, two issues are very important. First,
the control limits must be unique for a given part number,
process and test characteristic. If the control limits'
storage is based on a unique part and test characteristic
(as found in most SPC products), the user must be careful
not to mix processes within the same control limits. Tagging
data with the process and then filtering all but one process
isn't a good solution. Correct implementation with these
products requires a different file or setup for every process
that runs the same part. Software products that define control
limits using unique part, process and test characteristic
combinations are statistically correct and represent the
most efficient control-limit management approach. Make sure
the control limits are time-based (i.e., that updated control
limits support effective dates). Control limits from old
data aren't changed when updated limits are established
on newer data.
The other issue involves standard deviation calculations.
Specifically, the software must differentiate between the
sample (long-term) and estimated (short-term) standard deviations.
The long-term sigma (s, as calculated using the n-1 formula)
uses all the data values, independent of subgrouping. Long-term
sigma is used in Pp and Ppk calculations. Short-term sigma
(^s) is calculated based on the range chart. Short-term
sigma is used in Cp and Cpk calculations. Make sure these
differences are understood and correctly applied within
the software you choose.
Also make sure the software provider has industrial statisticians
on staff. This will help ensure the software's statistical
validity.
The SPC package you choose should be able to integrate
with other software products, such as those designed for
document control and corrective action. Launching a rejection
form when an event is triggered in the SPC program will
prove very useful. For example, with such functionality,
you're automatically able to tell an operator that the gage
being used must be recalibrated in three days, or that the
gage, based on the most recent gage R&R study, will
consume 26 percent of the measured characteristic's tolerance.
Some SPC products make such integration simple to implement.
If you require integration, make sure you ask the SPC software
provider how it can help you.
Subtle differences exist among the various products, so
as you close in on your ideal software, keep the following
tips in mind to ensure you're getting your money's worth:
Don't pay more than 15 percent of the software's purchase
price for the standard annual support fee. This should include
unlimited telephone, fax, e-mail and Internet support, as
well as all future updates.
Ask for copies of the company's software update logs. Look
at the update record. Make sure the updates include substantial
enhancements and new features, not just bug fixes.
Make sure that if the software supports e-mail, it isn't
an imbedded proprietary e-mail system that requires a software
license for anyone receiving the e-mail. The system should
be SMTP-compliant, allowing e-mail to be sent across the
Internet.
Look for a concurrent-use licensing scheme. The software
can be installed on an unlimited number of workstations,
but access to the software is controlled based on the number
of purchased concurrent licenses.
Don't pay extra for Web publishing. Some companies have
built-in utilities, at no additional cost, to allow automatic
Web publishing of the SPC charts and data.
Beware of any company that wants to sell you its data collection
and control charting software as well as a "high end"
statistical product for the analysis. This not only increases
the system's cost but also complicates deployment. Instead,
make sure the software supports simple data extraction into
any number of popular post-analysis statistical products.
Don't pay for validation documents.
Ask to use the software for at least 30 days at no cost.
If you want to conduct a pilot test, expect to be given
90 days.
Don't pay for expensive software if Excel add-ins will satisfy
your needs.
There's no better way to candidly discuss potential software
than to talk to companies that have already invested in
it. Ask the software company for three types of references--and
take the time to call them:
An organization that has made a recent purchase. Ask about
its selection criteria, initial implementation issues and
any buyer's remorse.
An organization that has been using the software for several
years. Ask how the company has been treated after the sell.
It should also have a wealth of knowledge in terms of lessons
learned.
An organization in your industry. It will be able to speak
your language. There are bound to be common problems that
all organizations in your industry share; ask questions
about how it has used the software to solve them.
Several providers offer excellent SPC software, so it's
important to first determine your needs before you start
shopping. This will help you limit your final selection
to a couple of products. At this point, start a pilot program.
Implement the product candidates in a limited fashion, and
let all interested parties play with the software and express
their opinions. Within a few weeks, the best choice for
you should be obvious. By following these steps, you'll
have the confidence that you're making the right decision.
Steve Wise, an industrial statistician and director of
statistical methods at InfinityQS International, is a co-author
of the original Boeing D1-9000 specification, Advanced Quality
System for
Boeing Suppliers. He's also co-author of the book Innovative
Control Charting, published by ASQ Quality Press. During
the last 15 years, he's worked with companies implementing
real-time, production-floor SPC and advanced quality management
tools. Letters to the editor regarding this article can
be sent to letters@qualitydigest.com.
Click here to find Quality
Digest's 2002 SPCSoftware Guide, a directory of more than
125 SPC software providers, complete with companies' contact
information, Web site address and a brief description of
their offerings.
Also visit Quality Digest's online SPC directory at www.qualitydigest.com/directories,
where you can search vendors by region or by their products'
specific functions (e.g., Pareto charts, simulation, regression
and more).
|
