Gaining Control with SPC Software
Charting the progress of statistical process and statistical quality control (SPC/SQC) shows a definite trend. It's straight up.However, the latest software promises to tame the charting beast. Vendors are integrating their wares more completely with supervisory control and data acquisition (SCADA) systems, targeting their products at audiences other than quality engineers, and opening pr...
Charting the progress of statistical process and statistical quality control (SPC/SQC) shows a definite trend. It's straight up.
However, the latest software promises to tame the charting beast. Vendors are integrating their wares more completely with supervisory control and data acquisition (SCADA) systems, targeting their products at audiences other than quality engineers, and opening programs up for easier data import and export.
These trends are evident in stand alone SPC packages from such vendors as Applied Statistics Inc. (St. Paul, Minn.), SAS Institute Inc. (Cary, N.C.), and SPSS Inc. (Chicago, Ill.). The same can be said for modules from such human-machine interface (HMI) suppliers as Intellution Inc. (Norwood, Mass.), National Instruments Corp. (Austin, Tex.), TA Engineering Co. Inc. (Moraga, Calif.), USDATA Corp. (Richardson, Tex.), and Wonderware Corp. (Irvine, Calif.).
Covering the basics
There are differences between the various programs, although most will meet basic SPC needs. All produce such standard SPC charts as X-bar/R and its various cousins. All but the module from TA Engineering also offer Pareto analysis and process capability statistics such as Cp and Cpk. The same group has run-time rules, such as raising a flag when x number of samples in a row have been high or low. In general the stand-alone versions offer more power than the modules do, although most will satisfy basic requirements. TA Engineering plans to partner with a software company to offer an integrated and more complete SPC solution. This partnership should be in place by second-quarter 1998.
One thing no SPC package can do is magically create expertise in statistical process control where none exists. Nor can software wrestle an unruly process into submission. That's why it's best to make sure the process knowledge basics are covered before any SPC software purchase.
"A first question [to answer] before people go and get SPC software is do they need and can they get SPC, period," says Lonnie Wilson, founder of Quality Consultants (El Paso, Tex.) See sidebar, "How to Get Started with SPC."
For those already underway, there is welcome news. One of the biggest trends in SPC software has been a move away from isolated analysis. In the past, SPC number crunching was done in an engineer's office after product was made and frequently shipped. Today, on-line SPC is increasingly favored. One advantage to this approach is less chance for error as well as a reduction in paperwork. And there are other benefits as well.
"You get the data off the manufacturing line and in a format the operator can see in more real-time," says Don Holley, industrial automation marketing manager for National Instruments.
SPC modules inside SCADA systems have a distinct advantage for on-line analysis. They're tightly integrated parts of the overall package. Since control points are already being monitored, all that's needed to implement on-line SPC is to set up which points to include in the statistics. There's only the HMI interface to contend with, making configuration and training relatively easy. The same tight integration also makes it simple to set up alarm conditions and should eventually lead to SPC being an active part of the process control loop.
The trend toward modules is so strong that HMI industry giant Rockwell Software Inc. (West Allis, Wis.) is in beta test with an integrated SPC module for its RSView product. It is based on work done earlier with sister company Datamyte Business (Minnetonka, Minn.). The older project was scrapped in favor of an internal Rockwell development. The new module is due for release second- quarter 1998.
As another indicator of this trend, Charley Rastle, vice president of system integrator Process Analysts Inc. (Lakewood, Colo.), reports that the last three quality systems his company installed have all been migrations from stand-alone packages to modules from HMI vendors.
Off-line isn't out of line
Stand-alone vendors are combating this by adding alarms to their products so that they too can be part of the overall control loop. As a result these programs operate under special constraints.
"It is critical for an SPC program to be flexible enough to work with a variety of gages and also have the capability to import data from a number of different data sources," comments Courtney Hye, manager of the Quality Products Marketing Group in the Quality and Manufacturing Business Area of SPSS.
QI Analyst from SPSS can, for example, gather data nine ways, ranging from reading ASCII files to importing spreadsheets, to handling direct input from devices. Offerings from the SAS Institute and Applied Statistics are just as adaptable. For the future, the ongoing development of open data exchange standards should make this task easier. However, the days of free-wheeling mix and match is still a few years off.
It could also be that both sides of the debate will prevail. The modules are at a disadvantage for those SPC tasks outside of a SCADA system. This can be a pressing problem in large organizations because SPC is spreading to nonmanufacturing areas. Some companies may want to standardize on a single package applicable to all operations. Furthermore, for extensive analysis the stand-alone packages offer more power. So it may be that the two flavors of SPC software will coexist—a view shared by some.
"Off-line software is complimentary to our software," declares John Eng, product marketing manager for SPC Pro from Wonderware. Mr. Eng won't divulge the exact number of users for SPC Pro, but he admits it is far fewer than the 70,000 Wonderware InTouch HMI users.
One result of this spread of SPC to the factory floor and beyond is that the programs are being used by a wide variety of people. Very few of them are quality engineers or statisticians. For the modules from HMI vendors the challenge is to make the software easy enough for an operator to use and understand. Stand-alone software has to appeal to a much wider group of people who likely have very differing needs.
"In manufacturing, we have a wide audience for our SPC software from the operator on the floor to the plant manager," remarks Donna Fulenwider, technical marketing specialist in the Statistical Quality Improvement Research and Development Group at the SAS Institute.
Customization fits all
To handle this spectrum of users vendors offer customization tools. The types of charts, frequency of data points, and control limits are all configurable. Several different alarms are also possible. Some of the simpler ones use a change in color; others can trigger line shutdowns. The stand-alone programs, in particular, offer extensive flexibility and personalization, so users should be prepared to spend some time experimenting with these options to get full use of the software.
A final trend for SPC designers is to move toward an easy interface with outside programs. All of these packages can talk to databases that support open data base connectivity (ODBC). That makes it possible to both export and import data across an enterprise.
As Bob Reed, strategic technical account development manager for USDATA, says, "The understanding of a better way to use data is coming into the marketplace today."
Some SPC Terms
Cp: Process capability calculated by Tolerance/(6 sigma). Only measures dispersion.
Cpk: Process capability calculated by the smaller of (Mean-Upper Spec)/(3 sigma) or (Mean-Lower Spec)/(3 sigma). It measures both dispersion and central tendency. A 6 sigma program implies Cpk of 2 or greater.
EWMA chart: Exponential Weighted Moving Average. It detects small recent variations by assigning an exponentially declining weight to data points as they age.
Pareto chart: A means of classifying problems from important few to trivially many. Defects are assigned a cause and categories ranked in order of number. The most significant problem will be the first.
Run rules: Guidelines to detect nonrandom process shifts. Some rules use six or nine data points on one side of the mean in a row. They are based on the likelihood that so many successive readings will not occur randomly.
X bar, R: This most popular of Shewart charts plots average (X bar) and range (R) for a sample.
How to Get Started with SPC
For those who've decided to implement SPC, software can be a "Jekyll and Hyde" situation. The right software will make things easier while the wrong will do just the opposite.
"Software that is too difficult to use will not be used correctly or at all," warns Sophronia W. Ward, a statistician affiliated with Statistical Process Controls Inc. in Knoxville, Tenn.
"The most important thing that a user should do is to define the software need carefully and who will be the user of the software."
It's important to make sure that the software satisfies the requirements behind any SPC system.
If SPC is being implemented because of a customer requirement or government regulation then whatever software is selected has to satisfy that demand.
Beyond that to be successful the software should:
Allow easy data entry;
Analyze the data correctly; and
Integrate the necessary tools to build knowledge about the process.
Furthermore, easy data entry should include some means for automating the process. While correct analysis should be a given, calculations should be checked.
The last point is an important one. With the proper tools SPC can be used to understand a process, minimize variation, and become a means for continuous improvement. However, no program can take the place of a capable control engineer.
As Ms. Ward notes, "Software does not improve processes, people do."
Putting SPC to Work
The Dow Corning plant in Elizabethtown, Ky., manufactures 1,000 different sealants and adhesives. Before March 1997, readings from those products were entered by hand into an SPC system.
Since then, the company has implemented a system based on QI Analyst from SPSS. The new software cut down on paperwork and errors while providing visual feedback for out-of-control points. Operators click on points to enter discrepancy explanations.
In another application, Coleman Co. Inc.'s plant in Wichita, Kan., makes outdoor camping gear. SPC is used in the press shop, where parts are made from large rolls of steel. Coleman runs an SAS software application on a real-time server over an IBM OS/2 network. When operators take a gage reading, the data automatically enter the database and update four different control charts. This provides immediate feedback.
"Charts come up on-line as they're running and show whether the process is in control," explains Dallas Hart, a quality engineer with Coleman.
Alarms to prediction
The Aluminum Co. of America (Alcoa) plant near Yankeetown, Ind., produces more than 300,000 tons of aluminum a year.
Prior to going into everything from cans to venetian blinds, the metal first has to go through a smelter. During the smelting process metal is checked for weight and thickness. With software now in use, an alarm automatically sounds when an SPC error occurs.
Under a partnering agreement with Rockwell Software (West Allis, Wis.), Alcoa looks ahead to more aggressive process control. According to Joseph Motz, a staff electrical engineer at Alcoa, the new software will also allow on-line SPC because of the benefits this brings.
Mr. Motz says, "We can react before it's too late. We can proactively react."