From the (Apr-May-Jun.) 2002
Collecting data is a messy costly business. But,
you have to do it. Sometimes there are no good ways of doing any given
task, just a way. However, there are small things that you may be able
to do that will reduce the cost by a few cents. Multiply that savings
by the number of points of data you collect and then multiply that by
about 365 and you will see that you can save some costs.
Auto-collecting data, or data logging, may seem like a really good
idea on the surface, but as you dig into it you may find some problems
that will actually cost you more in the long term. I got a call from
some magazine involved with metal stamping or making screws, (or
something like that), in the UK that thought by collecting data on
100% of the product that they would be able to do some kind of
statistics on that data. Yes and no. You can do statistics on the data
collected providing you just look at a sub-set of the 100% data.
If you try doing SPC,
(statistical process control), on all the data, you are not
doing SPC. Not only that but, you could still have a problem with the process
that may not been seen until it is too late. Which of course leads me
to a short story. It really depends on if you are collecting data all
day long, like in incoming inspection, or if you are an operator and
just collecting a few samples every few hours.
Auto-collecting data has been going on for years in the testing of IC
chips. With the testing of IC chips, you usually have a test fixture
connected to a workstation, connected to a computer. You load the
testing software into the computer and on the workstation is a box
with a button you push to start the testing and a display of some type
that will display a 'bin' number. This bin number can be used to sort
the IC chips. We rarely sorted ICs' unless engineering wanted to see
how many bin 'x's they were getting. All most always, a bin one, '1'
went in to be processed more and any other bin numbers were scrapped.
The bin numbers on that computer could be from zero to ninety-nine.
They were only using about half the numbers they could have used. I.E
1 to 50. So they had about 50 more numbers they could have chosen.
The workstation with the display on it also had a little hood on top
of the box to block glare so the operator could read the display
easier. What happen was the operator was sitting too high and the hood
blocked only the very first segments on the LCD. That was fine except
that bin sevens, '7' looked like bin ones, '1' because the line going
down on the 7 was straight and not at an angle. The operator ran all
night putting bin 7s' into the bin 1 tray.
So auto-collecting data is fine but also make sure your process
is as flawless as possible. I ask the supervisor, why TSE, (test
system engineering), was using bin 7. They still had a lot of
other numbers they could have used. If they made all the bin 7s' in
all the test programs some other number, like with 2 digits instead
one digit, then that problem would have never happened. Also using one
digit for a good part and 2 digits for a bad part allowed there to be
more disparity between good and bad parts when looking at the display.
So they changed at least that part from bin 7 to bin 75 or something.
We then knew that particular problem would not occur again. Common
If you do data logging using custom software, you
need to insure that the process of connecting the data logger to the
software to analyze the data is as easy to do as possible. It does not
do you any good to have to spend 15 or 20 minutes each time you set up
a process to collect data unless you are running enough of a volume of
parts to justify the time. It can take as little as 15 seconds to
enter numbers into our statistics program via the keyboard. And if
there is a timing problem between the data logger and the analysis
software, it takes even more time for the operator to fix the problem,
and this may make your operators less productive. The problem usually
comes into play where you try to make something do a task it wasn't
designed to do. For example, one would not try to pull a fully loaded
semi-trailer with a FORD F-150 pick-up. At some point it will break
because it was not designed or tested to do such a task. While this is
a simple concept, I see people trying to do stuff like this all the
time. It is however possible to retrofit machines, software, etc. to
do what you want them to do.
While the device you use to collect data into your data
analysis software may not have the ways and means to put the data
onto a PC, it will most certainly have a way to print the data.
Yesteryear, the printer was almost always a serial port type printer.
So what I have done on more than one occasion is take the printer
output port on the measuring device and hook it to the serial input
port on a PC. Then using any modem program, you can get the data to a
text file. Depending on the software used to print to the printer, you
will find some extra characters in the text file that are used to
control the printer. Like form feed, etc. Once you have the data to
the PC's hard drive, you can write a parsing program, a program to
strip out all the control characters, and then put it into your data
analysis software or create a comma delimited file to import it into
your data analysis software. And remember we can help you automate
your data collection no matter if you are using ZeroRejects or not.
I had a Six Sigma course that dealt with software. The course taught
that a three minute interruption was the same as a 20 minutes
interruption in the programming process. This is because you lose your
concentration and have to get back into the swing. Not
only that, but, this is not only true with programmers, but most
everyone who has to use their brain doing their jobs. It may be 5
minutes instead of 20, but an interruption will do that. It is not a
one to one ratio because we are humans. When you stop to collect data,
you just created an interruption. So you may not be saving time at all
by only measuring one-piece part instead of 2 parts. For example, if
it takes an operator 2 minutes to shut down and make a measurement, it
may be 5 minutes before they get back to being 100% productive. This
may be the same amount of non-100% production time if the measure is
of one piece or two pieces. And by measuring 2 pieces, you can do real
SPC. You can check your production rates at various times to see how
your company is.
So try, and it is not always possible, to make your
SPC measurements when the operator is going to be interrupted anyway.
Like before lunch or breaks. There is also a theory that indicates
that before lunch and breaks there is a higher possibility for an
operator to make a mistake. This is because they are less attentive
before breaks. After breaks, it takes them a few minutes to reach 100%
production. Use these gaps where production is less than 100% anyway
to make your measurements if possible. There may be times with some
processes in some companies where this is not applicable. But if you
can do it, you will actually be saving some money.
Motorola and Six Sigma are
Trademarks of Motorola, Inc.
Ford, and F-150 are registered trademarks of the Ford Motor Company