|
|
Today’s manufacturing is very
complex. There are many stages of operations, and many variables associated
with each operation in each stage. Even the most talented engineers, charged
with controlling variables to maintain consistent quality, encounter problems
that have unknown causes. These problems lead to product variability, rework,
and rejects. Process engineers scramble to understand the issues, and often
rely on guesswork to piece together relationships between variables. Data
mining takes this burden and gives it to the computer to quickly and
exhaustively find those relationships which are meaningful. Armed with this
knowledge, the process engineer can then focus on how to make the process
better.
Data mining is an amalgamation of
analysis, search, and modeling technologies developed over the last twenty
years. These include signal processing, clustering, association rules, decision
trees and multivariate regression techniques based on machine learning, such as
artificial neural networks (ANN) and multivariate adaptive regression splines
(MARS). Companies such as IBM, ORACLE, and SAS sell data mining software to
large organizations for their data warehouses or data marts. They target
marketing and financial applications, and generally require the employment of
people with sophisticated mathematical and computer skills.
More recently, better user
interfaces with advanced visualization have made data mining technologies
accessible to a broader range of people. Process engineers can now apply data
mining to manufacturing, but experience and knowledge of how the tools work is
still important. Users can encounter difficulty in getting their data loaded
into the software, or in knowing which technique would work best for their
particular problem. Once learned, however, data mining can give the process
engineer a powerful tool to uncover new truths and tackle the stickiest
problems.
In the automotive industry,
data mining can be applied to testing, quality control, painting, and
utilities/energy/environmental management. Each of these has several
sub-stages, or phases that could all be optimized with data mining tools and
techniques. For example, in the manufacture of components such as castings,
forgings, extrusions, and stampings, there is a myriad of variables such as die
design, lubricants, quenches, and heat treatments that affect final part
dimensions. Complex variable interactions can create tolerance stack up-related
problems in machining and assembly. In painting, variability in ambient
conditions, paint systems, and complicated electro-mechanical spray equipment
interact in unpredictable ways, such that high rework rates are common. Even
the most advanced virtual process CAE methods, such as feature-based CAD and FEA,
do not always lead to trouble free production trials and process operations.
With
data mining, process engineers can generate visualizations of how multiple
variables interact, and then create predictive models that are used to
determine how to minimize problems, or optimize processes. The 3D plot above
illustrates a multivariable process problem. It shows the response of a quality
variable (vertical axis) to variability in one controlled and one uncontrolled
variable (horizontal axes). The actual process, and therefore the predictive
model, involves several more variables. The response surface moves about
through the measured data points when these other variables are manipulated. In
this case, a high value for the quality variable is undesirable, so manipulating
the controllable variables to move downward along the response surface is
desirable. Search routines are combined with the model to tell how to best
configure the controllable variables to compensate for fluctuations in
uncontrolled variables.
Advanced Data Mining, LLC
(www.advdatamining.com) believes forward thinking companies can gain competitive advantage
through the discovery of new knowledge using data mining and advanced
visualization. Also, by employing this new knowledge to optimize their recipes
and process controls, these companies will realize additional benefits in
quality, yield, and productivity. The tools and services provided by ADM are
geared towards manufacturing and other physical processes. They let engineers
and scientists see their processes in new light, and make it easier to optimize
them on an ongoing basis. ADM has created a technique called multivariate phase
space reconstruction (MPSR) that optimally models processes where periodic,
chaotic, and noisy behaviors are inherent in the data. Using technology derived
for 4D medical imaging, ADM provides advanced data and process visualization
products for data mining application development and deployment.
Take a look at data mining
technology, and see where it might benefit your company. If an improvement in yield makes a big
difference in profitability for your company, you need to find out if your data
is trying to tell you something.