Design of Experiments
Design of Experiments (DOE) is the term applied to the use of sophisticated statistical
tools to economically optimize manufacturing processes. Companies that use these tools
typically have manufacturing costs, not counting materials, that are 20% to 50% of their
industry competitors. Cycle times are equally low and inventories can be reduced because
the processes are more reliable. Companies that do not incorporate these tools
often have unending quality and cost problems; the resolution is always on the horizon.
The companies that have developed the most efficient manufacturing
processes are committed to two kinds of continuous improvement:
- First, transformation processes are analytically monitored, well
maintained and minor, operator initiated, improvements are made on a regular basis.
Second, the processes are periodically subjected to rigorous, traditional
industrial engineering scrutiny: Can/should the process be modified, eliminated,
combined, simplified or should the sequence be changed?
One of the critical elements of managing manufacturing material transformation
is aggressively optimizing existing processes with economic performance as the
key optimizing parameter. Sophisticated statistical and
mathematical tools are
an essential part of the optimization process.
These tools are commonly known
as Taguchi techniques and Design of Experiments (DOE).
Statistical Tool's Power
Traditional process analysis involved changing one variable at a
time and examining the impact of the change. It is difficult to discover
or understand interactive effects using traditional techniques. In a
manufacturing process analysis the power of statistical tools includes:
The ability to expose relationships among interacting parameters. For
example in a metal forming process there may be a link between the metal
temperature, the type of forming lubricant and the surface finish of the raw material.
Statistical tools allow several variables to be changed simultaneously,
dramatically reducing the number of trials necessary. Both time and cost are reduced.
Occasionally a parameter that was considered to be immaterial will be
discovered to be important.
Understanding the relationships and impact of process parameters
is the key to improving and economically optimizing processes.
It is common for the analysis of early sets of experimental trials to reveal
unexpected results. For example, in some metal forming operations,
lubricant temperature and thickness have proven to be unexpectedly important.
Metal forming speed has also proven to be a surprising variable with specific materials.
Design of Experiments - The Process
Successful manufacturing process control and optimization experiments
proceed through a proven series of steps as follows:
- Select a multi-disciplined team familiar with
the manufacturing process, materials, experimental techniques and economic analysis.
Analyze the process, i.e. choose the independent and dependent variables
At initial stages it is important to capture all of the potential
variables. It is often tempting, especially for experienced people, to
dismiss factors that are "intuitively" irrelevant. In the course of
the experimental process and analysis surprises are usually discovered.
Design the experiment, the range of independent variables to
be studied. This is the "art" of the process. The most robust design
will include multiple "levels" of every independent variable.
The number of experimental runs required is equal to the number
of levels for each factor times the number of levels for every
other factor. The matrix can get large (therefore expensive) quickly.
Partial or fractional level factorial designs (Taguchi) are often used
to reduce the cost and time required for experiments.
Design/build the experimental equipment Production equipment is rarely either
adequately instrumented or adjustable for good experimenting. Measurement
and adjustment systems must be calibrated. This can be time
consuming, expensive and can require sophisticated instrument engineering skills.
- Perform the experiment(s)
- Analyze the data
- Iterate the process based on the results of the previous experiment.
- Reach technical conclusions about the critical factors and their impact.
Rationalize the chosen factors based upon their economic impact.
This often leads to short term factors based upon the capability of existing
production equipment, and longer term plans based upon modifying or replacing
the equipment to achieve better economic performance.
- Establish process "run" parameters.
There are also two differing basic approaches to executing the experiment.
Either can be equally effective.
- Run the experiment as a non-production exercise,
as a research activity - "off-line." This is the most common approach
Perform the experiment during production, on production equipment,
using natural process variations to adjust the process and
"modify" it in "real time." This is best accomplished when the
process parameters can be automatically changed and monitored.
Injection molding is an example where this approach has proven valuable.
The techniques of Design of Experiments are designed to reveal what
factors are important in a manufacturing process and what their settings should be.
The techniques are not designed to reveal any underlying theory.
Design of Experiments is largely an analytical process. Once the key
factors are understood an economic analysis is usually necessary as part
of the process of "optimization." The synthesis of the final "run" factors follows.
Design of Experiments leads to optimizing a process. It does not assure
that the process optimized is the best process for the job. It can be
seductive and can lead to "sub-optimization" of the entire manufacturing process.
Design of Experiments is largely an analytical process. Once the key factors are
understood an economic analysis is usually necessary as part of the process of "optimization."
The synthesis of the final "run" factors follows.
Sophisticated statistical and mathematical tools are an essential part of the optimization process.
Statistical tools allow several variables to be changed
simultaneously, dramatically reducing the number of trials necessary.
It is often tempting, especially for experienced people, to dismiss factors that
are "intuitively" irrelevant. In the course of the experimental process and
analysis surprises are usually discovered.
The techniques of Design of Experiments are designed to
reveal what factors are important in a manufacturing process and what their settings should be.