CO2_2016 - page 30

28
Chimica Oggi - Chemistry Today
- vol. 34(2) March/April 2016
KEYWORDS: Design of experiments, DOE, modelling, data mining, screening, optimisation.
Abstract
The method of Design of Experiments (DOE) was invented almost one hundred years ago by Sir Ronald
Fisher and has since become an invaluable tool in helping improve product and process performance in
various scientific and engineering disciplines. Still many academic and commercial organisations prefer to rely on expert judgment
alone, or experiment by varying one factor at a time and are not benefiting from the DOE method. We find organizations who have
not embraced DOE tend to have more problems than time available to solve them; regularly cut corners, so limiting their
understanding of the real drivers of process robustness, effectiveness, and efficiency; and regularly need to fix problems with existing
processes which divert resources away from innovating and developing new processes and products. This article will show how recent
advances in DOE and predictive modelling methods simplify the application of DOE offering organisations competitive advantage
through increased process and product understanding.
Making design of experiments work in industry:
A case study in optimising an existing chemical process
HOW CAN DESIGN OF EXPERIMENTS HELP MY BUSINESS?
Client A was unable to predictably scale-up and transfer
into production, processes for making new products. This
resulted in delayed product launches and poor predictability
of supply once in production. Engineering experimental
practice was to vary one factor at a time (OFAT) to try and
fix problems. They lacked DOE knowledge and considered
the learning overhead too high. To reduce the learning
threshold and perceived complexity of getting started with
DOE, a DOE application customized to their own terminology
and engineering language was developed. This enabled
their engineering community to adopt DOE easily and they
are now scaling-up and transferring processes right first time.
Process scale-up and transfer has become predictable
since engineers at many of the production plants worldwide
have adopted DOE. Compared with the old approach of
varying one factor at a time, they are optimizing and scaling
up production processes with fewer individual experimental
runs and estimate this is saving them €2M per production site
per year in reduced experimental effort alone. There is huge
upside from getting to market faster than the old approach
and doing so predictably.
Client B needed to double the capacity of a product line to
meet growing demand. They had limited understanding of
the key process steps, a large number of potentially important
variables that could be influencing throughput and limited
budget for experimentation. They used a Definitive Screening
Design – an innovative new design of experiments – to
minimise the experimental outlay. Statistically modelling the
resulting data delivered the know-how to help double the
production rate with no capital investment! The case study
that follows is based on this situation.
BACKGROUND TO DESIGN OF EXPERIMENTS
Pioneered by Sir Ronald A. Fisher (1) while at Rothamstead
Experimental Station, England in the 1920’s. Fisher first applied
DOE to increase crop yields in agriculture and it is widely
accepted that DOE has played a major role in increasing
agricultural production since the 1930’s as illustrated in Figure 1.
Fisher introduced four DOE principles:
1. Factorial Concept - varying all the factors together using a
factorial grid rather than varying one factor at a time (OFAT).
2. Randomization – randomizing the order of the individual
experimental runs in the factorial grid to avoid bias from
lurking (unidentified) variables.
3. Blocking - to reduce the noise from nuisance variables.
4. Replication - to reduce the potential masking of
experimental factors due to noise (unpredictable) variation.
The DOE method has been improved and enhanced in the
decades since Fisher invented the method:
-
-
In the 1930’s, Yates also at Rothampstead, simplified the
analysis of DOE data by introducing Yates algorithm,
PROCESS OPTIMIZATION
MALCOLM MOORE, PHIL KAY
JMP Division os SAS
Henley Road, Medmenham
Marlow SL7 2EB, United Kingdom
Phil Kay
Malcolm Moore
1...,20,21,22,23,24,25,26,27,28,29 31,32,33,34,35,36,37,38,39,40,...68
Powered by FlippingBook