Portfolio > 5b > 1. Previous Experience > 2. Experience at A.F.R.L.
Search this site

5. A reflective researcher with skills in:
(b) Conducting research

5b1.2 Notes and Reflections on Experience at A.F.R.L.

Portfolio Documentation
Experience as scientist, mathematician and statistician at A.F.R.L. (Australasian Food Research Laboratories).

I worked as a laboratory technician while I was a student at Avondale College, Australia. I was enrolled as an external student at London University, and my undergraduate degree was in science, with majors in Mathematics and Physiology. My workplace was the food research and development facility of the Sanitarium Health Food Company, and for two of my years at Avondale, the company gave me a full tuition and living expenses scholarship. In return, I agreed to give two years of service for each year of scholarship.

I completed the Bachelor of Science in the middle of 1969, and for the next eight years, I worked as a scientist in the laboratory. My role was largely to support the other scientists as mathematician and statistician, but I was also involved in research into analytical methods from time to time, and wrote up many standard procedures.

Scientific Reports Written

I have been able to locate only two of the reports I wrote up from my time as a scientist in the laboratory. Most of the research I did personally related to instrumentation and measurement methods. The first of the reports I was able to find, "The Relationship Between Outside Air Humidity And Finished Biscuit Moisture", was written in September, 1977. I used a Stepwise Linear Regression procedure to analyze the results, and concluded that "humidity had no effect on Finished Biscuit Moisture". The other report was written in January, 1978, and is an example of research on a measuring instrument. The A.F.R.L. had designed an instrument for quickly determining the moisture content of a cereal called Weet-Bix (still Australia's number one breakfast food). The instrument incorporated scales for measuring weight before and after, and used Infra Red (I.R.) radiation to heat the biscuit and remove moisture. The report was called "Voltages In I.R. Moisture Meters" and the research revealed some errors both in the manual for the instrument and in the actual use of the instrument.

Radioactive Isotopes

One experiment I remember very clearly involved a radioactive isotope of sodium with a half life of under five hours. One of the products the company made was Marmite, a product similar to Vegemite which is made by Kraft. Marmite was made largely from brewer's yeast, but the manufacturing process did not allow for sufficiently tight control over properties such as viscosity, color, and salt content. The finished product was produced as a blend of many different batches, using a weighted average to achieve target quality control limits. The radioactive isotope was used to resolve a question concerning the best way to operate the blender, a huge 2-story high cone-shaped container with an Archimedes screw that could rotate up or down as it revolved around the outer edge of the cone. Some operators claimed that the most consistent mix was obtained by rotating the screw up, others felt from subjective observations that a better mix was obtained by running the screw downwards.

A team of scientists and technicians set up shop at the end of one shift and prepared for 24-hours of non-stop sampling and analysis. The procedure was to introduce the radioactive isotope into the top of the large cone of Marmite being blended, then take samples to measure radioactivity at the very edge, one foot in from the edge, and 3 feet in from the edge at regular intervals, more closely spaced earlier in the experiment. I was equipped with a programmable calculator and the formulae for radioactive decay. As the samples were taken measurements were made with the Geiger Counter, I fed the readings into my program and plotted the results on graph paper for each of the three sampling positions.

We performed the experiment with the screwing driving down then up, and the results showed conclusively that effective blending is achieved in much less time driving down than up. After correcting for half-life decay, the variability in radioactivity readings was well within tolerances after only 3 hours of mixing when driving down, and was much better than that achieved after 9 hours of mixing up. The previous standard practice had been to mix upwards for 5 hours, so we felt very pleased that our marathon 24-hour experiment allowed us to achieve better mixing in better time. The Marmite from each of these experiments was stored in drums with large hard-to-miss labels warning of radiation danger, but the already low initial levels were virtually undetectable a day later. Each of us had our radiation detection badges processed after the experiment was over, and were grateful that through safe practice, no one had received undue exposure to radiation.

Response Surface Methodology

The Director of the food research laboratory put me in touch with some material on an experiment design technique called "Response Surface Methodology" or RSM. This methodology allowed for the efficient design of experiments with multiple variables, and gives optimum value ranges for the independent variables for achieving target values in the dependent variables. Quadratic functions are formed for each of the variables and the resulting matrix of simultaneous quadratic equations is solved using matrix inversion techniques. Sets of quadratic response surfaces are plotted for selected sets of independent variable values. One of the attractive attributes of the methodology in food research and manufacturing is that one is not only able to find a set of variables that yield outputs in the desired range, but one can choose a set where there is low sensitivity to variation in the independent variables. This is presented graphically as flat spots, plateaus, or places with a gentle slope on the response surface. One is able to detect saddlepoints, valleys and ridges with steep slopes, and avoid them.

The following examples of response surfaces are taken from the "Design and Analysis of Experiments" course at http://www.colorado.edu/EngMgmtProg/usrey/5550/rsm.html (on local server). In each of these surface plots, independent variables are plotted on the x and y axes. If there are other independent variables, then they are assigned constant values for each of these plots. The vertical or z-axis is the dependent variable.

RSM Figure 1
1. Assuming that larger is better, the "peak" represents a combination of independent variables for which the dependent variable is optimized.
 
RSM Figure 2
2. When there are many peaks and valleys, as in this example, it becomes more difficult to use RSM.
 
RSM Figure 3
3. A surface like this represents an ideal solution, because the dependent variable is relatively stable for changes in either of the dependent variables.
 
RS MFigure 4
4. The variability in the dependent variable is greater in the case of a saddle, since deviations in one independent variable cause increases in the dependent variable, while deviations in the other independent variable cause decreases.
 
RSM Figure 5
5. This is an even more desirable solution than the gentle hilltop in example 3. In this case, changes in one of the independent variables has no effect at all on the dependent variable. Quality control is thus made that much easier.


Using this methodology, I helped other scientists in the laboratory to design their experiments, and to analyze and interpret the results. This was during the 1970s, and it seems to me to have been an advanced methodology for its time. I very much enjoyed applying the methodology and using the statistical knowledge I had been learning in the Master of Engineering Science courses I was taking at the time at Newcastle University.

The Taguchi Approach

While RSM continues to be taught in statistics courses and to be applied in industry and in research, there is another more recent method that seems to have many of the same objectives, and that is the Taguchi Approach. I first learned of the existence of this approach while taking a course entitled "A Scientific Framework for Product Development" in 1996 as part of the coursework for the Master of Computing I earned from Macquarie University in 1998. According to Dr. Ray Offen, one of my professors for this course,

The Taguchi Approach utilizes experimental design to ...

  • design products and/or processes so that they are robust to environmental conditions
  • design/develop products so that they are robust to component variation, and
  • minimize variation around a target value. [1]

I can't help feeling that we would have adopted the Taguchi Approach if it had been available back in the 70s when I was designing and analyzing RSM experiments.


[1]Offen, R J, from 1996 Course Notes for Statistical Quality Control segment of course at Macquarie University, Sydney on Scientific Framework for Product Development


Return to 5b1 Previous Experience

Created: Wednesday, October 17, 2001 10:13 PM
Last Modified: Saturday, January 10, 2004 10:21 PM