"Quality is a virtue of design," says Genichi Taguchi, the Japanese guru on Experimental Design. Now, putting the truism to work is Ranbaxy Laboratories Ltd. which is revolutionising the way it conducts bulk drugs research through a simple statistical tool called Design of Experiments (DOE).The concept of DOE, interestingly, germinated in the field of agriculture, in 1919, when Sir Ronald A Fisher, a British statistician developed a strategy for systematic experimentation. Subsequently, DOE's popularity revved up in the automobile industry with nearly every global automobile major using DOE for the design of automobiles and automotive components.
Currently, while Ranbaxy is the first pharmaceutical company in the country to use DOE techniques to develop new bulk drug formulations and processes, SamClor Glass Ltd, a joint venture of Samtel India and Corning, USA is using the technique for developing black and white monitor glass.
At the heart of DOE lie templates for experimentation called "orthogonal arrays". An orthogonal array is much like a sophisticated switching system into which many different design variables and levels of change can be plugged. With the help of the orthogonal array, even a relatively inexperienced designer--or in Ranbaxy's case, a scientist--can extract the average effect of each factor on the experimental results. And thus reach reliable conclusions despite the large number of changing variables.
Here's why orthogonal tables are so important. At one end, the scientist has the choice of playing hit-and-miss and run a few experiments on the basis of factors he thinks are important. Or, if he is really conscientious, he can run the full brace of experiments in all possible permutations and combinations--but that would be like looking for a needle in a haystack.
However, with an orthogonal array, the scientist has to run just eight experiments in the order prescribed in the array, to reach virtually the same conclusions as he would have if he had performed upto 1024 experiments.
Says R S Chalapathi, a Ghaziabad-based DOE consultant: "When an instrument is played, it is rhythm which transforms the sound into music. Without rhythm it's just noise. Similarly, orthogonal arrays bring rhythm to the experimentation process."
So how exactly does Design Of Experiment work? There are seven steps in all:
One: Problem identification: Consider the manufacturing process for a piston, which has a high rejection rate of 30 per cent. Under DOE the root cause for the defect has to be identified first: is it shrinkage, porosity, cracks, or dents? To define the problem more precisely, a team of design engineers needs to undertake a Pareto analysis, use a cause-and-affect diagram and brainstorm the possible parameters which lead to the problem. Suppose the root defect is narrowed down to porosity in pistons.
Two: Selection of experimental design: The right orthogonal array is chosen depending on the number of factors (an element of a process which has an effect on the outcome) and levels (the values the factors will take), involved in the experiment.
In the case of rejected pistons for example, there could be four possible factors leading to porosity: mold temperature; mold design; riser design; and metal temperature. The team would now need to isolate two levels for each parameter--a high and a low--to run the experiments. So, metal temperature for example, can vary in the experiment at a low of 375 degree centigrade or a high of 500 degree centigrade. Given four factors at two levels each, the team would choose the L-8 orthogonal array.
Three: Conduct the experiments: Based on L-8, the team has to perform just eight experiments instead of the full brace of 16 experiments to track the defect. The team now runs the eight experiments prescribed according to the orthogonal array, and records the response. In the case of defective pistons, the response would be "percentage rejection"-- but the same DOE principle can be applied to track cycle time, yield, waste, or cost.
Four: Look For Signals: Under DOE, the resultant data speaks for itself through analysis of variance (ANOVA)--a statistical tool used to measure the impact of each individual factor on the variation in the final response. This step allows the team to prioritise which of the four factors have maximum impact in causing porosity. The impact of each factor is ranked and only those are chosen which are significant. Points out Chalapati, "This step invariably results in cost reduction."
Five: Determine Best Levels: The team now selects the optimum combination of factors and levels. In the piston example, feedback from the eight experiments could reveal that the optimum combination: temperature at 375 degrees instead of 500 degrees; mold design (new instead of old); higher metal temperature. This would then be the optimum combination, which yields the least porosity.
Interestingly, this combination of factors might well not be one of the eight experiments. Therefore, it is critical that right at the beginning, the team select the right factors, levels and parameters. Warns Chalapathi, "DOE will only give you results if your process understanding is complete."
Six: Conduct Confirmation Run: To reassure themselves that the optimum experiment designed by DOE does deliver, the team conducts the experiment. Comparing the actual rejection level with the theoretical rejection level soon proves the veracity of the new experiment--an error of more than one per cent means that some significant factor was missed and the whole process starts again. On the other hand a close match means that the team has figured out exactly what causes the porosity.
The last step? Says Chalapathi, "Congratulate the team!" For, by using DOE, it is the R&D lab which becomes a fertile ground for low cost, low time, and high efficiency products and production processes. And as Ranbaxy has discovered, in the crucible of competition, design of experiment is a handy catalyst for success.
Printer-friendly page
