I’m sure that most of you will have made paper aeroplanes when you were children. And you have thrown them into the air in an attempt to keep them flying as long as possible before they fall to the ground. I’m also sure that you realised that by changing the type of paper and the shape and dimensions of different parts you succeeded in improving the flight. You were carrying out design of experiments, or DOE.
In fact, in many DOE courses these experiments are used as a practical part of the training. Some of you will have done courses on DOE or Taguchi methodology and you will have had to do this exercise. Another similar exercise is to throw marbles of different sizes and from different heights on a floor covered in sand and measure the craters that are produced, simulating the fall of meteorites on a planet. To make it more fun you can place small dinosaurs in the sand and emulate the conditions that led to the extinction of dinosaurs on earth.
By doing random experiments we can achieve improvements, but a robust design requires the use of a methodology. We must know how to select the variables that we are going to use in the experiments. These variables include those that can be controlled to a very great extent, such as the parameters of a machine or some design features. But there are other more random variables, also known as noise, which we cannot control, but which must be taken into account in the experiments. In the case of paper aeroplanes, a random variable might be the wind, while the type of paper and the lengths of the different parts are variables which can be controlled.
The objective of the DOE in this case is to optimise the controlled variables so that we find which combination gives us the best result (which in the example of the aeroplane is the longest flight time). In the case of the noise variables, however, what we are looking for is which combination of the rest of the variables is the one that causes the least variation in the result (in the example it would be the combination of paper type or lengths that is least affected by the variation in the wind). A robust process design is not always the one that gives us the best result, but rather the one with the least variation in response to changes in the uncontrolled variables.
A typical problem in DOE is making the mistake that the effects add up. A classic example given in the DOE courses is that of alcohol and medicines. If a person is suffering from discomfort and takes a medicine it produces well-being, and if they drink alcohol it produces euphoria; but doing both does not necessarily result in feeling even better. These are called interactions: the opposing effects of two variables that separately improve things, but which together have the opposite effect.
Another difficulty in DOE is the time required when we have multiple variables and combinations. The solution to this problem was provided by Taguchi and involves simplifying the choice of combinations, since it does not require us to try all the possibilities but only a number of them – well chosen – that will provide us with a solution very similar to the one that would be given if we tried all the different possibilities.
Analysing the results well and reaching reliable conclusions requires a good knowledge of the methodology and only in this way will we be able to produce robust engineering designs. Computer programs like Minitab help us perform these complex statistical calculations so that we do not get lost in mathematics and can focus on understanding the data and designing the best combination that will lead to success in the experiment.
As electronics manufacturers, at IKOR we have many opportunities to apply the design of experiments (DOE) to improve our processes. If we want to achieve good soldering (reducing voids and micro solder balls, improving wetting and reliability in the long term or in exterior climatic conditions) or if we want to obtain circuits with high levels of cleanliness (eliminating flux residues or reducing ionic contamination).
To identify which controlled variables are going to have an influence, such as solder alloys, reflow times and temperatures, types of PCB finishes, screen printing screen mesh, etc.; and the uncontrolled variables, or noise, such as the variability of the components or the curing of the solder masks on the PCBs. To choose the values of the variables that best suit us and the matrix of combinations that can give us the most information with the least effort and time; to test, analyse the results, choose the best combinations and confirm these with new experiments. The process is lengthy, but if in the end we achieve the sought after improvements the effort will have been worth it.