We are approaching the end of this series, so the charts are becoming more specific. This category is mainly used in R&D and the more operational control of companies. After all, it shows the relationship between 2 or more process variables.
The "scatter plot", the simplest way to visualise relationships
This simple and singular example of a 'scatter' plot shows the relationship between process temperature (°C) and yield in a bio-process (kg).
However, it goes without saying that a causation test must be done first. Is the changing yield really a causal consequence of the changing temperature? You may assume so in this example.
A relationship can clearly be seen between the 2 parameters. Once this relationship is established, you can further analyse this relationship. For this, however, you need an extension to the "scatter" plot.
The "regression plot", the statistical approach to process relationships
- Modelling: Using the regression technique, you can determine the mathematical model that models the visualised relationship. It produces a formula that you do have to be careful with. This formula is reliable only in the area for which observations exist.
The example turns out to be a quadratic function: y=Ax²+Bx+C
- Prediction: Once the regression model is known, you can use it to forecast. With the right software (it can also be done manually), you can calculate at which temperature (x) you get the highest yield (y). In this case, that turns out to be 36.9 °C.
Steering: You can also use this knowledge to steer the process. In this example, you could set the process control so that the temperature is always between 36.4C° and 37.4C°. Now suppose that with the current installation it is not possible to steer more accurately, you would ensure anyway that the yield is always at the top of the curve and therefore always near the maximum value.
There is obviously much more to say on this subject. For this blog, I will stick to the simple application of relationship plots. But those who want to know more can follow the fantastically beautiful, statistical path of more complex regression analysis. This path will eventually lead to the wonderful world of “Design of Experiments.” Anyone in the lab or on the production floor who wants to optimise the process to unprecedented levels really needs to walk this path.
Again, I like to close with an anecdote that shows that scatter plots can also be useful in the office context. Not long ago, a 'greenbelt-to-be' very proudly showed me a graph. He had plotted sales volumes per customer against payment terms. Mind you, this is not about agreed payment terms, but the actual payment date counted from order date. The pattern of the relationship is not so much relevant in this particular example, but the customers outside the pattern are. The budding 'caller' had made an individual analysis for a handful of out-of-pattern customers. He could therefore explain in great detail how we should make adjustments for this select group of customers and thus get paid faster. And this without changing anything about our own organisation and the behaviour of the unaffected customers.
Next time, I will conclude this series with some personal reflections.
