On 9 December 2021, the WMLG held an online meeting to mark its 75th anniversary.
For this belated celebration of the 75th anniversary of the local group, Dr Tim Davis presented his take on an interesting paper by Jim Morrison (Morrison SJ (1957) 'The study of variability in engineering design', Applied Statistics, 6(2), pp133-138). The topic of statistical engineering echoes the origin of the group with the first talks being given by speakers from engineering and manufacturing industry, with in particular the first talk by Mr BP Dudding of General Electric on September 26, 1945.
After giving an overview of the history of Morrison's paper, including the follow up article by the same author no less than 41 years later, Tim discussed the details of the problem considered in the original paper which focuses on the manufacturing of glass beads and the associated variations. The objective in this paper is related to robustness in the sense that the underlying design must produce viable results despite the effects of noise factors. Glass beads production is an excellent example since there are significant variations in the associated process and the corresponding failure mode is two-sided, ie the volume of the glass bead can be either too small or too large for the final product to function properly.
Without careful analysis, one might think that the nominal value with highest variation is the one that will be the most problematic (and this is what GE was planning on doing before Morrison's study). However, this approach ignores the fact that the same variation in different nominal values might produce very different variations in the targeted quantity.
In the second part of his talk, Tim furthered the analysis in Morrison's paper by considering parameter design, ie by not only identifying which of the variations should be reduced but also changing the nominal values themselves to increase the overall robustness. This was not needed in Morrison's study since a simple way of reducing variations was identified. Parameter design requires assumptions on the relation between mean and variance for the nominal values as well as assumptions on the correlation between nominal values, and different models were considered and compared.
Tim concluded the talk by discussing the place of engineering within the RSS compared to other fields such as medicine and agriculture.
Questions and remarks:
- One comment was made regarding the difference in understanding and testing variations between Medicine and Engineering.
- One question was raised about adapting the approach to implicit objective functions. Tim mentioned the possibility of estimating such a function and highlighted the fact that only the gradient of the function needs to be known to analyse variations.
The talk was based on the article:
Davis, T. (2021). The study of variability in engineering design—An appreciation and a retrospective. Data-Centric Engineering, 2, E3. doi:10.1017/dce.2021.3