[trustable-software] System vs component properties: Safety
Daniel Firth
dan.firth at codethink.co.uk
Mon Sep 10 13:25:39 BST 2018
On 10/09/2018 12:16, Shaun Mooney wrote:
> Hi,
>
> I totally agree. After studying the MIT work it's easy to see that it
> cuts out a lot of the main flaws with SIL based safety and making
> safety a systems design problem is clearly the way forward.
>
> Summarising all of the STPA material and creating a re-useable, easy
> to follow tool for a safety system seems like the best thing that we
> can be doing.
>
Paul has just kindly given me an overview of this approach to me and I'm
blown away.
Please let's make this capture tool principled from the outset. The
'reliability' of the capture tool is critical, and it being itself a
component that produces signals it should form part of the overall
model. For reliability of the capture tool I think the following
requirements are crucial.
1) No bugs that are detectable a priori via algebraic data types.
2) The internal logic be derived from a faithful algbraic representation
of control systems theory.
3) Ad infinitum scalability of that representation via mathematical
composition.
And possibly:
4) It should be possible to annotate freely, any component or any
topological construct of connected components, with any kind of typed
data, and declare 'coverings' arbitrarily.
Haskell can manage 1 and 3 (essentially for free), and from my beginner
understanding of this material a large chunk of 2 (potentially all of 2
down the road) with some planning. The correct way to achieve 2 in my
mind would be the categorification of control systems theory which
should transform *fairly* losslessly into Haskell. There has been some
prior art on the categorification of control systems/dynamical
systems/coverings[1][2].
A faithful typed representation off the bat would allow us to leverage
any and all work done on control systems in the academic sphere at very
low maintenance overhead, as it would more or less translate 1:1 from
white papers. (new analytic techniques, covering techniques, etc)
[1] https://arxiv.org/pdf/1803.05316.pdf
[2] https://www.emis.de/journals/ASUO/mathematics/pdf25/24f.pdf
br,
Dan
More information about the trustable-software
mailing list