85 lines
3.5 KiB
Markdown
85 lines
3.5 KiB
Markdown
---
|
|
title: Scenario and event simulation
|
|
tags: simulation, scenario, discrete-event, montecarlo, validation
|
|
summary: What a simulation of a spacecraft system can and cannot tell you, and why the number it produces is not the answer.
|
|
---
|
|
|
|
# Scenario and event simulation
|
|
|
|
A simulation of a spacecraft system is a model of a model. What it is good
|
|
for is comparison between designs and exploration of a parameter space; what
|
|
it is bad for is certifying a single number.
|
|
|
|
## Definition
|
|
|
|
A **scenario** is a specified set of initial conditions and events: the
|
|
orbit, the failure mode, the environment, and the period simulated. Two runs
|
|
with the same scenario and different implementation are a comparison.
|
|
|
|
A **discrete event** is an instant at which something changes state: a
|
|
command is executed, a sensor reports, a component fails. An event-driven
|
|
simulation advances from one such instant to the next rather than in fixed
|
|
time steps, and its accuracy depends on how the events are scheduled rather
|
|
than on a step size.
|
|
|
|
**Monte Carlo** repeats a scenario with sampled parameters to obtain a
|
|
distribution rather than a value.
|
|
|
|
## Mathematics
|
|
|
|
### What the output means
|
|
|
|
A simulation result is a value of a quantity that was computed under
|
|
assumptions. The uncertainty in it has three parts, and they are usually
|
|
very different in size:
|
|
|
|
- The uncertainty of the model, which is the largest and is the reason to
|
|
simulate rather than to measure.
|
|
- The sampling uncertainty of the run, which shrinks as the square root of
|
|
the number of runs and can be estimated.
|
|
- The numerical error, which shrinks with refinement.
|
|
|
|
Reporting the second as if it were the first is the characteristic failure
|
|
of simulation work, and it is why a Monte Carlo figure without a stated model
|
|
assumption is not a result.
|
|
|
|
### Time in an event-driven model
|
|
|
|
The state is constant between events, so a propagation or an interval can be
|
|
computed analytically rather than stepped. This makes an event-driven
|
|
simulation of an orbital system both faster and more accurate than a fixed
|
|
step at the same work, and it is why the choice of when an event fires is a
|
|
modelling decision rather than an implementation detail.
|
|
|
|
## Unit conventions
|
|
|
|
The units of the model. The failure that matters is a scenario specified in
|
|
different units from the implementation, which produces a plausible number
|
|
that is wrong by a power of a thousand.
|
|
|
|
## Applicability
|
|
|
|
Simulation answers comparative questions well: which of two station keep
|
|
strategies costs less propellant over five years, given the same orbit
|
|
determination quality. It answers absolute questions badly, because the
|
|
result carries the model's error and no amount of runs removes it.
|
|
|
|
The strongest use is checking that a design does not have a failure the
|
|
analysts missed, by simulating the failure rather than by analysing it. The
|
|
weakest use is producing a number for a report, which is where the
|
|
distinction between the three uncertainties above usually gets lost.
|
|
|
|
Validation is the part that is skipped and that matters: a simulation is
|
|
credible to the extent that it reproduces something already measured. Where
|
|
that has not been done, the result is a hypothesis with decimals.
|
|
|
|
## Standards
|
|
|
|
- None required. Simulation practice is not standardised, and the reporting
|
|
of uncertainty is where the discipline is usually lost.
|
|
|
|
## See also
|
|
|
|
- [[numerical-methods]] for the numerical error that a simulation carries
|
|
- [[operations-planning]] for a planning question simulation answers well
|
|
- [[maneuvers]] for the model most simulations are built from |