“All models are wrong, some are useful”
-George Box
Have you ever simulated anything? Built an analytical model of your system?
If so, you’ve made many choices.
One of the most important comes near the beginning:
What level of abstraction will best give me the answers I need?
Say we’re modeling a three phase motor drive circuit.
We could approach the problem at at least three different levels of abstraction:
Low Abstraction:
Calculate each MOSFET switching time down to the microsecond.
Medium Abstraction:
Ignore the PWM stuff, and just feed the motor with a set of clean sine waves.
High Abstraction:
Treat the motor as a DC torque source and flatten the AC waveforms to their RMS values.
How do we choose?
It depends on the questions you’re expecting your model to answer, and how much detail is included in the inputs:
High Detail:
Does your firmware have timing errors that may cause H-Bridge shoot through or reverse polarity?
Medium Detail:
Does your field oriented control code work? Do you get the torque you’re asking for?
Low Detail:
How does my vehicle’s drive cycle impact energy consumption over the course of a typical day?
Typically, you want to decrease model abstraction as your question’s detail level increases.
But why don’t we just run the Low Abstraction model all the time? It’s more accurate right?
That accuracy comes at a cost, which is typically longer run times. Low Abstraction models often run slower (sometimes much slower) than real time, so modeling a whole day’s worth of data could take a week.
But that same analysis could be run on a High Abstraction model in seconds.
The High Abstraction model is “more wrong” - but it’s much more useful for the low detail question.
However, It’s completely useless for a high detail question, so we’re forced to use the slower, more accurate Low Abstraction model.
-Brian Schoolcraft
Comments