2. Outline
Design procedure – baby steps
Pseudo code
Fuzzy rules
Simulink / matlab interaction
Engine model
• System states
Driving schedule
NNet size and setup
Compare to original
• RPM fall ringing, overrev, stall
Calculations of original gains vs. final gains (avg)
80% of time in MATLAB design
Show non-linearity – throttle
Engine load constant?
Relationship to original presentation – diagrams, theory, approach, etc.
The fuzzy rule set: Feedback is analyzed in terms of rise time, settling time, percent overshoot, Mean Final Error, and whether it is ringing or sawtoothed.
Suggested changes are summed and scaled to yield crisp suggested gain changes.
First we begin by observing the plant and suggest better gains for the next time we face this situation.
This is a test of applying the gain rules to a system for several iterations, showing the final response. The search is reset when it stalls to avoid getting stuck in a sub-optimal local minimum.
Note that the bias is toward stability over performance.
Use the N-Net to choose controller gains for the current input based on past training from the observer.
Conventional PI control with fixed gains -- response to our test input series.