11/24/2019

Ali M. Sahlodin
Department of Chemical Engineering
AmirKabir University of Technology
1397 S.H

 Elements of hybrid models

 Types of events and model switches
 Event detection

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 2

1
 11/24/2019

 Event detection: confirm an event has occurred

in the past or current integration step
 Event location: find the time of the event
 What if multiple events occur in the same
integration step?
 Find the earliest one

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 3

 Event time: report the first point the event is detected.

 Will not give the exact event time in general
 Timing error: depending on step size

timing error

t*

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 4

2
 11/24/2019

Multi-transition system

h=0.001

h=0.003

h=0.03
h=0.3

Sequence of modes is different for each step size!

different trajectories

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 5

 Recall error estimate Err  M || x n1  x (0)

n 1 || 1
 Cuts steps until Err is less than the tolerance
 time event approximated
x Trajectory after event

Trajectory before event

Corrector failure due to large Predictor polynomial

error estimates, resulting in
repeated step cuts

t* t

Cons
•Repeated corrector failures  inefficiency issues
•Possibly total failure of the solver
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 6

3
 11/24/2019

1. Lock the current mode until a successful integration

step is taken.
2. Check if the any state transition has occurred during
the previous step.
3. Find exact location of the event.
Trajectory after event
x
Trajectory
before event
Discontinuity locked
simulation

tk t* tk+1 t
Any assumption?

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 7

 Zero-crossing of discontinuity function  y,u, t )  0

g (x, x,
x(t)

tk t* tk+1 t
g(t)

g (tk ) g (tk 1 )  0
tk t* tk+1 t

Can this condition always detect the event(s)?

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 8

4
 11/24/2019

 Rudimentary Approach x(t)

1. Estimate event time tˆ* from the

values of g(t) at tk and tk+1.
*
2. Integrate from tk to tˆ .
*
3. Evaluate g (tˆ ) . tk t* tk+1 t
* * g(t)
4. If g (tˆ )  0 , update tˆ . Go to
Step 2. Otherwise stop.
tk t* tk+1 t
Pro
•Useful when g(t) is not available in closed form.
Con
•Inefficient due to repeated integrations.
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 9

 Root-finding with interpolating polynomials

 Multistep methods (e.g., BDF)
 y,u, t) = 0
f(x, x,

Interpolating polynomials x p (t ), x p (t ), y p (t )
Construct interpolating polynomial for discontinuity function
Newton method
p p p
g (x , x , y ,u, t )  0 t*
Accuracy of the polynomial impacts accuracy of event location found.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 10

5
 11/24/2019

 Improveaccuracy of the polynomial

 Append discontinuity function to the DAE system

 y,u(t ), t )  0
f (x, x, Now under error control
z (t )  g (x, x,
 y,u(t ), t ) mechanisms of solver

Any change in the DAE index?

Any impact on the DAE solution?

Solve z p (t )  0
t  [tk , tk 1 ] t*
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 11