WIP: Delayed LTI #178

mfalt · 2019-01-24T17:58:24Z

Do not pull yet, still WIP

List of problems, feel free to add to it:

Should be resolved:

More extensive tests needed
Some discussions on the code below

Would be nice to have:

lsim doesn't handle D22 term
lsim interpolates x to get y (with D21 != 0), which is the only inhibitor for great accuracy (it is still very much usable, and x is correct at the grid points)
impulse: This is quite tricky, the first impulse can be handled, but if there is a direct term D21 to the delay, then we have to some way "store" the impulse train in there. I am not even sure how to begin to think about D22 != 0
Some sort of approximation of the delays would be useful to implement, e.g. for use with the design tools

src/types/DelayLtiSystem.jl

mfalt · 2019-01-24T18:03:14Z

src/types/DelayLtiSystem.jl

+# TODO: Think through these promotions and conversions
+Base.promote_rule(::Type{<:StateSpace{T1}}, ::Type{DelayLtiSystem{T2}}) where {T1<:Number,T2<:Number} = DelayLtiSystem{promote_type(T1,T2)}
+Base.promote_rule(::Type{AbstractMatrix{T1}}, ::Type{DelayLtiSystem{T2}}) where {T1<:Number,T2<:Number} = DelayLtiSystem{promote_type(T1,T2)}
+Base.promote_rule(::Type{T1}, ::Type{DelayLtiSystem{T2}}) where {T1<:Number,T2<:Number} = DelayLtiSystem{promote_type(T1,T2)}


These should probably be caught somewhere else by
Type{AbstractMatrix}, ::Type{<:LTISystem}
and
Type{<:Number}, ::Type{<:LTISystem}

mfalt · 2019-01-24T18:08:14Z

src/types/DelayLtiSystem.jl

+    DelayLtiSystem(psys_new.P, Tau_new)
+end
+
+function delay(tau::Real, T::Type{<:Number}=Float64)


Is this the canonical way, or maybe we should have:

delay(tau::T) where T<:Real

and

delay(::Type{T}, tau::Real) where T<:Number

Yes, any of those options are probably better.

Is delay a reasonable name for the function? It seems quite appropriate, but another option would be exp(-tau*tf("s"). However this transfer function notation is a bit unfortunate since time delays and transfer functions tend to be numerically troublesome.

mfalt · 2019-01-24T18:14:24Z

src/types/Lti.jl

@@ -4,6 +4,9 @@ abstract type LTISystem <: AbstractSystem end
 *(sys1::LTISystem, sys2::LTISystem) = *(promote(sys1, sys2)...)
 /(sys1::LTISystem, sys2::LTISystem) = /(promote(sys1, sys2)...)

+feedback(sys1::Union{LTISystem,Number,AbstractMatrix{<:Number}},
+         sys2::Union{LTISystem,Number,AbstractMatrix{<:Number}}) = feedback(promote(sys1, sys2)...)
+


We should think about being consistent here, but that discussion probably belongs elsewehere.
For /(sys1::LTISystem, sys2::LTISystem) = /(promote(sys1, sys2)...)
we dont allow matrices and scalars (and assume the specific implementation will solve it), but for
feedback we provide a fallback here.

Yes, I think there are many places where more we should be more consistent.

mfalt · 2019-01-24T18:18:01Z

src/types/PartionedStateSpace.jl

+
+
+# There should already exist a function like this somewhare?
+function blkdiag(A1::Matrix{T1}, A2::Matrix{T2}) where {T1<:Number, T2<:Number}


AbstractMatrix maybe for future-proofing

Although we would have to figure out a way of constructing the new matrix to be the promotetype, which might be tricky. So maybe we skip that

src/delay_systems.jl

coveralls · 2019-01-24T21:54:47Z

Coverage increased (+2.07%) to 58.087% when pulling 96acc55 on delayed_lti into 6312567 on master.

example/delayed_lti_system.jl

example/delayed_lti_timeresp.jl

baggepinnen · 2019-01-31T10:45:35Z

example/delayed_lti_timeresp.jl

+sys = feedback(1.0, ss(-1.0, 2, 1, 0) * (delay(2.0) + delay(3.0) + delay(2.5)))
+
+t = 0:0.02:8
+@time t, x, y = lsim(sys, t; u=t->[t>=0 ? 1.0 : 0.0])


Suggested change

@time t, x, y = lsim(sys, t; u=t->[t>=0 ? 1.0 : 0.0])

t, x, y = lsim(sys, t; u=t->[t>=0 ? 1.0 : 0.0])

gues this was a leftover from your benchmarking?

baggepinnen · 2019-01-31T10:46:26Z

src/ControlSystems.jl

@@ -75,9 +78,11 @@ export  LTISystem,


 # QUESTION: are these used? LaTeXStrings, Requires, IterTools
-using Polynomials, OrdinaryDiffEq, Plots, LaTeXStrings, LinearAlgebra
+using Polynomials, Plots, LaTeXStrings, LinearAlgebra
+using DifferentialEquations, OrdinaryDiffEq, DelayDiffEq


Isn't DifferentialEquations an umbrella that exports everything else?

src/delay_systems.jl

baggepinnen · 2019-01-31T10:51:30Z

src/types/DelayLtiSystem.jl

+    ny = noutputs(sys) - length(Tau)
+
+    if nu < 0  || ny < 0
+        error("Time vector is too long.")


Suggested change

error("Time vector is too long.")

throw(ArgumentError("Time vector is too long."))

What is the difference here?

An argument error is more informative, and people doing error handling in applications sometimes rely on the proper type of error being thrown. See https://discourse.julialang.org/t/error-vs-throw-argumenterror/18915

Being more specific also allows you to write tests with @test_throws that don’t just pass spuriously if the same type of error is thrown in some other part of the code under test.

src/types/PartionedStateSpace.jl

Delay lti2

olof3 · 2019-02-04T13:56:05Z

src/types/DelayLtiSystem.jl

+    ssold = sys.P.P
+    # Add to direct term from input to output
+    new_D = ssold.D
+    new_D[1:ny, 1:nu] .+= n


Don't do broadcast

…jl into pr-178/JuliaControl/delayed_lti

baggepinnen · 2019-05-22T05:36:11Z

I rebased this on master, I think I managed to pull it off without screwing up.

olof3 added 7 commits January 22, 2019 10:03

WIP: Delayed LTI systems.

fcbf840

Delayed LTI updates.

3eff27c

Updates.

4f33725

Updates to delayed LTI.

3d774b7

Updates delayed LTI systems.

9af3b83

Updates delayed LTI systems.

a51cc54

Minor typo.

2a574aa

mfalt commented Jan 24, 2019

View reviewed changes

src/types/DelayLtiSystem.jl Outdated Show resolved Hide resolved

mfalt commented Jan 24, 2019

View reviewed changes

src/delay_systems.jl Outdated Show resolved Hide resolved

baggepinnen reviewed Jan 31, 2019

View reviewed changes

mfalt and others added 4 commits February 1, 2019 21:07

WIP delayed lsim

5da7529

Working version of delay systems

cc0deec

Minor bug in test, and removed prints

eba8e5a

Merge pull request #191 from JuliaControl/delay_lti2

baac22d

Delay lti2

olof3 reviewed Feb 4, 2019

View reviewed changes

mfalt and others added 10 commits March 25, 2019 20:09

Added impulse to delay lsim

1a785d0

WIP: Delayed LTI systems.

12ce306

Delayed LTI updates.

45d0a3c

Updates.

ef6ba37

Updates to delayed LTI.

26e7e98

Updates delayed LTI systems.

39da60e

Updates delayed LTI systems.

9307374

Minor typo.

66d01a6

WIP delayed lsim

b373567

Working version of delay systems

eca6598

mfalt and others added 3 commits May 22, 2019 07:26

Minor bug in test, and removed prints

e94ca79

Added impulse to delay lsim

d64468e

Merge branch 'delayed_lti' of github.com:JuliaControl/ControlSystems.…

55fbe3e

…jl into pr-178/JuliaControl/delayed_lti

baggepinnen and others added 9 commits May 22, 2019 09:28

clean up rebase

b842af5

Update src/delay_systems.jl

834edf6

Apply suggestions from code review

e504b35

Replace function wrapper with @nospecialize

cd0a32f

Use builtin ODE interpolator instead of Interpolations

488d8c2

add corner case constructor

e9b9d8c

Forward kwargs to solver

cff8e69

increase tol and clean code

a3a9e6c

Merge branch 'master' into delayed_lti

96acc55

mfalt merged commit 9e5a846 into master May 22, 2019

baggepinnen deleted the delayed_lti branch September 24, 2020 06:53

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

WIP: Delayed LTI #178

WIP: Delayed LTI #178

mfalt commented Jan 24, 2019 •

edited by baggepinnen

Loading

mfalt Jan 24, 2019

mfalt Jan 24, 2019 •

edited

Loading

olof3 Jan 31, 2019

olof3 Jan 31, 2019

mfalt Jan 24, 2019

olof3 Jan 31, 2019

mfalt Jan 24, 2019

mfalt Jan 24, 2019

coveralls commented Jan 24, 2019 •

edited

Loading

baggepinnen Jan 31, 2019

baggepinnen Jan 31, 2019

baggepinnen Jan 31, 2019

mfalt Jan 31, 2019

baggepinnen Feb 1, 2019

baggepinnen Feb 1, 2019

olof3 Feb 4, 2019

baggepinnen commented May 22, 2019



		# There should already exist a function like this somewhare?
		function blkdiag(A1::Matrix{T1}, A2::Matrix{T2}) where {T1<:Number, T2<:Number}

	@time t, x, y = lsim(sys, t; u=t->[t>=0 ? 1.0 : 0.0])
	t, x, y = lsim(sys, t; u=t->[t>=0 ? 1.0 : 0.0])

	error("Time vector is too long.")
	throw(ArgumentError("Time vector is too long."))

WIP: Delayed LTI #178

WIP: Delayed LTI #178

Conversation

mfalt commented Jan 24, 2019 • edited by baggepinnen Loading

Choose a reason for hiding this comment

mfalt Jan 24, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

coveralls commented Jan 24, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

baggepinnen commented May 22, 2019

mfalt commented Jan 24, 2019 •

edited by baggepinnen

Loading

mfalt Jan 24, 2019 •

edited

Loading

coveralls commented Jan 24, 2019 •

edited

Loading