README.md 7.92 KB
Newer Older
1
# LabProcesses
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
2
This package contains an (programming- as well as connection-) interface to serve
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
3
as a base for the implementation of lab-process software. The first example of
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
4
5
6
7
8
9
10
11
an implementaiton of this interface is for the ball-and-beam process, which is
used in Lab1 FRTN35: frequency response analysis of the beam. The lab is implemented
in [BallAndBeam.jl](https://gitlab.control.lth.se/processes/BallAndBeam.jl), a
package that makes use of `LabProcesses.jl` to handle the communication with
the lab process and/or a simulated version thereof. This way, the code written
for frequency response analysis of the beam can be run on another process
implementing the same interface (or a simulated version) by changeing a single
line of code :)
12

Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
13
## Installation
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
14
15
16
17
1. Start julia by typing `julia` in a terminal, make sure the printed info says it's
`v0.6+` running. If not, visit [julialang.org](https://julialang.org/downloads/)
to get the latest release.
2. Install LabProcesses.jl using command `Pkg.clone("https://gitlab.control.lth.se/processes/LabProcesses.jl.git")` Lots of packages will now be installed, this will take some time. If this is your first time using Julia, you might have to run `julia> Pkg.init()` before you install any packages.
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
18
19

## How to implement a new process
20
21
1. Locate the file [interface.jl](https://gitlab.control.lth.se/processes/LabProcesses.jl/blob/master/src/interface.jl). When the package is installed, you find its directory under `~/.julia/v0.6/LabProcesses/`, if not, run `julia> Pkg.dir("LabProcesses")` to locate the directory.
(Alternatively, you can copy all definitions from [/interface_implementations/ballandbeam.jl](https://gitlab.control.lth.se/processes/LabProcesses.jl/blob/master/src/interface_implementations/ballandbeam.jl) instead. Maybe it's easier to work from an existing implementaiton.)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
22
23
2. Copy all function definitions.
3. Create a new file under `/interface_implementations` where you paste all the
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
24
copied definitions and implement them. See [/interface_implementations/ballandbeam.jl](https://gitlab.control.lth.se/processes/LabProcesses.jl/blob/master/src/interface_implementations/ballandbeam.jl) for an example.
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
25
4. Above all function implementations you must define the process type, e.g,
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
26
27
28
    ```julia
    struct BallAndBeam <: PhysicalProcess
        h::Float64
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
29
        bias::Float64
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
30
    end
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
31
    BallAndBeam() = BallAndBeam(0.01, 0.0) # Constructor with default value of sample time
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
32
    ```
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
33
Make sure you inherit from `PhysicalProcess` or `SimulatedProcess` as appropriate.
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
34
35
36
37
38
39
This type must contains fields that hold information about everything that is
relevant to a particular instance of the process. Different ballandbeam-process
have different biases, hence this must be stored. A simulated process would have
to keep track of its state etc. in order to implement the measure and control
methods. See [Types in julia documentation](https://docs.julialang.org/en/stable/manual/types/#Composite-Types-1)
for additional info regarding user defined types and (constructors)[https://docs.julialang.org/en/stable/manual/constructors/].
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
40
41
42
43
44
45
5. Documentation of all interface functions is available in the file [interface_documentation.jl](https://gitlab.control.lth.se/processes/LabProcesses.jl/blob/master/src/interface_documentation.jl)

## Control a process
The interface `AbstractProcess` defines the functions `control(P, u)` and `measure(P)`.
These functions can be used to implement your own control loops. A common loop
with a feedback controller and a feedforward filter on the reference is implemented
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
46
47
in the function [`run_control_2DOF`](@ref), where the user can supply $G_1$ and $G_4$
in the diagram below, with the process $P=G_2$.
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
48
![block diagram](docs/feedback4.png)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
49
50
51
52
53
54
55

The macro `@periodically` might come in handy if you want to implement your own loop.
Consider the following example, in which the loop body will be run periodically
with a sample time of `h` seconds.
```julia
for (i,t) = enumerate(0:h:duration)
    @periodically h begin
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
56
57
        y[i] = measure(P)
        r[i] = reference(t)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
58
        u[i] = calc_control(i,y,r)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
59
60
61
62
63
64
65
        control(P, u[i])
    end
end
```

Often one finds the need to implement a stateful controller, i.e., a function
that has a memory or state. To this end, the function [`sysfilter`](@ref) is
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
66
67
provided. This function is used to implement control loops where a signal is
filtered through a dynamical system, i.e., `U(z) = C(z)E(z)`.
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
68
69
Usage is demonstrated below, which is a simplified implementation of the block
diagram above (transfer function- and signal names corresponds to the figure).
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
70
```julia
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
71
72
stateG1 = init_sysfilter(G1)
stateG4 = init_sysfilter(G4)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
73
function control(i)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
74
75
76
    rf = sysfilter!(stateG4, G4, r)
    e  = rf-y
    u  = sysfilter!(stateG1, G1, e)
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
77
78
end
```
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
79
`G1` and `G4` must here be represented by [`StateSpace`](http://juliacontrol.github.io/ControlSystems.jl/latest/lib/constructors/#ControlSystems.ss) types from [`ControlSystems.jl`](https://github.com/JuliaControl/ControlSystems.jl).
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
80
`TransferFunction` types can easily be converted to a `StateSpace` by `Gss = ss(Gtf)`.
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
Continuous time systems can be discretized using `Gd = c2d(Gc, h)[1]`. (The sample time of a process is available through `h = sampletime(P)`.)


# How to implement a Simulated Process
## Linear process
This is very easy, just get a discrete time `StateSpace` model of your process
(if you have a transfer function, `Gss = ss(Gtf)` will do the trick, if you have continuous time, `Gd = c2d(Gc,h)[1]` is your friend).

You now have to implement the methods `control` and `measure` for your simulated type.
The implementation for `BeamSimulator` is shown below
```julia
function control(p::BeamSimulator, u)
    sysfilter!(p.state, p.sys, u)
end
measure(P) = vecdot(p.sys.C,p.state)
```
The `control` method accepts a control signal (`u`) and propagates the system state
(`p.state`) forward using the statespace model (`p.sys`) of the beam. The function
[`sysfilter!`](@ref) is familiar from the "Control" section above. What it does
is essentially
```julia
function sysfilter!(state, sys, input)
	state .= sys.A*state + sys.B*input
	output = sys.C*state + sys.D*input
end
```
hence, it just performs one iteration of
```math
Fredrik Bagge Carlson's avatar
Fredrik Bagge Carlson committed
109
x' = Ax + Bu \newline
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
y  = Cx + Du
```

The `measure` method performs the computation `y = Cx`, the reason for the call
to `vecdot` is that `vecdot` produces a scalar output, whereas `C*x` produces a
1-element `Matrix`. A scalar output is preferred in this case since the `Beam`
is SISO.

It should now be obvious which fields are required in the `BeamSimulator` type.
It must know which sample time it has been discretized with, as well as its
discrete-time system model. It must also remember the current state of the system.
This is not needed in a physical process since it kind of remembers its own state.
The full type specification for `BeamSimulator` is given below
```julia
struct BeamSimulator <: SimulatedProcess
    h::Float64
    state::Vector{Float64} # states defined by the file define_beam_system
    sys::StateSpace
    BeamSimulator() = new(0.01, init_sysfilter(beam_system), c2d(beam_system, 0.01)[1])
    BeamSimulator(h::Real) = new(Float64(h), init_sysfilter(beam_system), c2d(beam_system, h)[1])
end
```
It contains three fields and two inner constructors. The constructors initializes
the system state by calling `init_sysfilter`. The variable `beam_system` is already
defined outside the type specification.
One of the constructors provides a default value for the sample time, in case
the user is unsure about a reasonable value.
137
138
139
140
141
142
143
144
145
146

## Non-linear process
Your first option is to linearize the process and proceed like above.
Other options include
1. Make `control` perform forward Euler, i.e., `x' = f(x,u)*h` for a general
system model ``x' = f(x,u); y = g(x,u)`` and sample time ``h``.
2. Integrate the system model using some fancy method like Runge-Kutta. See
[DifferentialEquations.jl](http://docs.juliadiffeq.org/stable/types/discrete_types.html)
for discrete-time solving of ODEs (don't be discuraged, this is almost as simple as
forward Euler above).