This example shows how to use an Multiple MPC Controllers block and an Multiple Explicit MPC Controllers block to implement gain scheduled MPC control of a nonlinear plant.

The system is composed by two masses `M1`

and `M2`

connected to two springs k1 and k2 respectively. The collision is assumed completely inelastic. Mass M1 is pulled by a force F, which is the manipulated variable. The objective is to make mass M1's position y1 track a given reference r.

The dynamics are twofold: when the masses are detached, `M1`

moves freely. Otherwise, `M1`

and `M2`

move together. We assume that only `M1`

position and a contact sensor are available for feedback. The latter is used to trigger switching the MPC controllers. Note that the position and velocity of mass `M2`

are not controllable.

/-----\ k1 || F <--- | M1 |----/\/\/\-------------[|| wall || | |---/ || || k2 \-/ /----\ || wall||]--/\/\/\-------------------| M2 | || || \----/ || || || ----yeq2------------------ y1 ------ y2 ----------------yeq1----> y axis

The model is a simplified version of the model proposed in [1].

M1 = 1; % mass M2 = 5; % mass k1 = 1; % spring constant k2 = 0.1; % spring constant b1 = 0.3; % friction coefficient b2 = 0.8; % friction coefficient yeq1 = 10; % wall mount position yeq2 = -10; % wall mount position

The state-space plant models for this examples have the following input and output signals:

States: Position and velocity of mass

`M1`

Manipulated variable: Pulling force

`F`

Measured disturbance: Constant value of

`1`

which calibrates the spring force to the correct valueMeasured output: Position of mass

`M1`

Define the state-space model of `M1`

when the masses are not in contact.

```
A1 = [0 1;-k1/M1 -b1/M1];
B1 = [0 0;-1/M1 k1*yeq1/M1];
C1 = [1 0];
D1 = [0 0];
sys1 = ss(A1,B1,C1,D1);
sys1 = setmpcsignals(sys1,'MD',2);
```

-->Assuming unspecified input signals are manipulated variables.

Define the state-space model when the two masses are in contact.

```
A2 = [0 1;-(k1+k2)/(M1+M2) -(b1+b2)/(M1+M2)];
B2 = [0 0;-1/(M1+M2) (k1*yeq1+k2*yeq2)/(M1+M2)];
C2 = [1 0];
D2 = [0 0];
sys2 = ss(A2,B2,C2,D2);
sys2 = setmpcsignals(sys2,'MD',2);
```

-->Assuming unspecified input signals are manipulated variables.

Specify the cotnroller sample time `Ts`

, prediction horizon `p`

, and control horizn `m`

.

Ts = 0.2; p = 20; m = 1;

Design the first MPC controller for the case when mass `M1`

detaches from `M2`

.

MPC1 = mpc(sys1,Ts,p,m); MPC1.Weights.OV = 1;

-->The "Weights.ManipulatedVariables" property of "mpc" object is empty. Assuming default 0.00000. -->The "Weights.ManipulatedVariablesRate" property of "mpc" object is empty. Assuming default 0.10000. -->The "Weights.OutputVariables" property of "mpc" object is empty. Assuming default 1.00000.

Specify constraints on the manipulated variable.

MPC1.MV = struct('Min',0,'Max',30,'RateMin',-10,'RateMax',10);

Design the second MPC controller for the case when masses `M1`

and `M2`

are together.

MPC2 = mpc(sys2,Ts,p,m); MPC2.Weights.OV = 1;

-->The "Weights.ManipulatedVariables" property of "mpc" object is empty. Assuming default 0.00000. -->The "Weights.ManipulatedVariablesRate" property of "mpc" object is empty. Assuming default 0.10000. -->The "Weights.OutputVariables" property of "mpc" object is empty. Assuming default 1.00000.

Specify constraints on the manipulated variable.

MPC2.MV = struct('Min',0,'Max',30,'RateMin',-10,'RateMax',10);

Simulate gain-scheduled MPC control using the Multiple MPC Controllers block, which switches between `MPC1`

and `MPC2`

.

% To run this example, Simulink(R) is required. if ~mpcchecktoolboxinstalled('simulink') disp('Simulink is required to run this example.') return end

Open and configure the model.

y1initial = 0; % Initial position of M1 y2initial = 10; % Initial position of M2 mdl = 'mpc_switching'; open_system(mdl) if exist('animationmpc_switchoff','var') && animationmpc_switchoff close_system([mdl '/Animation']) clear animationmpc_switchoff end

Simulate the model.

```
open_system([mdl '/signals'])
sim(mdl)
MPC1saved = MPC1;
MPC2saved = MPC2;
```

-->Converting model to discrete time. -->Assuming output disturbance added to measured output channel #1 is integrated white noise. -->The "Model.Noise" property of the "mpc" object is empty. Assuming white noise on each measured output channel. -->Converting model to discrete time. -->Assuming output disturbance added to measured output channel #1 is integrated white noise. -->The "Model.Noise" property of the "mpc" object is empty. Assuming white noise on each measured output channel.

Using two controllers provides satisfactory performance under all conditions.

`MPC1`

OnlyRepeat the simulation assuming that the masses are never in contact; that is, using only controller `MPC1`

.

MPC1 = MPC1saved; MPC2 = MPC1saved; sim(mdl)

In this case, performance degrades whenever the two masses join.

`MPC2`

OnlyRepeat the simulation assuming that the masses are always in contact; that is, using only controller `MPC2`

.

MPC1 = MPC2saved; MPC2 = MPC2saved; sim(mdl)

In this case, performance degrades when the masses separate, causing the controller to apply excessive force.

bdclose(mdl); close(findobj('Tag','mpc_switching_demo'));

To reduce online computational effort, you can create an explicit MPC controller for each operating condition, and implement gain-scheduled explicit MPC control using the Multiple Explicit MPC Controllers block.

To create an explicit MPC controller, first define the operating ranges for the controller states, input signals, and reference signals. Create an explicit MPC range object using the corresponding traditional controller, `MPC1`

.

range = generateExplicitRange(MPC1saved);

Specify the ranges for the controller states. Both `MPC1`

and `MPC2`

contain states for:

The position and velocity of mass

`M1`

An integrator from the default output disturbance model

range.State.Min(:) = [-10;-8;-3]; range.State.Max(:) = [10;8;3];

When possible, use your knowledge of the plant to define the state ranges. However, it can be difficult when the state does not correspond to a physical parameter, such as for the output disturbance model state. In that case, collect range information using simulations with typical reference and disturbance signals. For this system, you can activate the optional `est.state`

outport of the Multiple MPC Controllers block, and view the estimated states using a scope. When simulating the controller responses, use a reference signal that covers the expected operating range.

Define the range for the reference signal. Select a reference range that is smaller than the `M1`

position range.

range.Reference.Min = -8; range.Reference.Max = 8;

Specify the manipulated variable range using the defined MV constraints.

range.ManipulatedVariable.Min = 0; range.ManipulatedVariable.Max = 30;

Define the range for the measured disturbance signal. Since the measured disturbance is constant, specify a small range around the constant value, `1`

.

range.MeasuredDisturbance.Min = 0.9; range.MeasuredDisturbance.Max = 1.1;

Create an explicit MPC controller that corresponds to `MPC1`

using the specified range object.

expMPC1 = generateExplicitMPC(MPC1saved,range);

Regions found / unexplored: 4/ 0

Create an explicit MPC controller that corresponds to `MPC2`

. Since `MPC1`

and `MPC2`

operate over the same state and input ranges, and have the same constraints, you can use the same range object.

expMPC2 = generateExplicitMPC(MPC2saved,range);

Regions found / unexplored: 5/ 0

In general, the explicit MPC ranges of different controllers may not match. For example, the controllers may have different constraints or state ranges. In such cases, create a separate explicit MPC range object for each controller.

It is good practice to validate the performance of each explicit MPC controller before implementing gain-scheduled explicit MPC. For example, to compare the performance of `MPC1`

and `expMPC1`

, simulate the closed-loop response of each controller using `sim`

.

r = [zeros(30,1); 5*ones(160,1); -5*ones(160,1)]; [Yimp,Timp,Uimp] = sim(MPC1saved,350,r,1); [Yexp,Texp,Uexp] = sim(expMPC1,350,r,1);

-->Converting model to discrete time. -->Assuming output disturbance added to measured output channel #1 is integrated white noise. -->The "Model.Noise" property of the "mpc" object is empty. Assuming white noise on each measured output channel.

Compare the plant output and manipulated variable sequences for the two controllers.

figure subplot(2,1,1) plot(Timp,Yimp,'b-',Texp,Yexp,'r--') grid on xlabel('Time (s)') ylabel('Output') title('Explicit MPC Validation') legend('Implicit MPC','Explicit MPC') subplot(2,1,2) plot(Timp,Uimp,'b-',Texp,Uexp,'r--') grid on ylabel('MV') xlabel('Time (s)')

The closed-loop responses and manipulated variable sequences of the implicit and explicit controllers match. Similarly, you can validate the performance of `expMPC2`

against that of `MPC2`

.

If the responses of the implicit and explicit controllers do not match, adjust the explicit MPC ranges, and create a new explicit MPC controller.

To implement gain-scheduled explicit MPC control, replace the Multiple MPC Controllers block with the Multiple Explicit MPC Controllers block.

```
expModel = 'mpc_switching_explicit';
open_system(expModel)
```

Run the simulation.

sim(expModel)

To view the simulation results, open the signals scope.

```
open_system([expModel '/signals'])
```

The gain-scheduled explicit MPC controllers provide the same performance as the gain-scheduled implicit MPC controllers.

[1] A. Bemporad, S. Di Cairano, I. V. Kolmanovsky, and D. Hrovat, "Hybrid modeling and control of a multibody magnetic actuator for automotive applications," in *Proc. 46th IEEE® Conf. on Decision and Control*, New Orleans, LA, 2007.

bdclose(expModel) close(findobj('Tag','mpc_switching_demo'))

Multiple Explicit MPC Controllers | Multiple MPC Controllers