{"id":3756,"date":"2014-05-30T15:27:25","date_gmt":"2014-05-30T20:27:25","guid":{"rendered":"https:\/\/blogs.mathworks.com\/seth\/?p=3756"},"modified":"2017-01-04T16:04:23","modified_gmt":"2017-01-04T21:04:23","slug":"plant-identification-using-the-pid-tuner","status":"publish","type":"post","link":"https:\/\/blogs.mathworks.com\/simulink\/2014\/05\/30\/plant-identification-using-the-pid-tuner\/","title":{"rendered":"Plant Identification Using the PID Tuner"},"content":{"rendered":"\r\n

In R2014a, system identification<\/a> capabilities have been added to the PID Tuner app<\/a> to create a plant model, allowing you to do the plant identification and controller tuning all in one app.<\/p>\r\n\r\n\r\n

I gave it a try with the new LEGO MINDSTORMS EV3 Support Package<\/a>, and I was really impressed by how easy it is to use.<\/p>\r\n\r\n

Let's see how it works.<\/p>\r\n\r\n

Acquiring Data<\/strong><\/p>\r\n\r\n

To identify a plant model, your first need experimental data. In my case, I created the following simple model that applies power to a motor and measures the resulting motion.<\/p>\r\n\r\n

\"Model<\/p>\r\n\r\n

Using External Mode and To Workspace<\/a> blocks, I acquire experimental data that will be used later.<\/p>\r\n\r\n

\"Acquired<\/p>\r\n\r\n

Control Model<\/strong><\/p>\r\n\r\n

I want to control the motor in position, so I create the following model:<\/p>\r\n\r\n

\"Control<\/p>\r\n\r\n

I open the dialog of the Discrete PID<\/a> block, I specify the type of controller I want, its sample time, and I click the Tune button to launch the PID Tuner app.<\/p>\r\n\r\n

\"Dialog<\/p>\r\n\r\n

Identifying the plant<\/strong><\/p>\r\n\r\n

As first guess, the PID Tuner tries to obtain a plant model by linearizing the blocks present in the model. Obviously, in this case this will not work because the Simulink model does not contain a plant.<\/p>\r\n\r\n

To identify a plant based on the data previously acquired, select the Identify New Plant<\/em> option:<\/p>\r\n\r\n

\"Identifying<\/p>\r\n\r\n

A Plant Identification tab will appear where you will be able to import the previously saved data.<\/p>\r\n\r\n

\"Import<\/p>\r\n\r\n

Theoretically, a motor like this one should be modeled using a second-order transfer function, with integrator (See this tutorial for an example<\/a>). However, based on past experience I know that the effect of the second pole is almost negligible for this Lego motor. In the Plant Identification tab I specify a one pole system with integrator and click the Auto Estimate<\/strong> button. As you can see, this does a pretty good job. I zoomed in on one of the transitions to show that the plant output is not exactly identical to the experimental data... which would be impossible.<\/p>\r\n\r\n

\"Identified<\/p>\r\n\r\n

Tuning the Controller<\/strong><\/p>\r\n\r\n

Once you are satisfied with the identified plant, click the Save Plant<\/strong> button and switch to the PID Tuner tab. Play with the sliders until you get a satisfying response, and once this is done click Update Block<\/strong> to apply the tuned gains to the controller block in your model.<\/p>\r\n\r\n

\"Tuned(Click to enlarge)<\/em><\/a><\/p>\r\n\r\n

Deploying the Controller<\/strong><\/p>\r\n\r\n

I tried running the control model on the EV3 and I was really impressed to see that the results were almost identical to what I could see in the PID Tuner:<\/p>\r\n\r\n

\"Experimental<\/p>\r\n\r\n

Now it's your turn<\/strong><\/p>\r\n\r\n

This shows only a small part of what this app can do. Here are a few additional resources for you to learn more:<\/p>\r\n\r\n