Fullscreen HTML5Med Res (??MB)HTML4

While many readers will be familiar with electronic circuit simulation by programs like Spice or system design software such as LabVIEW, there are also simulation programs to test mechanical systems like motors. Simulink is a graphical programming language developed by MathWorks and it can be used to test Arduino control systems. By Karthik Srinivasan* Real-time system modelling with Arduino and Simulink T his practical demonstration involves using Simulink to analyse the operation of a standard radio control servo (Hitec HS-422) under Arduino control. Simulink can simulate faults and it allows the user to make alterations to the system to correct any faults in real time. Note, that the rest of this article assumes some familiarity with control theory, linear regression and related concepts. When designing any system, criteria such as gain and phase margins can ensure acceptable performance when there are slight changes in system dynamics. However, if system dynamics were to change significantly because of a component or sensor failure, the result can be less than optimal or even catastrophic. To ensure that these types of failures do not produce an unexpected result, it is important to detect these events as they happen. A real-time model lets you compare actual system measurements with predictions from the model and detect failures when the difference exceeds a certain threshold. This article details how to build the base of the model, and also how to ac82  Silicon Chip quire data from an Arduino device. You can find it at www.siliconchip. com.au/l/aaak and it includes a download of the source code. A video version of this article can be found here: www.siliconchip.com. au/l/aaal and another related video here: www.siliconchip.com.au/l/aaam For this demonstration we use a Hitec HS-422 servo (widely used in radio-controlled model cars, boats and planes) mounted on a motor driver which is connected to an Arduino Duemilanove board (an Arduino Uno board can be substituted). In Fig.1, we have the servo track a square wave reference angle and after 20 seconds we introduce a disturbance Fig.1: the predicted (yellow) and measured (mauve) motor angle. The measured motor angle does not go back to a rest position once the fault has been introduced into the system. siliconchip.com.au Fig.2: the Zero-Order Holds in the control algorithm provide a specified time-delay for the inputs provided, here we leave it at the default value of 1. The polynomial estimator performs the comparison between the measured and predicted motor angle, it then calculates the absolute error and original parameters. The data then needs to be sent to a host computer so it can be plotted. to the system, stopping the servo motor from tracking the reference angle. Since we want to be able to detect this change in behaviour while the servo motor is running, we program the Arduino Duemilanove board to perform this calculation. Fault detection To program the Arduino Duemilanove we build a two-part model in Simulink using blocks from the System Identification Toolbox (Fig.2). The first part of the model is the control algorithm and it uses motor angle measurements and a PID (proportional-integral-derivative) controller to send a voltage request to the servo motor which then tracks the reference angle position. The PID controller looks at the set-point value provided and compares it to the actual value of the process variable; in this case the reference angle of the motor. It then returns low if both values are the same, or high if they differ too much. The second model handles parameter identification and fault detection and this is where the maths gets heavy siliconchip.com.au (After all, it is from MathWorks!). It uses a “recursive polynomial estimator” block located in the System Identification Toolbox which has inputs for the input motor voltage and measured angle. This is configured as an ARMAX (autoregressive-moving-average exogenous) model that is used to estimate an ARMAX polynomial with the form: A(q) * y(t) = B(q) * u(t – nk) + C(q) * e(t) This lets us model noise and dynamics independently. Choosing the right parameters for our ARMAX model is part science, part trial and error. Since any DC motor can be modelled as a second-order differential equation, we choose two The inner workings of the model DC motor and what blocks are required to take data from an Arduino device over a serial port. January 2017  83 Figures 3 & 4: an ARMAX polynomial is used for comparison between the measured and calculated motor reference angle and the value of each parameter should be set as follows in the menus above. The “Output estimation error” selection is needed so that we can plot the difference between our results. for the number of poles [A(q)], two for B(q) and one for C(q) as shown in Fig.3. For the input delay (nk), we record how long the servo takes to respond to a step input and divide this number by the estimator sample time, providing us with a value of two (milliseconds) for our model. The input delay can be treated as phase shift or a time delay applied to the polynomial. In Fig.4, the recursive polynomial estimator block gives us the option to enable or disable parameter estimation. We use “Add enable port” to perform ‘online’ parameter estimation for the first 10 seconds of runtime – this provides time for the parameters to converge to their steady state values. Once this period has passed, the parameter estimation block stops updating the servo motor with new values; instead it uses the previously estimated values to calculate the next step. These estimated values are used to predict the motor angle for a given motor input voltage (under normal operating conditions) and are then compared to the measured motor angle. This is how the error for the motor system is calculated. Simulink provides a way to calculate a steady state error of the system. To do this we enable the error port which outputs the one-step-ahead prediction error (the difference between the measured motor angle and the predicted angle). We use a low-pass filtered version (ie, another pole) of the fault detection 84  Silicon Chip algorithm, which is implemented as a two-state StateFlow chart; which is a finite-state machine (Fig.5). The StateFlow chart sets the fault flag high when the filtered error is greater than the threshold value which is currently 1 and sets the flag low when 10 seconds have passed and the error is less than 1. This Simulink model is now loaded onto the Arduino Duemilanove board by using the “Run on Target” feature of Simulink. results. We have the parameter estimation algorithm calculate the approximate servo motor dynamics during the first 10 seconds of runtime before it reaches its steady state. After around 20 seconds have passed we introduce our disturbance into the servo motor, this causes the error value to shoot up and the fault detection algorithm sets its fault flag high. Once the disturbance is removed, the system returns to normal and the fault flag is once again set low. Result Building on this approach Everything is set to detect changes in the servo’s dynamics while it is running and in Fig.6 we can see our For this project, we use real-time estimation to detect faults as they occur in our system. Fig.5: the StateFlow diagram is pictured below and to the right is what the inside of the diagram looks like. siliconchip.com.au Extra Reading Fig.6: the top yellow trace is the predicted (or control) motor angle while the mauve trace is the measured angle. The plot below shows the fault prediction error (the difference between the predicted and measured values). Below that is the fault flag value; it steps high when the error rises above a set constant. Common applications of this are in adaptive control, where this technique is used to modify some controller based on changes to the system helping to maintain a required level of performance. For example, some radio control (RC) servo motors use adaptive control to correct the performance of its mechanism. An odd example would be a swing driven by a motor with an adaptive control system to help sustain periodic movement. After making sure the prototype model is correct, Simulink provides an easy way to generate code for your model, letting you deploy it to your target hardware, similar to how we loaded the software onto an Arduino Duemilanove board using the “Run on Target” feature. Contact and pricing A free 30 day licence for the MathWorks suite which includes Matlab, Simulink, StateFlow and more can be applied for on the MathWorks website (https://au.mathworks.com/) by clicking the Trial Software link at the bottom of the page. Pricing details for each type of licence can also be found on their website and more example projects can be found at http://makerzone.mathworks.com/ SC * MathWorks https://au.mathworks.com/help/ instrument/direct-interface-communication-in-simulink.html – using serial communications in Simulink http://au.mathworks.com/help/ supportpkg/arduino/ug/run-modelon-arduino-hardware.html – running Simulink programs on Arduino https://au.mathworks.com/help/ ident/ref/recursivepolynomialmodelestimator.html – recursive polynomial estimator block http://au.mathworks.com/help/ident/ ref/armax.html – ARMAX function https://en.wikipedia.org/wiki/Autoregressive-moving-average_model – ARMA details http://au.mathworks.com/help/ident/ examples/comparison-of-variousmodel-identification-methods.html – model comparisons www.facstaff.bucknell.edu/mastascu/eControlHTML/Design/Perf1SSE. htm – steady state error details https://au.mathworks.com/help/ stateflow/gs/anatomy-of-a-stateflow-chart.html – StateFlow chart example www.landau-adaptivecontrol.org/ Slides%20Ch1.pdf – slides on adaptive control compared to conventional and robust methods w w w. g o o g l e . c o m . a u / p a t e n t s / US5833545 – adaptive control swing patent http://au.mathworks.com/matlabcentral/fileexchange/44416-simpleadaptive-control-example – example program using adaptive control Glossary Exogenous variable: variables independent of the process being measured; an example could be a shift in consumer confidence leading to lower sales, or a natural disaster affecting energy production. Online: the model or plant being run during the runtime of the system being tested, ergo in real-time. Pole: poles and zeros are terms applied to mathematical transfer functions. In electronic circuitry, every filter network has a time constant, a rolloff frequency and associated slope and a phase characteristic. Each filter network, which may be as simple as an RC low-pass network, is referred to as a “pole”. The filter network which is the main determinant of a circuit’s frequency response is referred to as the “dominant” pole. Plant: a term from control theory referring to the combination of an input and output signal with some component that is responsible for controlling a mechanism or system. Polynomial: a polynomial is an expression that can be expressed in the form: anxn + an-1xn-1 + … + a1x + a0 where a0, …, an are constants, xn are unknown variables and n > 0. Servo motor: a servo motor contains a DC motor, gear reduction unit, a position-sensing device (normally a potentiometer) and a control circuit. A servo receives some control signal that represents the desired output motor angle of the shaft and applies power to the motor until that angle is reached; the position-sensing device is used to determine which way the motor should move. Transfer Function: a mathematical function relating the output or response of a system to its input, eg, a filter circuit. siliconchip.com.au January 2017  85