DesignWorldDesignWorldNovember 2009 www.designworldonline.com

ALSO INSIDE:

SENSOR NETWORKS:Are wireless sensornetworks really readyfor industry?

Page 54

SENSORS:Acousticresonance: theheartbeat oftouch sensors

Page 70

FLUID POWER:Diaphragm pumps bydesign

Page 62

Motion Control:

Integratemodel-basedmachine design withmotion control Page 50

nov. dw cover-2:Layout 1 5/10/10 9:16 AM Page cov11

Reprinted from Design World for Kollmorgen, © 2010 WTWH Media, Inc.

Reprinted from Design World for Kollmorgen, © 2010 WTWH Media, Inc.

mented in computer code. Thisprocess can be hindered by the dif-ficulty of finding experts in controlsystem algorithms. It’s also veryeasy to make, and very difficult todetect, mistakes in hand-writtencode, especially during the earlyphases of development effort. Chances are the handwritten codemay only make sense to the person who wrote it, which maycreate communication difficulties in larger projects.

The next step is to simulate the operation of the system. However, writing the equations of motion to model all of theinteractions of subsystems can be complex. Furthermore, software validation normally can’t be addressed until late in thedevelopment cycle, so errors are usually not detected until thecode is run on the actual hardware. After each iteration, the design must be recoded and the source code recompiled, rerun, and debugged, which increases development time andexpense. In addition, when the code is finally ready for produc-tion, it must normally be recoded to run on the target processor.

Advantages of model-based designModel-based design offers an alternative to the traditional

approach to control system design. It provides an environmentto develop an executable specification in the form of softwarecode that incorporates all of the key elements of the controllogic and mechanical and electrical design.

A typical system-level model uses blocks that represent mathematical operations between input and output

Two factors have made the design of motion control systems in machines more challenging. One is that mechanical improvements have made it possible formachines to operate at much higher speeds than everbefore. But the resulting mechanical problems such as

resonance and vibration control have become a much greaterconcern. Solving these complex problems often requires in-depth knowledge of control systems as well as mechanicaland electrical systems. The other factor is the loss of engineerswith extensive control system design expertise due either to lay-offs or company restructuring. The problem is then how to solvethese increasingly complex problems with too few people andlimited time and funds.

A promising new technique called model-based design integrates the machine design function with the motion controlsystem. Model-based design makes it easier for engineers withlimited expertise in control systems to design machines andsolve the problems that arise. It generates executable code basedon the control logic and the mechanical and electrical specifi-cations of the system in a model-based environment, overcom-ing the limitations of more traditional approaches to controlsystem design.

Traditional approachThe traditional approach to designing complex control

systems involves first writing text-based specifications that define the requirements in as much detail as possible. Thesespecifications are then used to design a control strategy imple-

50 DESIGN WORLD NOVEMBER 2009 www.designworldonline.com

MOTION CONTROL

Integratemodel-based

machine designwith motion control

Integratemodel-based

machine designwith motion control

Model-based design methods cut the time andcost of building and test-ing motion control sys-tems.

Robert SteeleChief Technical OfficerKollmorgenRadford, Va.

Tony LennonIndustry ManagerIndustry Marketing, Industrial Automation and MachineryThe MathWorks, Inc.Natick, Mass.

Engineers can now build a control system model, debug and test it in software, and compile and download themodel to a motion controller,saving valuable time and cost.

Reprinted from Design World for Kollmorgen, © 2010 WTWH Media, Inc.

of the model, such as adding noise in thevoltage or temperature and magnetic satura-tion effects. You can further increase modelfidelity by replacing approximate mathemat-ical representations of the mechanical systemwith blocks that represent mechanical bodiesand linkages translated automatically from aCAD file. You can also use the model to simulate operation of the control system insoftware. This approach makes it possible totest a wide range of alternative design concepts without investing time and moneyin writing code or building hardware.

Until recently, a limitation of model-baseddesign has been that the pre-configuredblocks used to construct models were

available only for standard control algorithms such as PIV (Proportional-Integral-Velocity) and PID (Proportional-Inte-

gral-Derivative). This meantthe technology could onlybe used on less sophisti-cated control systems. Inaddition, code could not beautomatically generated forleading-edge hardware.Newer graphical buildingblocks make it possible tomodel more complex con-trol systems and downloadthem to RTOS (Real-TimeOperating System) environ-ments used on the most advanced motion controlsystems. These standardfunction blocks reduce the

need for spec i a l i z ed so f twa re programming expertise, letting mechanical engineers take control of the design process.

Newer functionblocks can also sup-port complex gearingand following meth-ods such as gainswitching, vibration control, and multipleinput multiple output(MIMO) plant mod-els. Combined withdata capture, loggingand visualizationtools, these functionblocks integrate me-

chanical, I/O and software data into one measurement environment. Plant model output parameters, such as motor drive and

positional feedback, can be used to evaluate the controller, making it simple to identify and adjust physical parameters suchas mass, length, and capacitance which can cause instability.You can predict the move performance, ringing and settlingtime, and motor and drive sizing of alternative control system

signals. For example, a model can contain a block representinga motor. Initially, the model may simply take a voltage inputand convert it to an output torque. As the design process continues, additional detail can be added to increase the fidelity

www.designworldonline.com NOVEMBER 2009 DESIGN WORLD 51

The updated controller uses position, velocity, and torque pre-filtersto prevent resonance.

The original control system for the pick-and-place machine was asimple PIV controller.

In this pick-and-place machine, increased gantry speeds led to anincrease in reaction forces, which caused excessive vibration.

The model-based design approach also protects intellectualproperty by eliminating the need to disclose control strategiesto vendors, while making them nearly impossible to reverse-engineer. The control system can be tested using hardware-in-the-loop testing with control algorithms expressed in code on atarget microprocessor, and the plant model expressed in codeon a real-time system. The next step is usually prototyping withthe control algorithms on a real-time system using real hardware. DW

Kollmorgen1.540.633.3545contactus@kollmorgen.comwww.kollmorgen.com

MathWorkswww.mathworks.com

concepts before prototyping the design. Test cases can be embedded in the model to check each design iteration againstrequirements and find and correct any mistakes.

Once the control strategy is developed and tested in simula-tion, the code algorithms are converted from continuous (analog) to discrete (digital) and code is automatically generatedfor the motion controller. This eliminates the need for code-writing experts, prevents the introduction of coding errors,and saves time.

52 DESIGN WORLD NOVEMBER 2009 www.designworldonline.com

MOTION CONTROL

A real-world example

Here’s how a model-based design approach with amechatronics toolkit solved one machine builder’s

control system problem. The company builds machinesused to assemble printed circuit boards (PCBs). ThePCBs are clamped to the machine base prior to place-ment. The placement head rides along the beams of anoverhead gantry mounted to the base to pick up compo-nents from the feeder, move to the appropriate positionon the board, and place the components.

Early versions of the machine used a relatively simplePIV control algorithm. However, as the speed of thegantry system was increased, the reaction forces also in-creased. These forces generated vibrations strongenough that the components could no longer be placedwithin the specified tolerance limits. Initially, the ma-chine builder addressed the problem by increasing thesize of the base, reducing the deflection generated bythe reaction forces. However, this approach increasedthe size and cost of the machine and limited future modi-fications.

Recently, the machine builder switched to a model-based design approach to solve the problem. The firststep was to model the existing PIV controller using func-tional blocks. The command position input comes fromthe program that indicates the position of the compo-nents on the PCB, while the actual position comes fromthe sensors on the gantry axes. The original control sys-tem model was then simulated with predicted levels ofvibration matching what was seen in the real world.

The control system was then upgraded by adding position, velocity and torque prefilters. These prefilters prevent the control system from generating a motionprofile that would cause the structure to resonate. Thesimulation found that it provided a significant but notsufficient improvement in machine performance.

Next, an observer block was connected to an ac-celerometer mounted on the base. The accelerometermeasures the actual vibration of the base and providesan additional input to the control system so that it cangenerate a motion profile to counteract the vibration.The simulation showed that this approach reduced vi-bration to levels that made it possible to substantially re-duce the size of the base.Different base masses weresubstituted into the plant model in order to determinethe minimum base mass that would make it possible tomeet component placement specifications.

It would have been expensive and taken months tocode each of these possible solutions, implement themin hardware, and run experiments to determine theirperformance. Instead, the machine builder control engi-neers were able to evaluate these alternate approachesin software in weeks. After deciding on the final ap-proach, the code was automatically generated for themachine controller. A prototype was built to test the newapproach and it worked exactly as predicted by themodel. By greatly reducing vibration, the new approachmade it possible to increase production rates from50,000 parts per hour to 100,000 parts per hour while re-ducing the weight of the machine.

The last step was to addan observer block and asensor to measure actualvibration. The controllerthen generates a motionprofile to counteract thevibration.