What is Control?

‘Control’ is a term for the use of binary calculation methods to execute a process or task.  I suspect it is as ambiguous a term as ‘mechatronics’.  I suspect that we cannot even agree on what control is, without getting into some depth on the all the possible definitions of the subject.

This is rather ironic considering the billions of dollars that are spent on control systems across all fields. Is control fundamentally any different if it is inside a car, automating lighting and HVAC in a large building, on an automotive manufacturing plant floor, in a biological resesearch laboratory, or in a giant refinery where chemical products are made.  It’s all control.  And the more we try to define it, the more inclusive we make the definition, the more vague and ambiguous the term becomes.

Efforts continue to increase the power of the PLC (programmable logic controller) across many vendors. By increasing processor speed, memory and capability PLC’s are becoming the universal platform of control as a discrete controller, process controller and motion controller.

Simultaneously, motion control specialty companies continue to increase speed, processing power and I/O structures in an effort to expand the dedicated motion controller as a competitive platform to the PLC.  This is a necessary migration to address control applications where an external PLC could be eliminated.

Is there an ideal mix of motion axes and I/O that will help resolve which hardware solution is best?  Not really.  The fact is that the majority of the market is made up of motion control using stand-alone axes that are triggered by logical conditions in the system.  Coordinated axes require the sharing of pulse to pulse position feedback information.  Stand-alone axes do not share data at that low a level in time.  Most PLC controllers are well able to handle stand-alone axes, especially if an intelligent indexer is used.  This off-loads the motion to the servoamplifier and only I/O handshakes are used.

Part of the ambiguity here is that control is the result of hardware and software together.  ’Control’ seeks to generate complex behaviors using digital methods.  The digital methods, processors, depend on programming techniques in order to implement the desired behavior.  So when we talk about Control, we are talking about hardware and software simultaneously.

What matters most to users of automation technology is both logic control and motion control programming exist in a single environment.  It doesn’t matter if the programming environment is a PLC with motion blocks inside it, or a motion controller with logic blocks inside it.  What matters is that all aspects of a control system can be programmed using a single editor.  Controllers from the major electrical companies like Rockwell Automation and others have opted for the logic-centric programming environment with motion blocks in the ladder diagram.

This approach eliminates the complexity of multi-processor solutions, each with their own programming language, that were commonplace a few years ago.  Multiprocessors have their own unique programming environments and a significant amount of programming to create proper interaction between the various platforms.

Missing from this description is the hard wired control that is part of system start up, power management and safety.  More on this in the next installment.

Servo Tuning

There are many opinions about servo tuning.  Every engineer who has ever worked with servo motors has experienced the difficult process of tuning the motor.  The difficulty is in the fact that the rules about tuning are hard to apply, because every situation is a little different.

Considering the rules for servo tuning first, the ideas are basically simple.  Based on the use of 0-10V velocity command, the control system is designed to regulate motor speed.  A Proportional, Integral and Derivative gain value is used to “tune” the command signal sensitivity to allow the control system to regulate the motor and load performance.  This strategy has been created over the years and is used by almost all servomotor and drive suppliers in the motion control industry.

The Proportional term is the most important value in this approach.  The proportional value is generally the amount the 0-10V command signal will be increased in response to a following error.  The more gain, the more velocity will be commanded.  This will allow the control system to correct for changes in load conditions.  The proportional gain is how the system responds to current conditions.

The Derivative term controls how quickly the control system can add or subtract energy from the load.  The derivative can take the form of dI/dt or dV/dt depending on the specific controller.  This term has two important purposes.  As stated, it defines how quickly the system responds to changes in the load condition, and it exactly parallels the breakdown limit of the power transistors used in the motor amplifier.

The integral term provides correction on a cumulative basis.  All previous error information is integrated over time to provide the system with correction to the control command that “damps” reaction to disturbances.

Notice that all the gains are directly tied to time.  The faster the motion, the more P and D gain is needed to provide adequate response in the control.  Many motion applications have low enough dynamics that servo tuning needs to be very low performance compared to the capability of the equipment.

For most applications, and depending on the gear you are using, the best thing to do is start with I & D gains set to zero.  Use the amplifier autotuning for the motor without the load.  Sometimes these are default settings that are already loaded into the controller.  Tune the motor and amplifier until the P gain seems the best for a step input.  Then add the load and run the autotune again.  If possible, use a step response input that is similar to the type of move you will do in your actual application.

Then gradually add D gain until the leading edge of the step response has no overshoot.  If there was little overshoot without P gain, the motion dynamics are probably very slow and that’s OK.  With D gain set for the application, slowly add I gain and see if the trailing edge of the step response is improved.

If the axis is “hunting” after the motion stops, there is probably too much play in the mechanical system. Using a gearbox with a lot of backlash or a timing belt that is a little loose will produce just enough mechanical error that the servomotor will detect.

Tuning, like everything in motion control, is as unique as each individual application.  There are more complex analytical techniques that can be applied to the subject of tuning.  But I hold to the theory that the majority of applications can be dealt with using simple techniques.

More Measurement and Motion

For all the measurement technology we have available there are some elements of motion control that are generally missing.  We have laser interferometer measuring tools that are accurate to a fraction of 1 micron.  There are rotary position sensors that can divide a circle into a million digital positions. Many of the semiconductor industry’s processes would not be possible without the incredible advances of measurement technology.

Sometimes the motion control aspect of a process is not the primary objective of the control system or machine being considered.  The process of clamping or crimping a can lid onto a can body is an example of this situation.  The motion control system must locate the can lid to the can body correctly, but the final process is the crimping or application of a thrust force to cause the parts to form a strong joint.  In this case the real process variable is the pressure that is exerted at the end of the motion.  The pressure is critical to joining the parts, especially when the can is an igniter for an automotive air bag.

Grinding and polisihing is another example.  The motion control application is required to bring the grinder or polisher into contact with the work piece.  The actual grinding or polishing is the amount of friction generated between the grinder motor and part being worked.  This is actually proportional to the current of the grinding motor, which can be measured and regulated.  If too much current is detected the part might be ruined and the control system can be commanded to move the grinder away from the workpiece.

Important physical attributes of motion include inertia, center of mass and momentum.  There are no convenient sensing technologies that help us with these seemingly basic attributes of the mechanical system.  This is probably why they are ignored in the control system.

However, if the machine was designed in a 3D solid modeling environment, then things like center of mass and inertia are directly available.  A momentum profile can be created as a product of the center of mass and the duration of the motion profile.  This gives us mathematical information that can be used to “inform” the control system in spite of the absence of a control signal that directly measures these properties.

With this in mind one can easily imagine a pick and place mechanism made from two linear stages mounted one on top of the other.  When the two axis are making high speed coordinated moves, the reflected forces of the upper axis put loads on the lower axis.  The data from the solid model becomes useful information in providing mathematical “filters” that can improve the motion in ways that are beyond the current technology of motion control.

There are ample opportunities for improvement in the control of mechanical systems.  We should be looking for new strategies that the modeling and simulation environments provide.

Motion, Measurement and Control

Motion control is all about control.  But you cannot control what you cannot measure.  So there is an important measurement component to the control of moving systems.  The difficulty lies in knowing what to measure, how to measure and what to do about things you can’t measure.

The obvious thing to measure is motor speed.  That part is easy.  Servo motors have built in feedback devices. In the old days, the preferred feedback device was a small generator that produced a voltage proportional to the speed.  In the digital age feedback is by quadrature encoder that outputs a digital pulse that is primarily used for position control.  Most control systems are able to easily integrate the pulse train to derive the speed of the motor.

Unfortunately, most applications require relatively low speed.  Most motors are engineered for high speed.  This is in an effort to package more work related power in a smaller physical package.  Often, the motor is connected by pulleys or gear reducers to get the speed of the motor to more closely match the desired speed of the load.

Some of the important attributes of motion cannot be easily measured.  In addition to speed, torque is extremely important to controlling motion.  Torque can be measured directly from the drive electronics, but this is rarely used for control.

Torque and current are direct equivalents with a slight variation due to the temperature of the motor winding. As the temperature of the motor goes up, the resistance goes up and the current required goes up at the same time.  Since high performance motors have fairly high internal temperatures, this swing can be in excess of 100 degrees centigrade, and should be considered in the control scheme.

Most of the emphasis on current control is in terms of protecting the motor and drive electronics.  The first derivative of current over time  is the limiting parameter of the power electronic devices and is an important boundary condition in safe operation of the electronics.

More important information can be derived by considering the region of the motion profile and the current or torque requirements that are presented.  In order to accelerate a load, a lot of current is needed to overcome the mass of the load.  But once the load is moving the torque requirement drops off.  This creates an opportunity to profile the current requirement while using the conventional error detection scheme of the traditional control.

Other variable that are part of the mechanical system are things like momentum and center of mass.  In multi-axis mechanisms, there is usually a dependency of one axis upon another.  The idea that the mass of one axis is changing it’s center of mass and momentum with respect to the other axis is generally ignored.  This too is an opportunity to gain increased stability in the control and possibly improve throughput by having a better model of the application from which to create the ideal control.

Looks to me like there is a lot of room for improvement.  Let me know if you agree or disagree.

Hardware, Software and Performance

In the industrial control world, its all about performance.  With every year that passes control system performance improves.  This is heavily influenced by the computer industry’s ever increasing cost effectiveness which has driven the convergence of so many industrial control products to PC based solutions.

As control technology has progressed over the last few decades, the performance requirements haven’t really changed much in a certain sense.  In the case of Computer Numerical Controls, the basic performance hasn’t really changed.  The big difference has been the cost.  The early CNC at $500,000 has been replaced by today’s production CNC at $50,000.

Certainly, there have been significant advances in capability, larger machine structures for aerospace applications, high precision applications and even the advent of the “maker bot” low end machines.  But the basic relationships required to shape parts in 3 dimensions are the same.

Yes, there are example of industries and applications that simply didn’t exist 50 years ago, like semiconductors, flat panel displays, etc.  Most of the improvements in electronic motor controls and motion control actually derive from the work done to improve the disk drive spindle motor.

But generally speaking, industrial control has migrated from specialized control hardware, to the PLC, a general purpose, high reliability form of controller, to personal computer based solutions.  Primarily due to the low cost and high performance computing power of the PC.

Some of the performance attributes have to do with how a given control is architected.  In the push for high reliability, the PLC goes to great lengths to insure that memory cannot be compromised, input-output hardware is extremely resistant to external electrical interference, and state changes are captured in a rigidly deterministic manner to insure that control code executes consistently.  In the PC there are PLC emulators that have been approved by various agencies that qualify their performance as equal to PLC’s.

Some attributes of performance are not obvious.  The operating system and its performance become part of the control system equation.  If the OS faults, the control system is down.  Hence the advent of Real Time Operating Systems, Linux and a host of other solutions aimed at reliability.

It can go deeper than that.  Exactly how the OS deals with interrupts can mean the difference between reliable operation and catastrophic failure.  Windows CE has different performance attributes over Windows for industrial control application.

In the PC architecture, how does data move from the disk drive to the operating program?  In the PLC how does the backplane impact the update rate of field I/O?

In the software realm, newer operating systems have made possible multi-threading programs that allow several task to execute simultaneously in the same hardware and software environment with complete coordination between tasks.  Robotic and CNC applications with multiple system operating in a single controller are able to execute separate programs with sophisticated anti-collision detection.

So exactly where the control system performance originates from is not exactly a function of a specific piece of hardware, but rather the result of hardware, OS and software combined to achieve a desired outcome.  Not a simple matter.

Putting the “Mecha” into Mechatronics

Motion control, machine automation, mechatronics, actuator technology, whatever you want to call it, the thing you have to keep in mind is that in motion control, tasking is mechanically bounded.  The boundary conditions of the system are defined by the mechanical solution.  It is generally impossible for the control system and power electronics to cause a large inertia mass to behave like a small inertia mass with high acceleration rate and high frequency dynamic response.

There is a lot that can be done from the control system to improve overall performance.  Some of that improvement takes the form of sophisticated control algorithms.  Some improvement may take the form of increased power electronics to provide added acceleration and deceleration capability.  And sometimes projects get scoped to do things that are marginally impossible.  This is because there is gap between our understanding of the difference between mechanical and electrical properties of the application.

The gap in understanding is partly due to the educational system that teaches mechanics and electronics separately.  With the newer mechatronic programs that are available at many schools this gap is decreasing.

But there is a deeper issue.  The issue is context.  Mechatronics combines electronics and mechanics.  The missing context is that everything we do in the electrtronics, is an  analog for something mechanical.  If you don’t know how the mechanical relationships work, it’s unlikely that you will get the best results in controlling the system.

There are all sorts of examples.  A belt and pulley reducer is a mechanical system that turns a high speed input to a lower speed output.  The analogy in motion control is electronic gearing.  We can arbitrarily program any given motor to follow a speed reference from an encoder, tachometer or other motor, and follow the input signal at a programmed speed or ratio.

Electronic line shafting is a similar application that seeks to operate several independent motors and loads as if they were connected on the same mechanical shaft.  This application requires very high angular precision between the following loads which requires high speed regulation between the motor and associated electronics, but mathematically is very simple.

From the mechanical standpoint moving something from point a to point b seems pretty simple.  But as the time requirement for the motion decreases, the forces acting on the system become very significant.  As the load is accelerated from rest, it gains momentum.  Overshoot at end of the acceleration profile is a property of the momentum or kinetic energy of the system.  Tuning the system properly involves understanding the mechanical properties of the load and specifically how the control systems relates to the mechanics.

Does overshoot even matter?  Not so much if there isn’t a following axis involved.  At the end of travel it may be a concern if the load is positioning under a fixture or a next mechanical operation.  In many systems, there is settling time required for the load to come to a complete stop.  In these applications, the “S Curve” is an ideal solution because it compensates for the mechanical properties of optimizing the acceleration so that little or no overshoot takes place.

Maybe it’s less about the tuning and more about intelligent trajectory planning.

Priorities and Projects

Getting one’s priorities in order is a big subject.   It’s the big subject that we try to wrestle through in the political arena.  What will we choose, as a nation, to spend money on?  What do we, as individuals, choose to spend money on?  Judging from the consumer electronics show, we still like our electronic toys.  As a nation, however, we need to do a better job of getting good information into people’s hands so as individuals engaged in political process, we can make better decisions about our country’s future.

In managing technical projects, there are similar issues.  In technology projects, there is stuff we like to do, and there is stuff we need to do.  And sometimes, the what gets done and in what order really impacts the results.

One aspect of project management that is often not considered is defining what is important.  And especially what the order of importance is.  Because there are always competing attributes in any automation project, and cost tradeoffs that need to be considered.  So a really valuable exercise at the beginning of the project would be to make a list of the project priorities and put the list in order of importance, so you and your team know what the important issues are.

The typical project priorities involve;                                                                                                                                                             speed as throughput in parts per minute or hour,                                                                                                                                   amortized cost which is the cost of the machine per part produced,                                                                                               and part precision or accuracy which can sometimes be measured as scrap rate

These are the most common issues in manufacturing.  There are certainly many unique constraints on project performance that can be identified in specific situations, but these are the most general and the ones with direct cost impact both in the development of the machinery project or process and in actual production practice.

The problem is that while the project is being developed, decisions will have to be made on the fly.  With a formal statement of what the project priorities are, it becomes easier to determine where money needs to be spent and how it relates to the final performance.  It also becomes easier to avoid pitfalls where spending money produces no benefits in terms of the project priorities.

Meeting part per hour production rates may be well within the  project forecast, but achieving quality goals and reducing scrap rates may have some cost tags that will need to be considered.  How much will it cost to reach the next improvement in production throughput?  How will it impact the amortized cost per production unit?  Will a new, expensive solution to part accuracy pay off in reduced scrap rates?

All of these questions get a little easier to deal with by knowing your priorities.

Competition in World Markets

As a follow up to “Made in America,” I would like to explore some of the dynamics of competition in manufacturing in a world market.  We have all seen the effects of American manufacturing outsourcing to Mexico, Japan, India and China.  Maybe this progression should take many years, maybe American political policy has contributed to how quickly the change is taking place.  And of course that makes it much harder to adjust to.

Reducing the cost of manufacturing is always an issue.  Who doesn’t want lower prices?  But labor costs alone are not the whole story.  And its hard to compete directly when labor forces are wildly disparate.  With US labor costs of $12-45 an hour competing with costs of $.64 an hour in China, it is hard to deny the merit of the strategy of just moving everything over.

In fact, intermediate macroeconomics will tell you that all markets gravitate to the lowest price.  As trade between countries becomes more widespread, using foreign labor becomes more prevalent.  The US moved a lot of its automotive parts to Canada and Mexico for that very reason.  And high scrap rates out of Mexico were tolerated as we tried to take advantage of the proximity of the labor force and low cost labor.

This also makes the point that the main barriers to using foreign labor are distance, which means transportation costs, and quality which means scrap cost.  Transportation costs are constantly changing, as recent fuel cost show.  Transportation costs, which were steadily decreasing in the 1990′s, may have contributed to the “offshoring” strategy.  As fuel costs increase, the cost of using foreign labor increases as well.

Scrap cost or serious product deficiency that results in recalls or injury to customers creates huge expense and bad press for the product involved.  These situations can severely impact revenue and the risk may not be worth it.

Tariff and duty charged by national governments are cost barriers that are political means of creating barriers to foreign goods.  The US has had difficulty with balancing trade with its trade partners. Special situations where foreign governments have large loans to the US make negotiations even tougher.

The big issue is the value of currency.  As the dollar falls in value, other currencies become more expensive, so the relative cost of labor changes.  Again, a very dynamic situation over time.  And this is exactly the case with our largest trade partner, China.  Not only is the currency going up, but China’s millions of emerging consumers are experiencing inflation at the same time.

But beyond the standard economic factors, there are hidden costs to using foreign labor.  A growing number of US manufacturers have found that offshore labor costs aren’t the bargain they initially appear to be when the foreign manufacturer requires constant on site support in order to maintain promised product performance or delivery schedules.  Off-shoring involves significant risks.

All of which leads to returning manufacturing to the US.

Critical Systems

The recent post on the Japan earthquake prompted quite a bit of response.  Some commentary focused on the technology of nuclear power and what the alternatives might be.  Other  comments were very insightful about the nature of engineering for critical applications.  I found the comments very thought provoking and will take this week’s post to expand the discussion.

With regard to the technology discussion there was a great post regarding the fabled Thorium reactor.  By far, the YouTube post linked by a reader was the most informative piece I have ever seen on the subject.  At 16 minutes, it’s well worth the effort.  Please check out  http://youtube/WWUeBSoEnRk or Google search Thorium reactor and note the Google Tech Talk from 2008.

There is so much knowledge available and yet so little makes it into the mainstream of political decision making.  And this, I think, is the essence of a couple of comments from readers.  Will our politicians make good decisions?  Is it even possible?  There seems to be so much “influence” going around.  Does General Electric or Westinghouse really care about the technology they are putting out, or is it more cost effective to just re-hash the 40 year old water cooled reactor of the late ’60′s and just keep truckin”?

New technology does not guarantee performance.  It is simply that the improved solution is expected to have similarly improved performance.  But that performance has to be proven.  The thing to do would be to fund a demonstration program to construct a variety of technologies  and get some idea of how they really work.  Working hardware is easier to evaluate than concept drawings.

On the engineering side, a number of comments were focused on the nature of critical systems.  In the controls arena there is a combination between the control system software and control system hardware that has to properly designed in order to achieve the robust, stable operation required in many critical applications.  Electrical practice also includes a number of important conventions that impact operational safety such as removing power from part or all of the control equipment.

More complex issues exist in understanding the nature of the critical systems in nuclear power plants.  Clearly the water supply and pump for cooling the core of the reactor are critical systems.  The engineering challenge is to make the cooling system reliable in the event of  a power failure and in the event of earthquakes.  These are very complex circumstances to engineer around, but that’s the challenge.

The need for establishing higher reliability power and backup systems is critical to our future.  In order to achieve the power goals of the modern world, the fundamental technology options must be explored an the same time sophisticated methods of managing the equipment in order to keep things running.

All worthy goals which we need to pursue.   With all the available options.

Fast, Faster, Fastest – Or Not?

Speed is relative.  Especially in the world of industrial control.  1 millisecond look ahead features in the machine control world used be considered “cutting edge” (pun intended).

The programmable controller, the standard of industrial control, has a speed of execution metric.  It is generally specified in thousands of instructions per millisecond.  At 1000 instructions every 2 milliseconds, for example, it is easy to assume that there are no applications that will be a problem.

Programmable controllers, PLC’s, are the very essence of dependability.  In fact, they are one of the few controller technologies recognized by all the major safety agencies.  PLC code execution is very robust, and in recent years, has even migrated to fault tolerant systems that are the next level for high reliability.

As these systems have migrated to faster and less costly processors, users have benefited from falling prices and increasing performance.  Enhanced features like Ethernet communications, math functions, importing values from Excel spreadsheet, and even motion control has made their way to the PLC platform as a common industrial hardware solution.

However, when you add motion control to any control system application, you must ask and answer the question, how fast is fast.  How “real time” does my solution have to be in order to behave correctly?

This actually gets down to the level of understanding digital sampling and analog to digital conversion.  You can look at a 12 bit  analog command signal and think that 4096 increments of the command voltage to the drive is plenty of resolution.  But you could also be wrong.

If the drive is a servo, it is entirely possible for the 2 kilohertz sampling rate of the drive to catch the little steps between output values and actually try to follow the step function, which was never intended.  This situation can cause current inrushes to occur which will either cause nuisance trips in the drive, or gradually cause the drive to shut down from overheating.

Think its far fetched?  Not at all.  It actually happened to me during a project on a PC board plating line at Hewlett Packard some years ago.  The fix for this might be output smoothing of the analog command, but that function doesn’t exist in PLCs.

But one of the critical aspects of insuring the reliable performance of the PLC also creates some problems for time sensitive  applications.  PLCs update all of their inputs first, execute all of their logic and then set all of their outputs. So latency is built into the process intentionally to protect the application from certain types of failure.

If you have to read a value from a sensor, the analog value has to be stored in a register. Then it has to go through a read cycle in order to be retrieved and operated on by the math calculation in the program.  So there can be several clock cycle of difference between when the data was read, when the program calculates a value, and when the output value is sent to the output to be updated.

These little timing anomalies creep into the execution and can appear entirely random and very difficult to de-bug.  And often you don’t know ahead of time that the problem is going to affect your project.  Until it’s too late.

Industrial PCs, as an alternative, have migrated from the old 25 megahertz 486 to the current 1.8 gigahertz Celeron chips.  These systems run hundreds if not thousands of times faster than PLCs.  Often for industrial application the Operating System can be Linux to increase reliability.

So ask yourself how fast your system really needs to be to meet its requirement.  Sometimes the fastest PLC on the block isn’t going to be the right choice.

« Previous Page — Next Page »