posted July 25, 2007 12:43 AM            

What kind of information do you wish were in front of you, but isn't, when doing process troubleshooting?

How do we design such systems to avoid being inundated with more-or-less raw data versus process knowledge?

What prompts me to ask ... I've read through several of the papers linked to the energy efficient extrusion site.

posted July 25, 2007 11:01 AM            

I have been involved in complex control systems. In most cases almost nothing about them is used. If they are used, the people using the information do not understand the implications of what they are looking at.

Believe me, as I try to live in the high end of extrusion technology. Most control occurs in the mechanical systems; screw design, blender design, die design downstream design.

Part of the blame is the nature of plastic resins and single screw extrusion. Both are "fuzzy" in their performance. Extruders seem to be affected by everything, and resins are never the same from lot to lot. SPC and the like have a difficult time dealing with the variations. This is why many processes are going to twin screw (co and counter) which provides more repeatable performance. Again a mechanical solution.

------------------
Best Regards,

Tom Cunningham

posted July 26, 2007 02:44 AM            

quote:

---

Believe me, as I try to live in the high end of extrusion technology. Most control occurs in the mechanical systems; screw design, blender design, die design downstream design.

---

Can't argue with that, Tom - its what I call the "Can't shine shat" principle. If the underlying system is incapable of doing the job then no amount of control system heaped on top of it will make it do so.

On the other hand, there is Ashby's law of requisite variety, and, in my opinion, sometimes KISS can be "too KISS" (Yiddish speakers may pronounce it 'tuchis').

For instance, when I first started working around plastics equipment control systems maybe 20% of the heat-only zones were still using 20 or 30 second percentage timers, and some of these zones didn't even have a thermocouple meter to see what the resulting temperature was ... and remember one old-timer complaining about PID controllers, and how much more complex they were. And its true - PID controllers are more complex, but after you learn how to tune them their performance is superior to recycling timers.

That said, and given that I haven't seen many control systems outside of our plant I can only conclude you are correct in the following.

quote:

---

I have been involved in complex control systems. In most cases almost nothing about them is used. If they are used, the people using the information do not understand the implications of what they are looking at.

---

I'd be interested to know more details about these observations, and especially the second part. In those cases was the information they were looking at clearly presented, but the user didn't know what to make of it, or was there a problem in both areas?

Would you say these more complex systems fall down perhaps because the underlying complexity wasn't sufficiently hidden from the end user?

posted July 26, 2007 08:26 AM            

The first test of extrusion control is; is the extruder running? This is of primary concern and what everyone in the plant is taught to care about. All the focus is here.

The second test is; is it running smoothly enough so it does not require much operator attention? After the first test, then this is most important.

Third; does the product pass QC? If not there is a real problem; Which dials to twirl or is it the raw material?

Fourth; Are the operators keeping up with the line as far as feeding it and taking away product?

Now if items 1-4 are taken care of, what is the problem? It's not running the same as last time you say? Don't touch it or you will screw it up! I got another line to get running and need the help over there. Zone number 3 is overriding? So what?; the line is running OK. Must be that screw design is no good for the material. The product looks screaming hot!? Well, melt temperature is within specification. Do you have an isertion thermocouple to measure the melt temperature? A what?

Fact is even though I gather multiple full set of measurments, all of the design information, all the material information and the best extrusion simulation software, it can take me hours or days to figure out exactly what is going right or wrong with a process. It is just too damn complicated to figure out. How is an operator or process engineer going to deal with it quickly and easily? Trial and error take time and expensive extrusion time. Then the cause and effect of those trials and errors are not clearly understood so progress is difficult.

At one point it was proposed that a computer simulation of the process run side by side with the process so the cause and effect can be seen and correclty compensated for. Won't happen in my lifetime.

A similar situation occurs in injection molding. Great design efforts are put into the mold in terms of computer modeling. That is the mechanical solution, and with that solved and the remainder of the process properly sized it takes care of itself. Same goes for extrusion. 90% of control is in selecting the correct equipment. That correct equipment absorbes a lot of other sins.

------------------
Best Regards,

Tom Cunningham

posted July 27, 2007 04:04 AM            

quote:

---

At one point it was proposed that a computer simulation of the process run side by side with the process so the cause and effect can be seen and correctly compensated for. Won't happen in my lifetime.

---

Well, one of these days it will come to pass, but I can't see it happening any time soon. Even if we were able at this time to fully specify how such a system should operate I'm guessing it'll be a quarter of an LDV (Leonardo da Vinci) - about 125 years - until the pieces are in place for it to become practical on the plant floor, or, as R.A. Heinlein might say, "When it's time to railroad, you railroad". We aren't even close to railroading ...

quote:

---

Fact is even though I gather multiple full set of measurements, all of the design information, all the material information and the best extrusion simulation software, it can take me hours or days to figure out exactly what is going right or wrong with a process. It is just too damn complicated to figure out. How is an operator or process engineer going to deal with it quickly and easily? Trial and error take time and expensive extrusion time. Then the cause and effect of those trials and errors are not clearly understood so progress is difficult.

---

Well said, and again, no arguments from this quarter. However, this is precisely where, in my opinion, extrusion control and visualization ought to be focused. We can't make the big jump to the simulation vs. real time modeling scheme, but can be taking baby steps.

Roughly the first and second generation of computer-based HMI interfaces I've seen essentially clone this same functionality of a completely hardwired extruder control system, but did it first in columnar lists, then on to graphic representations of analog and/or digital meters, and often a blend of these two approaches. There wasn't much, if any, net gain in understanding the underlying process, and perhaps a step or two backwards, depending on the implementation.

My opinion here is decidedly mixed - there have been cost saving associated with virtualizing meters and switches in a computer-based HMI rather than buying actual, "hold 'em in your hand" gear, but, since screen space is still at a premium, screen-based indicators and switch objects need to be fairly tiny, and are invariably harder to interpret, especially at a distance, than their 'real' cousins.

I'm fond of keeping certain types of start/stop and other functions in the realm of physical pushbuttons, selector switches, and other operators, and key display indicators as digital panel meters rather than virtualizing them exclusively on the screen. When you are in the middle of a startup you don't want to be punching frantically at a touchscreen searching through menus, or looking for process displays!

To my mind, we are still not quite to railroading here as well, but a darn sight closer .... I'm guessing 5 to 15 years on the outside for large format plasma screen technology to mature, and demand in the consumer electronics sector to drive costs down enough for impressively large industrial touchscreen HMI interfaces (say, the width of a typical electrical enclosure door, say 32" wide) to become a reality. When that happens then many of the menu nesting and small screen object size problems ease up, although they never go away completely.

Once you have this piece in place (and other useful things will be evolving over that time frame as well, and may also be "railroadable" by then) then I think we have a chance for a revolution.

I'm running out of words for today, and am still mired in prefacing the point I want to make, but will push ahead to this limited extent.

Take extruder screw and melt pump control as an example. On a hardwired machine, or one with a computer HMI (albeit one offering merely a virtualized version of the traditional displays) you'll have a suction pressure controller, and a 5000 PSI probe for the suction side, it pushes out 0-10 volts DC to a DC extruder screw drive (0 to 100% screw speed), panel meters are provided for screw speed (scaled in RPM) and motor load (could be scaled in armature amps, or % load), and, depending on the OEM, a physical needle type pump discharge pressure meter, or maybe a digital display instead.

The details may vary (a digital suction pressure controller rather than one built around analog op amps, LED-based digital panel meters instead of d'Arsonval meter movements, 1/16th DIN digital temperature controllers versus monstrous vacuum tube based controllers, etc.), but not too different in spirit than what would have been available in a mid 70's vintage machine.

Exactly how to implement the next generation of extrusion HMI and control has a lot of ironing out ahead, but one of the things I'd like to see (although perhaps not up on a main screen) are displays in terms of 100% machine capability as well as in the more traditional engineering terms of RPM and amps.

i.e. - The extruder screw speed of 125 RPM may be 90% of capability (with the existing motor/gearbox combo). For a 300 HP, 500V DC motor a current reading averaging 450 amps is about 95% of full load capability, and so on with other parameters that can be treated in this manner.

While you may need to be an expert to know that 125 RPM and 450 amps are getting you close to the wall for that 300 HP motor showing them in terms of maximum output percentages are easier to interpret ... if you are hitting 95 to 100% then you are out of machine, and/or need to do some major surgery.

quote:

---

... it can take me hours or days to figure out exactly what is going right or wrong with a process. It is just too damn complicated to figure out. How is an operator or process engineer going to deal with it quickly and easily? Trial and error take time and expensive extrusion time. Then the cause and effect of those trials and errors are not clearly understood so progress is difficult.

---

You are still right ... but I think the challenge is to figure out how to devise systems that simplify interpreting machine and process conditions, to do a better job of historical trending (at least an order of magnitude better than what I've been exposed to so far; would go a long way toward understanding the cause/effect relationships you correctly point out are often murky, captured as they are in 'the heat of battle') - essentially, to do a lot of the time-consuming legwork otherwise required - and do it without costing a couple of arms and legs.

posted July 27, 2007 09:02 AM            

I wish any of my customers would run just one formula on their extruder. Then a tight control scheme could be built. I have never seen this however. Usually regrind and reclaim is introduced and they are the bad actors.

I don't live too far from your location. I'd like to visit your operation some day to see what you have done.

I like local control elements. A hand turned potentiometer for downstream speed control right at the extruder head and such. 10 turn pots are a problem sometimes because they are typically scaled for all 10 turns and take forever to get to gross settings. So somehow a gross control and then a 10 turn for fine control would be good. I know it has been done.

Automatic controls have a hard time with startup and upsets. If you could easily switch from manual to automatic without a big bump in operation that would be good. Get it settled down in manual mode, then switch to automatic.

PID is a big problem when it comes to startup, upsets, and instability. For heat control it is good, but for some other systems it is a problem. PID is great when thing are close to setpoint and stable, but can go crazy during upsets. In most cases the ID portion can be turned off with no problem and the P or gain can be turned down a lot. This really helps settle thing down for a lot of equipment. Extrusion should be stable, so correction needs should be small and slow. PID attemts to follow the upsets with big corrections making things worse when the extruder has an upset. This connects up with the previous paragraph. Bring it in manually, wait till things settle down, then switch to mild automatic control.

Most extruders will run with a steady amp draw, and a steady head pressure. I set up controls where the average value was recorded for the last five minutes. If a real time value (still averaged for 15 seconds or something like that) exceeded +/-20% (or appropriate programable value) then the alarms went off, or the line went into automatic shutdown. This would help to prevent a lot of other problems or even save the line from going down if possible.
Again, start up the machine, stabilize, and then activate this alarm system.

Do you use water cooled extruders? I could go on forever about controlling that system!

------------------
Best Regards,

Tom Cunningham

posted July 29, 2007 09:30 AM            

A good point not often understood by control systems designers.

------------------
Best Regards,

Tom Cunningham

posted July 30, 2007 01:29 AM            

quote:

---

Steve: A system must be inherently stable and easily controlled manually before it should be put under automatic control. You cannot put controls on a brick and turn it into an airplane, so to speak.

Tom: A good point not often understood by control systems designers.

---

I'm not making any argument favoring the use of advanced control techniques to turn extrusion lemons into oranges, but don't think it is outside the realm of possibility.

After all, it is possible to control an F-117 Nighthawk jet, but the necessary avionics and other control system engineering are the culmination of many millions of man-hours and trillions of dollars (figuring from the evolution of U.S. military jet technology starting with the F-80 Shooting Star), and building this plane so it is essentially uncontrollable without all the gizmos was done for a specific purpose (stealth).

In a like manner, it may be possible to control a manifestly unstable extruder satisfactorily by piling on enough control system tweaks, but seems to me we'd need to understand the dynamic processes much more thoroughly, and pour in a lot of engineering money to make it work - it is simply a lot more cost effective to build a system that is stable from the git-go.

quote:

---

This is a good conversation. I would hope that others would add to it. However in-house control schemes are considered proprietary so likely many won't comment.

---

It is prudent to be sensitive about giving away too much of how particular control systems operate, but I don't worry too much about discussing generic control strategies. The devil is in the details, and my view is talking about generalities can't let the cat out of the bag. Simply being exposed to a concept shouldn't be enough to spell out how to do a thing, and, if it is, then that concept is fairly trivial to begin with.

For instance, a common way to describe screw output is in terms of specific energy consumption (HP per lb-hr), but I have yet to see an extrusion system presenting this information although it isn't difficult to formulate how to do it in a general sense.

If a digital motor drive has instantaneous power data available in a data register then glom onto it, and do whatever math is necessary to convert it to horsepower. If not, then monitor the drive using whatever analog inputs and conditioning are necessary (for a DC motor, armature voltage and current), then do the math. We can manipulate it further for whatever screen displays are desired (current power demand, % of drive/motor power rating, etc.), and also accumulate power usage for some period of time. What to do here is somewhat dependent on the blender system - if continuous gravimetric then perhaps a fixed 15 minute time period would do, and for a batching blender, after x number of blend cycles have occurred.

Provided we're talking about a loss weight blending system with access to accumulated weight data, then pull the accumulated station values at time x, then again at time y (the fixed 15 minutes, or after x batch cycles), and calculate delta weights and elapsed times. Since you are getting it anyway, you may as well leverage it - show individual component percentages, virgin vs. reclaim percentage, and so on, as well as calculating the accumulated pounds over the period.

OK, now we have power usage and material usage over the same period - a bit of math, and voila, HP per Lb-Hr.

It'll be necessary to wrap some logic around this to take into account power failures, other planned and unplanned extruder shutdowns, and other 'outlier' conditions so the calculated results aren't nonsense.

HP/Lb-Hr shouldn't really change substantially for a particular screw design provided that operational conditions and material factors aren't changing. Precisely those factors will make it change a bit over shorter time periods, but it shouldn't fall out of a particular range - if it does, then something is probably wrong.

What constitutes a reasonable range of HP/Lb-Hr variation? I really don't have a clue, since gathering and processing raw performance data into this measure is time-consuming, and so many other factors enter into the picture.

posted July 30, 2007 08:15 PM            

People use kw-Hrs/kg, HP-Hr/lb, and Lbs/hr/HP.

In my early days Lbs/Hr/HP was poplular. We got numbers anywhere from 6 to 16. Interesting, but knowbody knew what to do with them. As efficiency went up, so did the number. As work went up the number went down. Useful for specifying motor size.

Then I got involved with CRFI (Co-Rotating Fully Intermeshing) Twin screws. The Germans were concerned with how much work they put into the material, so KW-Hr/kg was employed. As the number went up so did the work put into the polymer. Typical numbers were 0.1 - 0.3 Kw-Hr/Kg

Now single screw English speaking people are beginning to use HP-Hr/Lb.

So I wish I knew your first name rawelk, your process is sucking some power because you are at the high end at 6-7 Lbs/Hr/HP and .2-.3 Kw-Hrs/KG.

Either your material is very viscous, your head pressure is very high, you are pumping a lot of heat down the drain, your melt temperature is hot or some other issues is causing you to use a lot of power.

Gross assumptions about a process, as above, are the only uses I have found for these numbers.

------------------
Best Regards,

Tom Cunningham

posted July 31, 2007 03:41 AM            

Merde! Thanks for pointing out I was using the reciprocal to what most other people use (I've since amended the worksheet to show LB/HP-Hr, and added the conversion for kw-Hr/kg).

We tend to use high reclaim levels, and run the first stage zones set relatively cold (and are almost alway deep into cooling on them). A couple of months ago we started running a higher throughput job, and the extruder became badly unstable - large swings in suction pressure with occasional sustained dropouts.

The only thing that seemed to make a difference was increasing melt pump suction pressure, which I'm guessing helped by keeping the screw mixer section more full. Now, stability is good (+/- 30 PSI around setpoint), but with the higher power requirement in trade.

It is a 24:1, 6" extruder, and when those values were measured the screw was running ~118 RPM at 465A at ~3000 PSI head pressure, and using ~1960 #/hr to yield the 6.63 Lb/HP/Hr (0.248 kW-Hr/kg) number (check my math, and see if it passes the smell test this time). Unfortunately, I can't go into the screw details.

posted July 31, 2007 08:42 AM            

Regarding your process it seems outside the norm. RPMs are higher than typical for a 6" unless you are processing low viscosity melts. Output seems to be 2/3 of expectation. As always it it difficult to understand a process from a distance.

My process philosophy is very conservative. However in thinking about it, it is consistent whether I'm working on a single screw or twin screw. I believe this philosophy permits cosistency in the process so the measurements are more consistent, informative and believeable.

I design the process so that I am sure the polymer is melted. This is not as easy as it sounds, nor do a lot of processes show outward signs of unmelts.

I look for efficiency. Spinning a lot of heat into the polymer over a long period of time is a waste, and degrades the polymer, but it happens.

I make sure the melt is mixed enough. This provides a quality melt with uniform temperature.

I also consider pressurization, as a lot of bad things can happen if the pressure profile (along the screw) is off in the process.

If all these things are done right, then a stable, well developed melt will be delivered with some tolernce of variation.

Variation usually comes from the resin feed as the resin maker has some variability in the raw material, purchasing buys different grades and regrind is introduced. These can be dealt with up to a point. But if advances control techniques are to be used, such as SPC, then variation must be exercised out of the system from stem to stern. This is rarely if ever done in extrusion as far as I can tell.

So I'm back to the mechanical therory. That is; generally we build some hunk of metal to make it come out right. Hence my conservative screw designs.

------------------
Best Regards,

Tom Cunningham

posted August 20, 2007 12:21 PM            

I'm looking for your contact information.

Please send it again.

------------------
Best Regards,

Tom Cunningham

posted September 24, 2007 12:08 AM            

quote:

---

... an extruder with a little A.I. would be nice for financially strapped companies.

---

I'm not looking to do anything quite so ambitious, at least, not in the foreseeable future.

What I do foresee is an incremental change toward more comprehensive control systems with some degree of built-in diagnostics. The downside is that the instrumentation techs tending for them will need to know more detailed information, and it won't come to pass unless we can tame the effects of control system obsolescence.

For instance, I recently read an inquiry (on Eng-Tips, maybe?) where the question was how to detect imminent heater band failure. Well, there isn't any way that doesn't require good historical documentation of what constitutes "normal" operation, and, even then, some percentage of heaters will blow out either from undetectable aging effects, or, through "misadventure " (rough handling while doing other maintenance, etc.).

What I'd like to see become a standard approach are control systems which are also repositories for their component information, and that can compare measured performance against it.

Say we have a cast-in barrel heater half rated 4500W at 240V. Using Ohm's law and the power law we can determine current in amps, and element resistance in ohms for the nameplate values.

For the control system to monitor heater condition it needs to go into more detail, and differentiate between 'hot' (operational temperature) and 'cold' (room temperature) resistance (and hence, current, and actual operating wattage) since the elements have a positive tempco, and resistance will be a bit higher (and current and power a bit lower) when at processing temps. It also needs to account for line voltage variation which can change up to 10-15 volts over the course of several days depending on a number of factors.

Ideally, it would have a 'heater good' allowable current range for both conditions, and only throw alarms if current fell outside of the window.

Other tests (i.e. - using a megohmmeter for short circuit testing) can be useful in both historical, and 'go/no-go' testing, but isn't easily integrated for automatic testing into the machine. What could be done here is provide the ability to key in measurements from external test equipment along with other key facts (date, time, test voltage, etc.).

About a year ago we went through one of our extruder barrels, and learned a couple of things. ~10% of our elements were starting to show abnormally low insulation resistance to ground, and were retired. We also noted the original heaters had slightly lower terminal resistances that new replacements, and found it due to a change in manufacturing (slightly smaller wire diameter).

I suspect the required degree of detail to build such a control system - one that would be able to handle things like this change in heater manufacturing in some sort of intelligent way - is a major, if not *the* major stumbling block to putting more intelligence into control systems.

Although it isn't free, the major cost factor isn't in detection hardware, but rather in developing standards for what information to track, how to store it, how to get at it again, how it should interact, what alarms to generate, and developing software to do these things.

Without an industry supported standard then everybody who tackles it ends up inventing many sorts of slightly different 'wheels', and, what with control system software obsolescence, may end up doing it all over again two or three times in their career.

Heck, this pretty much is the way we stand now, but it isn't too much of a problem with simple systems.

Start to integrate them tightly with component characteristics, and other information enough to be useful, and generate sufficient payback to justify the additional work, and the lack of standards and control system obsolescence becomes a deal breaker.