PDA

View Full Version : Temperature Control +/- .5 deg C



snowman
07-07-2009, 10:03 AM
I am in need of an incubator. Those available on the market do not have PID controllers, or anything fancy, but state that they are able to maintain temperature within 1/2 deg celsius. I have at my disposal, PID controllers, but I believe that setting it to 1/2 deg would result in a constant pulsing, which is hard on the relays controlling the heating element.

I need the incubator for running hematology tests on wildlife and domestic animals, as well as culturing microbes from infected wounds if resistance is seen in standard antibiotic therapy. All that I would be doing is transferring the culture to an appropriate lab, I don't mess with microbes in concentrated forms...

I really have no clue how manufacturers are controlling temps to such a fine degree of precision, without the use of fancy electronics. I have plenty of heaters on hand, but really have no experience in controlling temperatures like this, my experience is high temperatures +/- 10 degs.

Any thoughts?

An RTD seems like it should work, but so does a simple rheostat, as it's not like the incubator will be setting outside in the elements, it'll be in a breeze free, mostly temperature controlled room.

Evan
07-07-2009, 10:07 AM
Dump the relays and use a solid state relay. Problem solved.

JoeFin
07-07-2009, 10:29 AM
Yes we did just that for a DoD lab operated by 3M - (I have no idea what they were doing)

Part of the trick was they had to be constantly circulating the air through a Laminar flow ventalation system. The other part of the trick was a fairly sophisticated control system (PC based) that incorporated "Fuzzy Logic" (which means it Learns) and a whole slew of Algorithm - and yes a PID was one of them

mechanicalmagic
07-07-2009, 10:36 AM
Precise temperature control is a bit of an art, but it goes something like this.

Calculate the heat loss at temperature (for discussion, 100 watts).
Use a heater that's double the loss (thus 200 watt heater).
You need a relatively large, good thermally conductive plate attached to the heater, to spread the heat. Most of the floor is good, 1/2" Aluminum works.
The temp sensor must be located VERY close to the heater.
A fan inside the incubator enclosure is required to eliminate thermal gradients.
Evan is right about the SSR, and the internal relay in your PID should last 20 years with no load.

Watlow has some handy info:
http://www.watlow.com/reference/index.cfm

Dave J.

Evan
07-07-2009, 10:52 AM
For best results look for a solid state relay that also incorporates zero crossing detection and switching. They are made specifically for heating element control. It eliminates noise generation on the power line as well as radiated RFI and also makes the heating elements last much longer.

RKW
07-07-2009, 11:09 AM
If you have a true PID controller I don't know why this would not be possible.

It might take a fair amount of tweaking but once you get the P, I, and D variables dialed in it should work fine with out overworking the relay. You might also consider a SSR instead of a mechanical one but keep in mind the failure state of some SSRs is ON!

dp
07-07-2009, 11:22 AM
A good precision controller would know what the controlled temperature is, but also the ambient temperature so it can provide anticipation to induced changes. The problem will be lag in the system, no matter what, so that is also an issue. Since incubators need only control air temperature, the entire loop should deal only with air temperature and ignore surface temperatures.

Humidity is also critical in incubators, of course. My work in systems like this demonstrated that controlling the atmosphere external to the incubator worked best - the controlled air was then injected into the incubator at low volume. A two-stage system, in other words. I raised 24 turkeys from eggs this way, and don't recall how many chickens, and never had any problems that reveal themselves in the last week (too much humidity produces horrific results inside the egg, frinstance). For all I know that system is still working after all these years (built in 1972).

fixxit
07-07-2009, 12:23 PM
Another way of compensating for temperature variations is to add some thermal mass such as a water tank with the heating element located inside.

vincemulhollon
07-07-2009, 12:33 PM
You might also consider a SSR instead of a mechanical one but keep in mind the failure state of some SSRs is ON!

Which brings up the point that you need multiple thermostats.

1) Fancy PID to control the temp

2) Cheapo that sets off a mildly annoying alarm when the temp is too low. But not too annoying of an alarm since you'll be listening to it while the system initially heats up.

3) Cheapo that sets off a really annoying alarm when the temp is too high and also cuts power to the heating element and/or the whole PID system and/or cuts power to the whole thing.

4) A fusible link in series with the heating element in case of total disaster.

barts
07-07-2009, 12:50 PM
Note that traditional PID control assumes linear behavior, bang/bang control of a heating element certainly isn't that....

If you really want tight control, pulse-width modulate the heating element w/ that solid state relay Evan suggested. Given the large thermal inertia of the elements, a 10 hz chopper frequency would be more than enough; the duty cycle on the heating element would be a linear function of the temp. error signal, with saturation (either 0% on or 100% on) occurring at a error of perhaps 5 degrees.

- Bart

toastydeath
07-07-2009, 01:13 PM
I got a chance to talk to some of the guys who worked on the first few instances of temperature control for machining awhile back. The following method is still in use at LLNL for their diamond turning machines, and I can only presume other places as well since they set a lot of the early industry standards.

Two circuits are used. One is your adjustment line, and in your case it will be heated. However you want to do this is fine. An electric heater would work, with a thermostat. Doesn't have to be particularly accurate. The water in this line circulates constantly to try and keep it homogenous. Doesn't have to be particularly swift, just get it circulating decently.

The second line is your control line. Run one end of this line to your radiators/fans, and the other to a series of liquid-liquid heat exchangers. You can make this, none of this stuff is fancy. However, this line must be pumped very quickly; the bigger lines and bigger pump you put on this, the better off you'll be. Don't bother adding tanks or whatever to add mass, that doesn't matter.

The idea goes like this. You have a solenoid, on a two or three second pulse width. You monitor your control line's temperature AFTER it has gone through your radiators. Based on this, you set what percentage of the pulse width the solenoid is open for (which may be near 0%). The solenoid, when open, allows the heated fluid to be pumped into the liquid-liquid heat exchangers. The control fluid pumps over these heat exchangers, and drops off (or picks up) whatever corrective energy is necessary to maintain a constant temperature on the radiators in the application area.

It takes some fiddling to get the variables right, but once it's set, you should have temperature control far better than .5 deg/c. The main things you'll have to play with are:

- the correlation between control line temperature past the radiators, and the application working temp
- the valve open time per tenth or hundredth degree difference in desired temp.
- The temperature of the heated fluid. The closer this can be to the desired temp, the better and the less oscillation you'll see in the system.

There's no fancy logic to this, and it works shockingly well.

JoeFin
07-07-2009, 05:37 PM
It takes some fiddling to get the variables right, but once it's set, you should have temperature control far better than .5 deg/c. The main things you'll have to play with are:

- the correlation between control line temperature past the radiators, and the application working temp
- the valve open time per tenth or hundredth degree difference in desired temp.
- The temperature of the heated fluid. The closer this can be to the desired temp, the better and the less oscillation you'll see in the system.

There's no fancy logic to this, and it works shockingly well.

Fuzzy Logic was able to learn and apply all of your values AND constantly update and fine tune them as well - thats why we used it. The more data aquired the more accurate it became.

But as I was trying to illustrate it has now gone way above and beyond the average PID controller. It now requires a Data base and it would really be nice to have more then 1 processor. You'll also want to monitor and maintain data pertaining to the ever changing outside temps as this allows the cooling or heating equipment to ramp up extra storage in anticipation of outside temperature changes


I got a chance to talk to some of the guys who worked on the first few instances of temperature control for machining awhile back. The following method is still in use at LLNL for their diamond turning machines, and I can only presume other places as well since they set a lot of the early industry standards.

and Chuck Mayfield (PhD Mechanical Engineering) learned on the 3M facility (1992)

Dawai
07-07-2009, 05:54 PM
The setting in the pid controller, called deadband.

Hope it helps..

I did a binary type arrangement through a triac once.. the diac (exciter) was fed with binary valued incremental resistors.. making it come on in variable stages.. was wonderful melting metal till lightning struck it.. Seems I used 8 resistors and one hex word to control it with a old pc running dos-basic out the parallel port.. Perhaps that was about 88?? 87? I forget..

RKW
07-07-2009, 05:58 PM
Fuzzy Logic was able to learn and apply all of your values AND constantly update and fine tune them as well - thats why we used it. The more data aquired the more accurate it became.



I really wasn't sure if fuzzy logic had stuck around or not. Really have not heard the term from anyone lately. Sounds like PID with AI added in.

JoeFin
07-07-2009, 06:09 PM
I really wasn't sure if fuzzy logic had stuck around or not. Really have not heard the term from anyone lately. Sounds like PID with AI added in.

Perhaps I'm dating myself in regards to Temp control projects with that one. I moved over to petroleum refining since then.

Non Sinusoidal Sine Waves (Harmonics) was also a hypothetical argument at the time too. But after the chill water loops shut down when the VFDs over taxed the Ckt Breakers (due to harmonics content) it wasn’t hypothetical to the many PhDs and MEEs working on the project. Damm Chuck was about to throw me under the bus in the board room until I reminded him of the logarithmic nature of what was then a hypothetical argument.

Evan
07-07-2009, 06:47 PM
Fuzzy Logic was able to learn and apply all of your values AND constantly update and fine tune them as well - thats why we used it. The more data aquired the more accurate it became.



Fuzzy logic? Har har. There ain't no such thing unless you are dealing with a ... skip it, I could get in trouble.

Ever since that term came out I choke every time I hear it. Logic is deterministic or it isn't logic at all. Even machine generated "random" numbers aren't random unless some sort of inherently unpredictable event is used as a seed and even then the following sequence isn't random, only random appearing. Making real random numbers is really difficult since however they are generated they must be sampled and the sampling process introduces a non random event. Logic can't be fuzzy for the same reasons, both in theory and in practice. At the basic level it's just a matter of using more than a single bit to represent a value. Extend that principle to the higher levels and you have what appears to be an adaptive system but it isn't fuzzy.

Even a Roulette game is predictable.



In 1981, with a computer built into my shoe, I walked into a Las Vegas casino and beat the house. This was the advent of the wearables age, when personal computers, barely invented, were already being shrunk into prosthetic packages capable of doing useful things - like taking money from a roulette table and putting it in my pocket.


http://www.wired.com/wired/archive/6.04/wearables.html

toastydeath
07-07-2009, 07:41 PM
I think fuzzy logic refers to being in two states simultaneously, but only partially in each. Not really anything to do with non-determinism? Maybe I am talking out of my ass here, but that's never stopped me before.

For instance, rather than having the chillers come on 100% right at 68 degrees when some bit flips, the chillers come up gradually as the machine goes from one state to the next. Instead of having a multitude of discrete power levels for the chillers based on set temp points (at 68, power 63%, at 69, power 64%, etc) it's easiest to just treat it as a continuum between states.

So you could have HOT, COLD, and ON TEMP as your states. The cold state would turn the chiller down to 0%, the hot state would turn the chiller up to 100%, and the ON TEMP would leave it be. But rather than being three distinct bands, there'd be transition zones. So halfway between ON TEMP and whatever point you've defined as HOT, you'd have the chiller power increasing running at 50% (or some other slope). That way, as the temperature fell back, the chillers didn't just shut off and give you an oscillating system, they'd damp it back out as the temp came back into range. So it's fuzzy in the sense that you get two "yeah, sorta" answers when you ask "Is it hot?" The machine's answer is "Yeah, sorta," instead of true or false.

Anyway, the system I know about came from the early 70's or so if I'm recalling correctly, and there definately wasn't any fancy control. I image the system as it exists now is pretty much as JoeFin describes it, with adaptive and predictive control moving the points around based on historical trends.

However, for a home based incubator, I'm not sure that a multicore box and programming the logic are really necessary. The temperature they got with just the dumb system with hardwired, basic logic was able to produce cuts with 1.5 microinch total form accuracy over a 2" bore for some electron microscope project.

And even beyond that, there's some pretty good suggestions besides this one on how to do it. I just wanted to throw something out there I know, for sure, has been implemented somewhere and was reported to be not particularly difficult.

Evan
07-07-2009, 09:33 PM
You can acomplish the same thing with less than a dollar's worth of linear circuits. Back in the early days there was a bumper sticker seen at times in Silicon Valley: "One op amp is worth a 1000 gates". Nobody realized that it wouldn't be very long and you could put a million gates in the same space.

clutch
07-08-2009, 12:21 AM
We have a couple applications that need tight temperature control. In addition to being able to heat the liquid, the unit we use has refrigeration to cool it.

There is some art to this.

Clutch

fixxit
07-08-2009, 12:41 AM
I Have repaired several laboratory systems of this type, often billed as a "temperature controlled waterbath" :
_______________________________________________
"We have a couple applications that need tight temperature control. In addition to being able to heat the liquid, the unit we use has refrigeration to cool it."
________________________________________________

The system consists of a small anemic (underpowered) refrigeration unit which cools a waterbath, and a huge temperature controlled heat source. The refrigeration system runs all the time and the huge heat source overwhelms the anemic refrigeration system and maintain the desired temperature.

Not very efficient.... but it works.

Fixxit

darryl
07-08-2009, 12:50 AM
I have some experience controlling temperature in a closed space, but not a lot. Most of the problems I had were with temperature swings because of the thermal time constant of the sensor, and the time between the heater coming on and the heat reaching the sensor. My conclusions are echoed by some of what has been said here by others.

Having a large thermal mass integral with the chamber, or the walls of the chamber itself being the thermal mass would keep the interior temperature more constant.

Heating the thermal mass with a heater only about twice as powerful as required to reach and maintain temperature would be about right. A PWM circuit with a zero crossing triac should be used, where a manual control sets the temperature. This would be a set and test thing, where you get the control to the point where the interior temperature ends up sitting close to what you want in this 'open loop' mode. Input from the sensor could now result in minor changes plus or minus from this set point. This way the heater is never really on or off, just varied a bit from your manual set point. The sensor circuit itself can be very sensitive without causing wild temperature swings of the thermal mass.

The sensor circuit basically compensates for room temperature changes, changes due to items being added to or removed from the controlled chamber, and temperature swings due to the chamber being opened and closed.

Sensing the temperature of the thermal mass itself or sensing the air very near to the thermal mass would result in better regulation of the chamber.

Outside air leakages can be counted on to mess with temperature control.

Again, I'm not an expert on this, but this is what makes sense to me.

One of the projects I did was to regulate a camera housing on a frozen food chain. My thermal mass was an aluminum box basically, with the heater being nothing more than a series of power resistors siliconed onto the aluminum surrounding the glass window. By calculating resistor values it becomes easily possible to size the capacity of the heat source to be just about right.

Evan
07-08-2009, 02:21 AM
There are couple of aspects of closed loop control systems that are counterintuitive. In particular, for a closed loop system to regulate it must have an error signal. That error signal is amplified and used to cause the state of the system to change.

The other is that the shorter the time constant of the system the easier it is to regulate the system. This means that in the case of temperature in an actively controlled system the absolute minimum of thermal mass is best. The system should be highly insulated so that the ratio of active control terms to passive change terms is maintained as high as possible. This permits the use of the least amount of gain in the error amplifier which is the most stable condition. The faster the response of the system to changes in input the more accurately the system may be controlled.

Dennis mentioned this earlier, control only the air temperature and let the thermal mass of the system take care of itself. If the system is well insulated it will be the air temperature that regulates the system temperature. Since the heat capacity of the air is relatively small the air will have a short time constant and will allow much more accurate regulation than measuring the variation of a long time constant component.

By regulating the air temperature loop gain can be kept at a minimum and the error signal at a maximum making accurate regulation far easier. The temperature excursion of the entire thermal mass of the system will be in a ratio of the thermal mass of the air to the system mass. If the air is regulated to +- 5 degrees and the system has 100 times the thermal mass the system thermal mass will vary only by 1/100 as much or +- 0.05 degrees.

Dawai
07-08-2009, 06:50 AM
PID:

Proportional output setpoint to span..
Intergral of time base to scan..
Derivative of error added into equation.

I have saw one pid controller with a auto-tune button that worked.
It takes a ramp angle corrective sense to keep a heat controller on course without significant deviation.. I typed a old Nasa guided missle math equation into a basic program and it worked better than any PID I had done in the "NON-math" oriented turbo basic language.. I think I posted the math clip over on Metalillness website.

Tuning a flow loop with pid, well.. sometimes it can keep you up all night.

JoeFin
07-08-2009, 10:53 AM
Fuzzy logic? Har har. There ain't no such thing unless you are dealing with a ... skip it, I could get in trouble.

That was one of those "Click Goggle before inserting foot in... Ah skip it...ain't worth wasting my time" moments Even

Here is a good definition of Fuzzy Logic you should become familiar with


WHAT IS FUZZY LOGIC?

In this context, FL is a problem-solving control system methodology that lends itself to implementation in systems ranging from simple, small, embedded micro-controllers to large, networked, multi-channel PC or workstation-based data acquisition and control systems. It can be implemented in hardware, software, or a combination of both. FL provides a simple way to arrive at a definite conclusion based upon vague, ambiguous, imprecise, noisy, or missing input information. FL's approach to control problems mimics how a person would make decisions, only much faster.

http://www.seattlerobotics.org/Encoder/mar98/fuz/fl_part1.html#WHAT%20IS%20FUZZY%20LOGIC?

The difference here Evan is I have experience in control schemes that maintain the spec the OP is inquiring about. You’re not going to get it in any scale and dependability with a simple PID and a couple of temp sensors

By nature mechanical sensors lag, changing ambient temperatures effect mechanical equipment, and closed loop systems have a response time. Without taking all these factors into account and anticipating their effects one would be hard pressed to attempt to maintain the spec the OP asked for

JoeFin
07-08-2009, 11:25 AM
I think fuzzy logic refers to being in two states simultaneously, but only partially in each. Not really anything to do with non-determinism? Maybe I am talking out of my ass here, but that's never stopped me before.

For instance, rather than having the chillers come on 100% right at 68 degrees when some bit flips, the chillers come up gradually as the machine goes from one state to the next. Instead of having a multitude of discrete power levels for the chillers based on set temp points (at 68, power 63%, at 69, power 64%, etc) it's easiest to just treat it as a continuum between states.

No – not from my experience in working with it.

You could spend a lot of time and effort in attempting to calculate and account for all the factors contributing to your control error. Or you could measure, record and compensate for them.

A simple Proportional–Integral–Derivative controller (PID controller) helps greatly but is still limited. By Data Logging the results in a Data Field, performing even simple averaging equations and then imputing the results into the PID equation the system continually fine tunes itself = “Fuzzy Logic”. The more data aquired the more accurate it becomes

As for the clinical definition of “Fuzzy Logic” (which is contested amongst mathematical scholars) perhaps I should not have used that “Buzz Word” But in the early 90s when I was doing a lot of this type of work it helped sell $Millions in contracts

And as far as "Mathmatical Scholars" are concerned - anyone who thinks an equation can be "Sexy" is a little off in my book too

dp
07-08-2009, 11:46 AM
There are couple of aspects of closed loop control systems that are counterintuitive. In particular, for a closed loop system to regulate it must have an error signal. That error signal is amplified and used to cause the state of the system to change.

There are third-order systems that don't require an error signal - they respond predictively and don't even use feedback. Rate control systems that I've worked on for ship-board autopilots used no rudder feedback, for example, as a second-order system does. It senses the rate at which the vessel is falling off course and applies compensating counter-rudder based on the gain of the system and the rate of change.

I suppose some could see this rate of change as an error signal, but it is quite different than one generated by a governor-controlled system or the fan-driven throttle of a lawn mower engine.

That's a bit off topic unless one is interested in building a very sophisticated controller for an incubator, and I did. It responded very well to the worst-case step response to stability: curious kids opening the hatch :)

Evan
07-08-2009, 12:15 PM
That was one of those "Click Goggle before inserting foot in... Ah skip it...ain't worth wasting my time" moments Even

Here is a good definition of Fuzzy Logic you should become familiar with




WHAT IS FUZZY LOGIC?

In this context, FL is a problem-solving control system methodology that lends itself to implementation in systems ranging from simple, small, embedded micro-controllers to large, networked, multi-channel PC or workstation-based data acquisition and control systems. It can be implemented in hardware, software, or a combination of both. FL provides a simple way to arrive at a definite conclusion based upon vague, ambiguous, imprecise, noisy, or missing input information. FL's approach to control problems mimics how a person would make decisions, only much faster.



That "explanation" is utterly useless. It provides no answer to the question asked but instead merely indulges in a lot of hand waving and then finishes by providing as an example of how it is done by a system (humans) that we don't understand at all.

Total BS.

In particular, the highlighted sentence describes a process that doesn't exist, not for machines or humans. What is missing from that statement are the words "wild guess" which IS the province of humans but not machines. The machines only use weighting factors to assign mathematical probabilities according to precisely defined rules. Like a random number generator, if given the same input twice in a row the exact same output sequence will result, twice in a row. Digital computers are wholly deterministic devices. When they stop operating as such they stop operating entirely.

Dawai
07-08-2009, 12:38 PM
Fuzzy logic is implemented by some of the newer cars that alter the vehicle tuning to suite the drivers actions..

ie: throw gas mileage out the window and under the bus when the step child borrows the vehicle.. they also remember seat positions by key-codes.. I am awaiting the day when my car calls the state patrol to inform them I am speeding and driving erratic..

Best I remember fuzzy logic was just creating solutions according to past performance and adapting on the fly. Although there may be a egghead definition that is totally different than I remember it being defined way back then..

THE more complicated you make anything, the more troubles you have from it. If it has tits or wheels, sooner or later you got problems.

A turn-left only robot is a application of fuzzy logic.. if it can't go forward, it turns left till it can.. I got a box full of them sensors somewhere..

toastydeath
07-08-2009, 12:54 PM
http://en.wikipedia.org/wiki/Fuzzy_logic

Evan
07-08-2009, 12:55 PM
Here is an actual descrition of "fuzzy logic" as it applies to logic in general.
The fuzzy logic used in computer control systems is a limited subset.


Fuzzy Logic
First published Tue Sep 3, 2002; substantive revision Sun Jul 23, 2006
The term "fuzzy logic" emerged in the development of the theory of fuzzy sets by Lotfi Zadeh (1965). A fuzzy subset A of a (crisp) set X is characterized by assigning to each element x of X the degree of membership of x in A (e.g., X is a group of people, A the fuzzy set of old people in X). Now if X is a set of propositions then its elements may be assigned their degree of truth, which may be “absolutely true,” “absolutely false” or some intermediate truth degree: a proposition may be more true than another proposition. This is obvious in the case of vague (imprecise) propositions like “this person is old” (beautiful, rich, etc.). In the analogy to various definitions of operations on fuzzy sets (intersection, union, complement, …) one may ask how propositions can be combined by connectives (conjunction, disjunction, negation, …) and if the truth degree of a composed proposition is determined by the truth degrees of its components, i.e. if the connectives have their corresponding truth functions (like truth tables of classical logic). Saying “yes” (which is the mainstream of fuzzy logic) one accepts the truth-functional approach; this makes fuzzy logic to something distinctly different from probability theory since the latter is not truth-functional (the probability of conjunction of two propositions is not determined by the probabilities of those propositions).


http://plato.stanford.edu/entries/logic-fuzzy/

Dawai
07-08-2009, 04:27 PM
Pretty much what I remember, finding a "variable truth" that fits within it's preset either hardwire or program guidelines for accepting the variable truth or false. No longer a black (0) no, or white (1) yes.. but a shade in between confirming to the guidelines defined by the process.

Wikipedia talks around the truth and explanation also, they have several confusing descriptions on the same page, with a complaint written across the top by a reader about it being vague.

The binary-analog output controller I mentioned was a fuzzy logic, if >greater than, but <less then "schedules to follow" to do a variable logic output..

THE whole input thermocouple logic table could also be interperted as a non-linear-fuzzy logic linear conversion subroutine. Orifice-flow Fluid dynamics is also a non-linear problem that needs either a square root conversion or a logic device with fuzzy logic to convert to a realistic flow range output.

Frankly, it all makes my head hurt. Every new piece of instrumentation has more cpu power than my first 4mhz 8 bit pc..

Evan
07-08-2009, 05:26 PM
"Fuzzy logic" is an academic scam that has been used repeatedly to gain a degree without actually having to do anything original. When the applicant is questioned and asked to explain he gives a fuzzy description. If called on that he then explains that it is fuzzy logic and as such descriptions of it must also be fuzzy. It then goes on with explanations of an infinite continium of possible values of rightness and wrongness between the certainties of true and false, 1 and 0 etc.

What a load of crap. Standard practice for generating pseudo random numbers is to generate values between 0 and 1. You then multiply by whatever value you wish if you need a different range, say 1 to 100.

There is no difference between a "fuzzy logic" value that is given as lying in some nowhere land between 0 and 1, true or false, and a value between any other two points on the number line. As soon as you shift the seemingly impossible and mysterious "it's somewhere between zero and one" nonsense to a different scale all the ambiguity vanishes and the logic is not at all fuzzy except in the minds of the purveyors of that silliness.

A computer has a certain degree of absolute precision which it can use to approximate mathematical values. This is determined by the type and nature of the software/hardware/both that deal with the representation of numbers, both integer and floating point. Since there are infinitely more irrational numbers than there are rational numbers the computer will always produce infinitely more wrong (approximate) answers than right (exact) ones. Mathematics is always about approximations because of this. Since a computer cannot calculate infinity infinite values must be "normalized" and assigned finite values. There is no exception to this rule, only variation in degree. As the computer can only deal in finite values nothing it does can be "FUZZY".

JoeFin
07-08-2009, 06:00 PM
Total BS.

In particular, the highlighted sentence describes a process that doesn't exist, not for machines or humans. What is missing from that statement are the words "wild guess" which IS the province of humans but not machines. The machines only use weighting factors to assign mathematical probabilities according to precisely defined rules. Like a random number generator, if given the same input twice in a row the exact same output sequence will result, twice in a row. Digital computers are wholly deterministic devices. When they stop operating as such they stop operating entirely.

No it is not a random number generator. The same form of logic was used in the military’s Fire Direction Control computers as well. The computer calculates the result, records the amount of error, and then modifies the PID based on that amount of error.

http://www.expertune.com/images/TEACH1A.GIF

In the +/-.5 deg world this is not acceptable


Best I remember fuzzy logic was just creating solutions according to past performance and adapting on the fly.

Yes David – that is accurate assessment as I understand it also

In Temp. control if one was using a Boiler Loop, the computer would monitor the amount of time it takes for the correction, (increase in temp) to return, the amount of corrective control action / resultant boiler loop temp increase, and the amount of thermal loss or gain / outside temp. The resulting corrective factors would then be placed in the their corresponding data fields with 100s if not 1000s of the simular derived corrective factors. New factors are then generated and modified based on past performance.

So it just appears that the control "learns" the system

Evan
07-08-2009, 06:30 PM
No it is not a random number generator.

???? Who said it is?

LIKE a random number generator, if given the same input twice in a row the exact same output sequence will result, twice in a row.

toastydeath
07-09-2009, 03:45 AM
I'd like to bring something up here that doesn't have to do with fuzzy logic that I was thinking of.

The whole "Error, than proportional response to correct the error" thing is why the guys at LLNL use the dual fluid circuit instead of another method. If your controller has to wait for the temperature to change before it makes a correction to the temperature, you've already lost the high precision temperature control game. Because chances are, if the control sees it go from 68.1 to 68.2, the temperature is already away and gone somewhere else in the environment.

If you have a long heat exchanger, the application air is going to get pretty close to 68 degrees before it exits the exchanger even if the incoming air changes temperature. Any fluctuations in temperature mean the coolant is going to come out warmer than it was before, but the air still has that "extra" buffer length of exchanger to drop off the excess energy its carrying and get closer to 68 than you would have with a heating system.

This is the real reason why it works so well - it doesn't have to wait for the control to pick up an error in temperature for the correction to take place. They provide a ton more fluid at 68 degrees than is necessary to cool the air to 68, so that when warmer air comes through the exchanger, nobody cares. That, and all the air in the room is going through the exchangers five or six times per minute. The machine will eventually see the exhaust line's temperature going up, and adjust the energy in the line a tiny bit. Meanwhile, life continues at 68 where it matters.

For example (if I recall correctly), DTM3 at LLNL uses over 300 gallons per minute of cooling oil. Three hundred. That's a lot more coolant that would ever be required to change all the air in the room from 68.5 degrees to 68.1, so obviously the temperature never does that.

For things where there are wild temperature swings, this method obviously doesn't work as well as the correction detection and ramping the coolers up with PID or what Joe's calling fuzzy logic. But for an environment where you mostly care about slow changes in temperature due to losses through the enclosure and temperature changes outside, it's difficult to beat.

Evan
07-09-2009, 06:47 AM
You are referring to the LODTM at LLNL?

It is temperature stabilized by a very large tank of water that is held at a constant temperature to within .0005 degree and is gravity fed to the spindle housing and the metreology frame in a 100% loss system. The absolute temperature isn't as important as the constancy of the entire volume of water since the metrology frame compensates for positional errors of the rest of the system. As long as the metrology frame doesn't change the rest is under control by it and accuracy is maintained. The rest of the system is temperature controlled by the air supply which is a factor of ten less precise than the water system.

In the case of the water system the machine metrology frame temperature isn't actively controlled at all. It's open loop and depends of the temp controlled water flowing through it to maintain the required stability. If you think about it the metrology frame can't be controlled by a closed loop system. What would it use for a reference? It IS the reference.

toastydeath
07-09-2009, 09:42 AM
Actually, I'm not referring to the LODTM. I don't know what the LODTM uses, and I'm willing to bet what you described is right. The LODTM is a pretty big departure from all the other diamond turning lathes at LLNL.

The machines I'm referring to (both DTM1s, DTM2, and DTM3) use recirculated oil (what i've been describing) as the coolant, sprayed directly on the machine and part. The only difference is they use flood spray, rather than radiators, to keep the machine temperature in check.

Some of the metrology equipment they have uses the radiator method I described instead of direct spray, since they obviously don't want oil on the measurement equipment.

Dawai
07-09-2009, 10:34 AM
ONE thing to note about the incubator..

MASS, causes the temperature swing to be slower. Ie: My home being built on a slab of concrete when it gets cold outside it takes the heat sometimes two days to come on. WHEN IT DOES, it heats the air, the air back-heats the slab.

I wonder what a slab of hydronic heated concrete would do under a incubator? if you set the controller deadband to less than the desired flutuation.

www.grainger.com had a nice lil pid 1/32nd din controller for about $150.

dp
07-09-2009, 11:28 AM
You are referring to the LODTM at LLNL?

It is temperature stabilized by a very large tank of water that is held at a constant temperature to within .0005 degree and is gravity fed to the spindle housing and the metreology frame in a 100% loss system.

I used a similar two-stage design for my incubator. The second stage had far fewer destabilizing problems (heater lag, sensor lag, lag in general) and was well insulated such that the exhaust air was pretty darn close to the inlet air for temperature/humidity. The controlled space (not very different from a convection oven in principle) had adequate capacity to deal with all issues an incubator faces except power outage. I didn't have a battery backup. But it was far better than a turkey can produce.

I was never so deep into it though that I tried to find a way to simulate the hen going off to chat it up with other hens, use the loo, or grab a snack. The eggs are completely exposed to the elements during these periods. My curious kids solved that one for me by cracking the hatch open to show their friends.

As incubators go it was entirely overkill.

Random number generators do a pretty good job, but getting a random seed is problematic. Nearly impossible, in fact.

dp
07-09-2009, 11:41 AM
ONE thing to note about the incubator..

MASS, causes the temperature swing to be slower.

Mass in a closed loop circuit introduces lag as you've observed. Lag is poison to control circuits. This is especially true for controllers that have to maintain a variable at non-fixed points, quickly. Not applicable to an incubator, but for an environmental test chamber that has to simulate a flight profile for a military jet taking off from a desert strip at sea level and get to 60,000' in less than 10 minutes, it's a big issue. So all this means is one has to consider response time needs for their particular control loop and design accordingly.

I got into it originally because I was designing single side-band radio transmitters and had a need for +- 20 for frequency stability, and crystal ovens were the method of choice (1970's). The principle influence for the controllers was heating of the system during transmission, and gyrating environmental environments (Alaska crab boats to Gulf shrimp boats).

The thing that made dealing with all the poles and zeros in the math less brain numbing was my HP calculator :)

ckelloug
07-09-2009, 12:11 PM
dp,

I think I'd modify your statement to lag that hasn't been characterized is poison to control systems. Yes, the lag will cause a pole but that isn't necessarily catastrophic.

darryl,
If you characterize the frequency response of your thermometer, you can get an instant reading by sensing the derivative of temperature rather than temperature itself. If you know the exact temp the thermometer says it is at and the rate of change, you know the exact temperature of the thermometer's surroundings. I wouldn't know this but they made us do a couple problems and a lab exercise about this when I was in school.

--Cameron

Dawai
07-09-2009, 01:04 PM
The old house I lived in had one of them gas heaters in the floor, old grate over it. Basically a vented can that had gas injected into a burner on one end, a carburetor throttle to mix the air.

TO replace the 50mv controlled valve. I could not get a replacement anywhere. I ended up putting a variable flow-gas pressure valve in..

you could turn the gas flame up, or down.. now this all seems simple to the first thought.. but.. it ran too high of a flame, would not heat the house evenly.. kick on, go to 100+ degrees for a 78 setpoint.. lower the pressure, tune it to blue tips again.. it would not heat the house during a very cold snap, stayed on most all the time..

Lead-lag-overshoot-thermostat deadband insanity.. Since I was no small guy even back then, laying in the mud underneath the house adjusting gas pressure and flame mixture was a chore that lasted for days. I can still vividly remember the "POOF" each time it lighted. It had a port where I had a inches water column guage. Reading it with a flashlight and turning the screw, then waiting on it to come back on.

My stepdaughter still has a scar from that floor grate.. As a toddler She kept running toward it, Her mom kept grabbing her.. I let her step one foot onto it then grabbed her. a tough lesson but much better than her falling headlong onto it. Hey, did you know matches burn too? Or if a kid bites you and you bite them back they understand it hurts.

Them old heaters are dangerous. My cousin still has a dress pattern burned into her leg from a wood stove.. she is my age and it happened before she was in school.

JoeFin
07-09-2009, 04:48 PM
Mass in a closed loop circuit introduces lag as you've observed. Lag is poison to control circuits. This is especially true for controllers that have to maintain a variable at non-fixed points, quickly. Not applicable to an incubator, but for an environmental test chamber that has to simulate a flight profile for a military jet taking off from a desert strip at sea level and get to 60,000' in less than 10 minutes, it's a big issue. So all this means is one has to consider response time needs for their particular control loop and design accordingly. :)



Fuzzy logic/Data Acquisition was only part of the equation. The effect of the resultant correction is calculated like a trajectory (Anticipation Algorithm) using the acquired PID values prior to being introduced into the environment. The result was a smooth non-linear curved response directly to the desired set point.

http://www.omega.com/techref/images/control.gif

Here is a link to Omega’s website discussing the process of “Fuzzy Logic modified Anticipation Algorithms”

http://www.omega.com/techref/fuzzylogic.html

Now in response to the OP’s original question


I need the incubator for running hematology tests on wildlife and domestic animals, as well as culturing microbes from infected wounds if resistance is seen in standard antibiotic therapy.....

A simple PLC Microprocessor ($100 - $200 on Ebone) with an appropriate amount of analog inputs and outputs and a dedicated PC. Preferably on a UPS.

He would need the system manufacture's software and using MOV statements to extrapolate the acquired data to a database maintained in the PC were the required algos could be applied. The modified operators would then be imputed as values to the controller.

Could be as simple as a “MySQL” database and an Excel Spread sheet to perform the mathematical operations

JoeFin
07-09-2009, 06:13 PM
A simple PLC Microprocessor ($100 - $200 on Ebone) with an appropriate amount of analog inputs and outputs and a dedicated PC. Preferably on a UPS.

On second thought I think I'm liking this I/O - PC interface i found on Omega's website

http://www.omega.com/DAS/images/OME-A-8111_m.jpg

$ 249.00 OME-A-8111

• 12-Bit A/D Converter

• 8 Single-Ended Input Channels

• 30 KS/s Sampling Rate

• One 12-Bit Analog Output Channel

• Software Programmable Gain: 1, 2, 4, 8, 16

• A/D Trigger Modes: Software Trigger, Pacer Trigger

• A/D Data Transfer Modes: Polling, Interrupt

• 16 Digital Inputs and 16 Digital Outputs

http://www.omega.com/ppt/pptsc.asp?ref=OME-A-8111&Nav=dasd01

Dawai
07-09-2009, 06:35 PM
I used something similar to data log some machinery, got myself a executive parking slot and a salary. It was a heat set machine that produced nothing but seconds till I worked the bugs out of it.
A first-out system on a carpet dye line did the second PUSH to a nice payroll. Was really big stuff about twenty years ago. That was a opto-22 422-serial com system.

What is it you are hoping to accomplish, I might have some junk around. If it is private and not a commercial job..

I start out with a I/O list with needs, then add wants if the job can afford it. I used to do internal diagnositics with timers and use one light (plc systems). I'd say at 4am, count the flashes and tell me what the number is between the pause.. Tell them how to fix it, then roll over and curse I was on salary.. before I'd charge them from the time they woke me up till the time I went back to sleep.

Evan
07-09-2009, 06:42 PM
Those happen to be almost exactly the specs for an old VIC-20 computer. Also, the VIC-20 has a video interface, ram expansion bus interface, two joystick ports and an 8 bit parallel I/0 plus a serial port and has the very unique feature of being able to run on 9 to 18 volts ac or dc input non polarized. It also has available a slew of emulators, cross compilers and and tons of other related software for the PC. Same goes for the C-64 except it requires regulated 5vdc power.

dp
07-09-2009, 08:47 PM
dp,

I think I'd modify your statement to lag that hasn't been characterized is poison to control systems. Yes, the lag will cause a pole but that isn't necessarily catastrophic.

Yes, true enough. I was thinking more of a scenario using an ad hoc method of adding mass rather than than rolling it into the math. Like throwing sacks of shot pellets into a pail until it works :) This actually does provide an empirical means of fine tuning things.