Sous Vide: Recipes, Techniques & Equipment (Part 5)

[Robert Jueneman]

Back to the overshoot question. My laptop is at work, so I can't run the data logger, but I tried my 1500 watt 10 liter rice cooker with my SVM 1500B (in Celsius mode) using P=100, I=100, D=0.

The temperature climbed from 28C (well above ambient, but the water was still warm from last night) to the set point of 55.5C in about 10 minutes. Unfortunately, it kept climbing, and hit 59.4C 10 minutes later. 45 minutes after I started, it was still at 57.0.

That's not very user-friendly, so we'll write off that combination. I assume my previous suggestion of I=125 for a similar cooker would produce comparable results.

EDITED: After 1:20 minutes, it was back down to 55.3, or 0.2C below the set point, which is the problem I was trying to eliminate in the first place! EDITED #2 -- Now it's back up to 55.7, so it is oscillating, but not too badly -- unless it gets worse.

So... what do we know? Assuming the cookers are reasonably comparable, I=100 is too low (it gets too hot too fast), but I=400 doesn't work either -- it may produce oscillations, and takes forever to stabilize.

I guess I'll split the difference, and try I=250. If that still is ridiculous, I'll try I=1, and see what happens.

And yes, e_monster, I will also try the auto-tune settings (P=13, I=900, D=300 -- as close as I can get by re-entering them), just to see what they do as well, even though they seem outlandish.

Edited April 29, 2009 by Robert Jueneman (log)

[blackp]

As per e-monster's suggestion my machine is now autotuning. As suggested I have nothing to lose.

It's been going for about 5 hours so far and I'll let it run until it stops. I'll provide more information as it comes to hand.

If the autotune doesn't give me better results my next attempt will be 50% power. There doesn't seem to be any problem with the unit heating up - even from room temperature it is quite quick, the main problem seems to be the latent heat in the unit after the element has been turned off, coupled with good insulation so that the heat doesn't go out of the unit quickly. After that increasing the "P" value will be my next step.

Peter.

[nickrey]

I have been through this exercise with the SVM as a consequence of overshoots very similar to those described in the posts above.

Using a smaller JNP-1800 10-cup cooker, I found the following settings work well.

The PID controlled advanced settings were used to set the unit power to 75% to moderate the speed at which the unit heated.

The PID settings that I arrived at after much experimentation (plus a rejection of the auto tune results which still resulted in overshoots) are P 100, I 60 and D 80. These seem to hold the temperature rock steady over the entire cooking period.

The other variable that seems to affect reaching temperature and temperature maintenance is initial water temperature. When filling the cooker, I run the tap water until it is at a temperature around 2C higher than my target temperature. On adding the cool sous vide packages, the temperature is typically near the target so the heating requirement is not as dramatic. The unit thus only has to maintain a constant temperature rather than heat to target.

[blackp]

The autotune of the SVM finished just now - 11.5 hours after starting.

It came up with: P=19, I=998 and D=499.

Intuitively that sounds wrong.

If I get a chance tomorrow I'll give it a go to see what happens.

Cheers,

PB

[nickrey]

Sounds very similar to the values I got.

[blackp]

OK - here is what it looked like with the Autotune settings:

This would be a good result if only the set temp had been 57C and not 54C.

Next experiment I'll try Nickrey's settings.

Only thing on my mind is that my cooker is twice the size and has more wattage so the settings may not be appropriate.

I guess another alternative could be to re-calibrate the SVM with a 3 degree offset and keep the autotune settings, although those spikes don't look good.

One thing I observed is that the insulation is very effective - I turned it off last night when the temp was 56.8C, and 6.5 hours later when I got up this morning the temp was still 43C even though the ambient temp was 17C.

Cheers,

Peter.

[blackp]

More information:

After the Auto-tune experiment I tried nickrey's settings (P=100, I=60, D=80, Power 75%). The overrun was similar or worse than the default settings (P=100, I=700, D=40, Power 100%). I aborted that try, as it clearly didn't fix my large overrun problem.

Next I doubled the P setting to 200 (P=200, I=60, D=80, Power 75%) and the result was better, but the overrun was still too high.

The last experiment for today was to eliminate the I and D settings (P=200, I=0, D=0, Power=75%). While it took 20 minutes for the target temperature to be reached, once there it was very stable within a few 10ths of a degree.

Buoyed by this new stability I got an ice brick from the freezer and put it into the cooker at the 75 minute mark. The ice brick weighs 475g and the cooker was filled with 9 litres of water - that equates to the ice brick being only 5.3% of the total mass in the cooker.

After 5 minutes the temp had dropped 2.5C and stayed between there and a further 0.5C lower for 15 minutes. It then took a further 13 minutes to recover to the target of 54C.

I don't know if this is expected, but when time permits I'll repeat the same experiment with the "D" setting increased to see if the time spent hovering around a lower temp can be overcome.

More info as it comes to hand.

Cheers,

Peter.

[nathanm]

Buoyed by this new stability I got an ice brick from the freezer and put it into the cooker at the 75 minute mark.  The ice brick weighs 475g and the cooker was filled with 9 litres of water - that equates to the ice brick being only 5.3% of the total mass in the cooker.

After 5 minutes the temp had dropped 2.5C and stayed between there and a further 0.5C lower for 15 minutes.  It then took a further 13 minutes to recover to the target of 54C.

←

The latent heat of fusion for ice is 334 kJ/kg

So, melting 475 grams of ice takes 158,000 joules of energy.

Taking 475 grams from -20C to 54C is 146,000 joules.

So melting the ice and bringing it up to temperature would require 305,000 joules.

It takes 4.18 joules to change the temperature of 1 gram (1 cc) of water by 1 degree C.

So if you just dropped the ice in without the heating element on, the temperature of a tank with 9 literes of water should drop 8 degrees C, by the time the ice was all melted.

But you did have a heating element on, and 1 watt = 1 joule/sec.

So, a 1500 watt heating element would take about 3.6 minutes to put 305,000 joules into the tank at 100% efficiency.

The efficiency is probably no more than 50%, and the heating element probably was not on high heat the whole time because the PID controller was gradually regulating the heat to avoid overshoot.

Remember that the PID only "saw" a 2C to 3C temperature drop, so it would not turn the element on to full power. The P parameter determines how the power goes proportionate to temperature difference.

So, your result on the dropping the ice into the tank is reasonbaly consistent with all of this. If you could record the PID % power output it could all be calculated very closely.

If you had dropped crushed ice in, you would have gotten a lower temperature spike on dropping it in. The ice brick took a while to melt, that is why the temperature was on a plateau for 15 minutes.

[joesan]

Robert (Jueneman) - thanks for the interesting observations around the SVM controllers. I'm thinking about adding to my SV detritus by adding a SVM controller. I really want accuracy to the 0.1C level with this next one.

I can see on the SVM site that the 1500B model you mention has resolution of 0.1C but accuracy is only 1%. This is obviously a big difference.

1% doesn't seem much but it could have a non-negligible effect over prolonged periods of time I guess.

In your testing have you been able to confirm that this model is more accurate long term than 1%?

[e_monster]

With I and D off, I think you can probably lower the P value. As a test, run it in just P mode with the P value you got from Auto-Tune and if that doesn't give satisfactory results double the P. This should also increase the recovery rate.

I wonder if the weird behavior after Auto-Tune (stable but several degrees too low) could have anything to do with running the power at less than 100%. Maybe there is a software bug that causes problems for the the auto-tune if the power isn't 100%.

More information:

After the Auto-tune experiment I tried nickrey's settings (P=100, I=60, D=80, Power 75%).  The overrun was similar or worse than the default settings (P=100, I=700, D=40, Power 100%).  I aborted that try, as it clearly didn't fix my large overrun problem.

Next I doubled the P setting to 200 (P=200, I=60, D=80, Power 75%) and the result was better, but the overrun was still too high.

The last experiment for today was to eliminate the I and D settings (P=200, I=0, D=0, Power=75%).  While it took 20 minutes for the target temperature to be reached, once there it was very stable within a few 10ths of a degree.

Buoyed by this new stability I got an ice brick from the freezer and put it into the cooker at the 75 minute mark.  The ice brick weighs 475g and the cooker was filled with 9 litres of water - that equates to the ice brick being only 5.3% of the total mass in the cooker.

After 5 minutes the temp had dropped 2.5C and stayed between there and a further 0.5C lower for 15 minutes.  It then took a further 13 minutes to recover to the target of 54C.

I don't know if this is expected, but when time permits I'll repeat the same experiment with the "D" setting increased to see if the time spent hovering around a lower temp can be overcome.

More info as it comes to hand.

Cheers,

Peter.

←

[Robert Jueneman]

A fascinating series of posts, from all over the world.

Peter, I won't say that the I value is useless, as I haven't exhausted all possibilities -- especially for smaller cookers. But I've tried I=1, I=10, I=100, and I=250, and all of them have resulted in excessive overshoot, and in most cases takes too long to stabilize.

My previous tests with the 10 liter cooker showed that P=50 and P=75 produced too much overshoot, while P=100 produces maybe 1C overshoot, and P=200 takes too long to come up to temperature, so P=100 seems pretty close to optimum (with I=0, D=0). If you were cooking a small amount of something, e.g., a small pice of salmon, where he overshoot might matter, you ought to wait for everything to stabilize. But if you throw in a 500 g frozen steak, the overshoot would actually help.

So far, I have not been able to eliminate about a 0.2 to 0.3C long-term undershoot with that combination of PID parameters and the 1500B. Of course, with the 1500A or 1500C, you would never see it. Maybe I'll just give up and change the set point.

When you said an "ice brick" did you mean real ice, or one of those "Blue Ice" brick-like sealed containers? I'll be very interested to see the D values you come up with -- I'm going to be doing the same thing.

Joesan, let me differentiate between four different terms:

Resolution refers to the minimum observable difference in readings of settings. The 1500B offers 0.1C resolution. The 1500A and 1500C offers 1 degree resolution -- F or C, take your pick. 1C is too coarse for sous vide, at least to my way of thinking, but 1F is probably OK.

Accuracy (in my terminology, at least) refers to the relatively short term repeatability at a single reference temperature. In other words, if the SVM displays XX.Y today, and I calibrate the offset to match my reference thermometer within 0.05C, will it display the same result tomorrow? So far, it does. As to long-term accuracy/stability, ask me in a year from now!

Linearity (again, my terminology) means the difference between the SVM and my reference, across a wide range of temperatures. As I recall (I'm at work and don't have my notes in front of me), the 1500B had a linearity of plus or minus 1.5C from 38C to 85C, with the larger differences occurring at the higher temps. By contrast, the 1500C, which has a probe that is better at measuring higher temperatures but only offers 1F resolution, had significantly better linearity -- within 1F over the entire range. Its probe is also less likely to be damaged by very hot (boiling) water. Unfortunately, the 1500A/B probe is not interchangeable with the 1500C.

So...

To date I use the 1500B when cooking meat at 55.0 to 55.5C, because even 0.5C makes a significant difference in doneness; and I use the 1500C at 83C for vegetables, which (so far as I know) are not all that sensitive to the precise temperature.

Would I like a device with a probe that was dishwasher safe, would withstand boiling as well as freezing, and would be accurate to 0.1C or better across the entire range? Yes.

Would I pay $2000 each for two PolyScience circulators that would do that, if in fact they will (and I have yet to see any accuracy claims or test results that would substantiate that)? Not until I win the lottery!

And now a question. Has anyone tried inserting one of the Taylor one-piece digital dial-type thermometers into meat, and then sealing the entire thing in the sous vide bag? I'm not certain whether the entire thermometer will withstand high heat, but it ought to hold up to 55C OK.

Is there any obvious advantage in measuring the internal temperature, or are Douglas Baldwin's tables good enough?

And wouldn't it be cool to have a probe with an integral RFID chip in it, for a remote readout?

[joesan]

Thanks Robert - I'm aware of the difference between resolution and accuracy. For me the bottom line has to be accuracy after all what's the point of your nice display showing 54.3C if your actual temperature is 58C. If you notice the specs for most controllers tend to have a larger percentage accuracy for F than C, basically because F is more granular.

I think that for SV the most critical range that has to be accurate is between 48C to 55C or 60C with 0.1C accuracy. Or obviously the equivalent in F. Outside of that I don't think the temperature has to be to so precise. Seems like this is obtainable with the SVM 1500B from your experiments.

As regards the Polyscience circulators I think they're nice to have but not strictly necessary if you can get to 0.1C with a much cheaper option. The extra $1500 you have left over buys a hell of a lot of Pata Negra and fine cuts of beef...

[Mallet]

I would argue that stability is just as important as accuracy, actually it's probably more important (you can always correct for systematic error ).

[Robert Jueneman]

joesan, I hope I didn't insult you by defining my terms. I sometimes forget that on this blog, at least, there are probably more science geeks than cooks!

And Martin, I agree that stability is more important than accuracy, assuming you do adjust for systematic error. However, if you are using your equipment over a broad range of temperatures, linearity may be even more important.

[e_monster]

Robert Jueneman wrote:

My previous tests with the 10 liter cooker showed that P=50 and P=75 produced too much overshoot,

Was this with I and D 0?

(Oh, it is also worth mentioning whether you are using the unit in C or F mode since the values of P I and D are related to the mode that the unit is operating in)

[Robert Jueneman]

Good point. P=180 (F) or 100 ©, with I=0 and D=0. The 50 and 75 values were for the 1500B, which I only use in Celsius.

[blackp]

Robert (Jueneman)  - thanks for the interesting observations around the SVM controllers. I'm thinking about adding to my SV detritus by adding a SVM controller. I really want accuracy to the 0.1C level with this next one.

I can see on the SVM site that the 1500B model you mention has resolution of 0.1C but accuracy is only 1%. This is obviously a big difference.

1% doesn't seem much but it could have a non-negligible effect over prolonged periods of time I guess.

In your testing have you been able to confirm that this model is more accurate long term than 1%?

←

When I was researching my purchase Frank Hsu from SVM recommended the 1500C over the 1500B. It has a newer type of thermocouple which is more accurate at typical Sous Vide temps. Even though 1500C only displays whole degrees as opposed to the 1500B which displays 10ths, he said that with less error from the sensor overall you get a more stable result. I was surprised that he was recommending a cheaper alternative. Unusual for a company to "down sell".

Also the 1500B model is Centigrade ONLY - the display is not wide enough to enable Fahrenheit.

Regards,

Peter.

[blackp]

See my comments in red below:

A fascinating series of posts, from all over the world.

Peter, I won't say that the I value is useless, as I haven't exhausted all possibilities -- especially for smaller cookers.  But I've tried I=1, I=10, I=100, and I=250, and all of them have resulted in excessive overshoot, and in most cases takes too long to stabilize.

I will try keeping P=200 as that got me to nearly zero overrun.  I want to try the same settings @ 100% power, also applying some "D" value to see if it helps counteract the drop when adding food.  In theory it should as it reacts to rate of change rather than amount of change (that is the "I" value).  In any case I will make a couple of tests - I don't want to change too many things at once or I will not be able to tell why things changed.

My previous tests with the 10 liter cooker showed that P=50 and P=75 produced too much overshoot, while P=100 produces maybe 1C overshoot, and P=200 takes too long to come up to temperature, so P=100 seems pretty close to optimum (with I=0, D=0). If you were cooking a small amount of something, e.g., a small pice of salmon, where he overshoot might matter, you ought to wait for everything to stabilize. But if you throw in a 500 g frozen steak, the overshoot would actually help.

So far, I have not been able to eliminate about a 0.2 to 0.3C long-term undershoot with that combination of PID parameters and the 1500B. Of course, with the 1500A or 1500C, you would never see it.  Maybe I'll just give up and change the set point.

I am using a 1500C but measuring temperature with a Thermoworks thermocouple thermometer.

When you said an "ice brick" did you mean real ice, or one of those "Blue Ice" brick-like sealed containers?  I'll be very interested to see the D values you come up with -- I'm going to be doing the same thing.

What I added was a plastic bottle nearly filled with frozen water.

Joesan, let me differentiate between four different terms:

Resolution refers to the minimum observable difference in readings of settings.  The 1500B offers 0.1C resolution.  The 1500A and 1500C offers 1 degree resolution -- F or C, take your pick.  1C is too coarse for sous vide, at least to my way of thinking, but 1F is probably OK.

Accuracy (in my terminology, at least) refers to the relatively short term repeatability at a single reference temperature.  In other words, if the SVM displays XX.Y today, and I calibrate the offset to match my reference thermometer within 0.05C, will it display the same result tomorrow?  So far, it does.  As to long-term accuracy/stability, ask me in a year from now!

Linearity (again, my terminology) means the difference between the SVM and my reference, across a wide range of temperatures.  As I recall (I'm at work and don't have my notes in front of me), the 1500B had a linearity of plus or minus 1.5C from 38C to 85C, with the larger differences occurring at the higher temps.  By contrast, the 1500C, which has a probe that is better at measuring higher temperatures but only offers 1F resolution, had significantly better linearity -- within 1F over the entire range.  Its probe is also less likely to be damaged by very hot (boiling) water.  Unfortunately, the 1500A/B probe is not interchangeable with the 1500C.

So...

To date I use the 1500B when cooking meat at 55.0 to 55.5C, because even 0.5C makes a significant difference in doneness; and I use the 1500C at 83C for vegetables, which (so far as I know) are not all that sensitive to the precise temperature.

Would I like a device with a probe that was dishwasher safe, would withstand boiling as well as freezing, and would be accurate to 0.1C or better across the entire range? Yes.

Would I pay $2000 each for two PolyScience circulators that would do that, if in fact they will (and I have yet to see any accuracy claims or test results that would substantiate that)?  Not until I win the lottery!

And now a question.  Has anyone tried inserting one of the Taylor one-piece digital dial-type thermometers into meat, and then sealing the entire thing in the sous vide bag?  I'm not certain whether the entire thermometer will withstand high heat, but it ought to hold up to 55C OK.

The Thermoworks device I have can be used to pierce a bag of food through some high density foam tape.  the tape seals around the needle probe and the bag remains sealed.

Is there any obvious advantage in measuring the internal temperature, or are Douglas Baldwin's tables good enough?

And wouldn't it be cool to have a probe with an integral RFID chip in it, for a remote readout?

←

[Robert Jueneman]

Peter, this is all true.

Frank is justifiably concerned about the basic accuracy and linearity of the probe, and according to my tests the 1500C is demonstrably better in that regard.

However, I bought the 1500A originally, and then I got both the 1500B and 1500C, and I use each one for the purpose for which it is best suited. Even though I'm more used to thinking in Fahrenheit (what a ridiculous, obsolete scale!), the additional accuracy and resolution of the 1500B is well worth it, I think, even for those of us in the backwards, non-metric United States, and especially when cooking meat at relatively low temperatures where the additional resolution/accuracy is more important. Sigh.

Unfortunately, my 1500A's sensor became erratic, perhaps because I tried to calibrate it with boiling water -- don't do that! And a young kitten chewed through the 1500C sensor's cable while it was sitting on the floor. So at the moment, I'm checking the mailbox daily, and waiting for some replacement/spare probes to arrive.

Re the "down selling." Frank is or was a chef, and his wife still operates a barbeque joint. So he knows the theory as well as the practice. I doubt that he is making millions or even thousands off the SVM, but I admire his intellectual honesty, and I am happy to recommend his products and the company. May his tribe increase.

[joesan]

I guess all the parameters matter to a degree (no pun intended).

Robert I took no offence at all, these forums are for learning and discussion, it's just that the actual point that I was making was that one should take care to avoid confusing a high resolution display of the temperature with it's actual empirical accuracy.

But now I'm confused again because of SVM recommendation of the 1500C. You've used them all, Robert, - which do you recommend as best able to accurately maintain 54.4C over 8 hours say?

[blackp]

More experiment results:

I firstly tried increasing the power to 100% (leaving the P=200, I=0, D=0) and the result was about a 30% decrease in the time to come up to temperature, but also about an overrun of the same magnitude. Probably should have been expected. As the temp stayed between 0.5C and 0.75C over the target for a while I aborted that trial as the 75% power setting got me much closer to the target consistently.

Next I tried going back to 75% power and setting D to 40 - I put the same plastic bottle of ice into the cooker at the same time (1:15). Not a big difference.

The astute will see that my starting temps were not the same so I'm not certain that that fact didn't have an impact.

I'm giving up manual temp tracking for now. Apparently my USB data logger is only about 2 weeks away. Then I can remove any "stupid human" errors and keep the test running for longer.

Cheers,

Peter.

[blackp]

I guess all the parameters matter to a degree (no pun intended).

←

You are correct with respect to the "P" parameter - it matters to a degree.

The "I" and "D" parameters relate to time so they matter for a while.....

java script:add_smilie(%22:wink:%22)

[Mallet]

Taking this thread back to cooking for a bit, has anyone experimented with SV mushrooms? slkinsey did some SV shiitakes (they were in a bag with chicken and scallions), but other than that I haven't been able to find much in the way of guidelines/recipes. For example, Under Pressure has not a single mushroom preparation sous-vide, as far as I can tell.

Is it just a waste of time? Perhaps proper browning of mushrooms is too integral to their flavour. Wild morels are coming up soon, and I don't want to waste any (assuming, of course, I manage to find any!).

[slkinsey]

Some questions for the data-loggers:

1. Am I correct in assuming that these are setups which do not have active circulation?

2. To what extent are you testing for consistency of temperature throughout the water bath?

3. Are any of the data-loggers seeing accuracy and consistency markedly better than what blackp with their non-circulating PID setups?

4. To what extent does the recorded temperature accord with the tempurature displayed by the PID?

[Robert Jueneman]

Joesan, once the SVM stabilizes, it seems to hold steady no matter which unit (1500A, 1500C, or 1500C) you use. I often cook things for 48 hours, and only occasionally do I see the displayed temperature vary by a single digit. (I still don't have the kinks work out in my data logger to automate some of these measurements.)

So the primary question is one of resolution, rather than stability. If you want to hold the temperature at 54.4C, rather than say 54.1C, then the 1500B is your only choice. If 1 degree

F is close enough, than any of the three ought to be good for you.

I would not be particularly surprised to find out that the three units all share precisely the same PID engine, and that the only difference is the display, and the ability to specify fractional degrees. In other words, the 1500C might be just as stable as the 1500B, i.e., within 0.1C or 0.2F, except for the fact that you wouldn't know and wouldn't be able to specify WHICH fraction of a degree it was shooting for.

I do see that with P=100, I=0, D=0 on my 1500B, the SVM stabilizes at about 0.2C lower than the set point, for reasons I don't understand. Maybe the power output at that point just isn't enough to keep up with the heat loss, when I am using a circulator pump and have only a sheet of Lexan over the top, rather than closing the lid. (This is so I can continue to measure the temperature with the Traceable 4000 thermometer.)

In my still limited experience, 1F or 0.5C seems to make an observable difference in the doneness of beef or lamb, salmon, or the perfect egg, but I'm not sure anything less than that is significant.

What has the experience of others been -- is tighter control beneficial, or even noticeable in practice?

slkinsey, although I don't yet have my data logger set up to completely automate the data logging, I am using an inexpensive submerged garden fountain pump for circulation, although I have never seen any significant difference or hot spots without it. The SVMs probe is normally sitting on a perforated metal disk about two inches off the bottom of the cooker, and the reference thermometer is suspended through a hole in a Lexan sheet and touching the same metal disk with its tip.

Once everything has stabilized, they agree within the limits of resolution and rounding error.

When using the I parameter, I sometime see differences between the display and the thermometer, but I haven't fully characterized them. I tend to think that what is being displayed may be the result of the integrated value, rather than the direct readout from the probe, but I don't have any way to confirm that.

Bob

Edited May 2, 2009 by Robert Jueneman (log)