Active Beer Cooler Sous-Vide Setup w/Machine Learning Support from an Amazon Web Services Cluster

The second iteration of our bootleg beer cooler sous-vide system is coming online now, and a lot of thought has driven the design of a new system.  First, we still have a design goal to make the beer cooler sous-vide system consist of everyday household things that cost very little. However, it became clear that our testing system needs more power.  In this post, I’ll spell out the new hardware and what we’re using it for.  In a future post, I’ll report back on the testing protocol.

The Parts Manifest

Internet Explorer Blanket + Beer Cooler + Bubble Wrap + Hot Water + A ThermoWorks TW-USB-TC-LCD data logger + One type 113-182 Thermocouple Penetration probe + Macbook Pro computer running Parallels + ThermoWorks EasyLogUSB Software + an Amazon Web Services Cluster running a variety of machine learning and graphing applications.

The Bubble Wrap

One of the first breakthroughs in our home sous-vide system is recognizing the bubble wrap forms an effective insulation layer on the water.  The idea for this came from managing the temperature of my pool using a solar blanket.  Regardless of the design of the solar blanket, as long as a layer of plastic lays across the top of the pool, you get a substantial increase in heat retention — as much as a swing of 10°F.  On the sous-vide setup, we are just beginning the testing, but we appear to be able to retain heat after the insertion of food much better than before.  So this is a substantial breakthrough in temperature management, especially if the bubble wrap doesn’t degrade when exposed to high temperatures.

The ThermoWorks Data Logger and Probe

The new NIST certified data logger and temperature probe allow us to capture temperature readings to a computer at a rate of one measurement per second.  The probe + device have a 2 to 3 second start-up time and are accurate to +- 2°F (NIST certified).  A shout-out to ThermoWorks for helping us with the measurement technology.  The free ThermoWorks software allows the probe readings to be imported into a CSV file (or an Excel spreadsheet) for processing.

The probe and software can take a maximum of 32,000 readings per experiment at variable resolution.

The Machine Learning Software

Once the data is imported into the Macbook Pro, it is SCPed into my AWS cluster for processing.  Using supervised learning, we’ll estimate the thermal loss of the system at various temperatures.  We’ll use this system to measure differences in the insulating properties of various water bath vessels, the impact of starting and sitting temperatures, and to develop water replacement strategies.

Once we have run a number of experiments to collect training data, the system will be able to tell us (or an Arduino Board microcontroller) the exact parameter expectations for water heating/replacement for a sous-vide cooking recipe.

Wrapping Up 

We’re excited about the new setup and I look forward to giving a report late next week on the usefulness of the system.  We’re also searching for an intern to help us build and test this system.  It should be a great “big data” and machine learning manipulation experiment for someone interested in learning these skills.

17 notes

Show

  1. spurpura posted this

Blog comments powered by Disqus