Networked Industrial Thickness Measuring System – May 01, 2004

Lime Daley's (even before the name "Lime Daley" existed) first project was to create a device that could measure the spacing between two giant rollers (40 feet long, 6 feet in diameter) in a paper mill.  Measuring the thickness of deformation of the rollers against each other (with the sensing device in between) we were able to accurately display how far apart the rollers were from each other. The idea is to make sure that the rollers are aligned properly to produce paper of an even thickness.

The existing system did the job, but could only measure at 16 points across the 40 foot span, and the customer desired more data points, but without all of the extra wires.

Lime Daley's system was a network of ATMega8 embedded controllers, each responsible for getting measurement data from 16 sensors over a two and a half foot span. Each two foot piece could be connected serially, up to the customer desired maximum of 16 pieces, though the network design was written to support up to 127 measurement devices.  There was a "master" device (an ATMega128) that talked via USB to a laptop, and to the slave devices via the RS-485 network. The RS-485 network was necessary due to the environmental noise on the factory floor.  The network protocol we used was modeled after a master/slave network protocol written for home automation, and it worked quite well for this project.  The network and protocol was easy to "sniff" for debugging purposes, and using some custom-made 485-to-USB converters, we could use multiple serial ports to watch (and inject) messages on the network.

An additional specification was that the whole system needed to be waterproof and the customer decided to use a laminator to encase the entire system in plastic.  The interesting thing about that was that the height of our components and circuit boards had to be less than 120 mils in order to fit through the laminator.  Fortunately, we were able to find components that fit under that limit, and we didn't need to use thinner (and more expensive) circuit boards.

The final system was able to take accurate measurements from all 256 sensors every two seconds.  This was well within the customer specified ten second limit.

The most interesting "gotcha" from this project was that due to the laminated wires that were used to connect each slave to each other, the wires had significant resistance (1 to 2 ohms per foot).  The result of this was that by the time we had plugged in 4 or 5 devices in series, the power and ground voltages were approximately the same voltage, and well below the 5.5V minimum that was required to accurately drive the A/D converter circuitry.  We ended up solving that by running separate power and ground lines to each board.  The network and "ripple" lines still ran on the laminated connections.

The entire system was designed to be powered off of three AA batteries or the laptop's USB connection, but it turns out that some laptop manufacturers don't follow the power specification (imagine that!) when adding USB to their laptops.  So, in some installations, the customer had to use batteries, which wasn't quite as elegant, but the system didn't use that much power, and so the batteries lasted a long time.

Another feature that we thought was neat was that the entire system didn't need to be configured with jumpers or otherwise tell each slave what his position was.  Instead, when the system was first powered up, each slave powered on in sequential order, using the one "ripple" wire, where each slave signalled the next slave to turn on and each slave simply watched the network and counted each slave as it registered, and then when its ripple wire was activated, it announced its presence on the network, and turned on its LED.  When slaves stopped announcing, the master recorded the number of devices on his network, and dynamically addresses them for the rest of the measurement cycle.  This also provided an easy way for technicians to see where any wiring connections were shorted and/or broken since any slaves with their LEDs turned off hadn't been "activated", and by watching the sequence of LEDs as the devices turned on, and devices that activated simultaneously were also easy to spot.