I'm working on a stationary system with a conical vessel that holds about 5 gallons of water. That will be maintained at 70 - 90F for 5-7 days. To be clearer, depending on what is being 'processed,' the system will maintain a set temperature, for example, 77F, over that period. About every 10 hours, the water will be drained by gravity to about half the volume, and freshwater will be added to return to the target water volume.

We have a fairly low-cost budget for this device, so I'm trying to use a simple, compact load cell from TE (FX29) connected using I2C to the microcontroller. I am getting substantial errors between what we should read and what is read. The readings from the load cell are always less than what has been added to the tank. So I am fairly convinced that the problem is transfer of force at supporting points that is not being measured by the load cell. I am trying to 'float' the tank by having it sit directly on the load cell at the bottom and then using four 1" ball transfers (see link) spaced at 90 degrees around the tank near the top. I adjust these so that the tank is level across the top.

Is there a better way to support the tank in reducing load transfer to points off the load cell?

Just a bit more information that might be helpful. The unit is in a room-temperature lab that has very little temperature fluctuation. In these initial tests, the water being added is at room temperature and is tap water. We are not heating or cooling it in the tank. There is no stirring, aeration, or agitation currently. I'm trying to keep it simple until we figure out accurate water volume measurements.

 

Taking temp out of the equation your best approach (at least from an industrial perspective) is to mount the framework on a four load cell arrangement, one in each corner and then connected in a wheatstone bridge configuration. But rather than trying to configure it for a bridge configuration you would be better off connecting each load cell to an individual channel on load cell amplifier. The reason for using an amp is because the amp will deliver a constant and precise excitation reference and it will amplify you feedback signal (which is usually in microvolts or millivolts) so that you can output in the usual industrial analog references of 0-5v, 0-10v, 4-20mA. Your microcontroller (Arduino, PIC etc) can handle these voltages much easier than it can handle the direct output from the load cells.

 

At this point, we do not expect temperature to be an issue. We've begun to do exactly what you have described. We now have four of this load cell. They are I2C devices which eliminates external bridges, amplifiers and A/DC. The load cell we are using is the TE Connectivity FX29 series. An extra advantage of the four load cell approach is that we can use devices with lower full load ranges (10lbf vs. 50lbf) so we get a higher resolution. The disadvantage of course is cost but the cost is immaterial if the thing won't work accurately. Unfortunately, the FX29 off-the-shelf units only come in on I2C address so we'll have to do a bit of multiplexing to use them but such is life. Thank you for your advice. You confirmed what we decided to do. FYI - this device is completely standalone and will not be used in a factory with other systems so that also gives us some flexibility in interfaces.