I think the idea of digital summing is indeed the best. I would look at a system structure wherein a single multi-channel A/D converter was used to gather the digitized readings from the outputs of the individual amplifiers. You should be able to compensate and calibrate better around one A/D instead of three or four of them. The multi-channel A/D can be simply placed onto an analogue section of the controller module. If you have concern about noise between this analogue section and the digital part of the controller then run the two sections on separate voltage regulators, use an A/D with a serial SPI type interface and optically isolate the few interface signals between the A/D and the MCU.

As Kai has pointed out, the software of the controller can take the readings from the appropriate A/D channels and use math to add and average the readings instead of trying to do that in hardware or separate MCUs. When Kai mentions that a plausibility check is possible when the individual channels are brought into software separately he is referring to the following types of checks.

1) Checking that all the load cell readings are withing a known and acceptable range of each other.

2) Checking if the reading on one channel does not change when it did on the others.

3) Checking if any channel is running flat line at or near min or max scale range.

4) Checking to see of some pairs of channels seem to have the same reading (within a small tolerance band).

All of the above will cover various types of errors with the load cells, their connectivity and their channel input amplifiers.

If you still want modularity you can put the channel input isolation / gain amplifier onto a small channel input board that would piggy-back onto our control board.

Michael Karas