Reliability of low-cost accelerometers in commercial applications

[marellasunny]

My question relates to the reliability of low-cost accelerometers (MMA7455,ADXL345 etc.) in commercial products with a expected life expectancy of 5 years or more. The tilt angle reading from the accelerometer will be used as a feedback parameter to orient a single-axis plate-reflector positioning mechanism. From testing, I gather that the tilt sensitivity of both accelerometers are coarse but this doesn't really affect my positioning accuracy that much. So, sensitivity isn't a big issue.

The accelerometer is placed in a IP65 box to protect from external weather conditions. The box is always shaded by the reflector plates are is not directly exposed to weather per say. Temperatures can vary between +15 degrees Celsius to +45 degrees Celsius. I use I2C bus protocols to communicate with the accelerometers.

Price:
These accelerometers cost around $5.
I am willing to spend upto $30 because service costs can dig a hole into my budget. ADXL103/203 seems to be a good alternative but then shipping costs to my country cost more than the product itself!

My second query-
Redundancy in engineering design: I have yet to come-up with an alternative in-case the accelerometer fails. Maybe using 2 accelerometers would do the trick? Please advice.

 

[Aaron Crowl]

If this application is critical (like aircraft attitude) I wouldn't trust a single cheap accelerometer. Home electronics that use cheap accelerometers accurately typically have more than one chip which produces an average. Even game controllers tend to have this setup. As for reliability, you may need the advice of an industry expert. It's hard to answer that question without some resources like a large product deployment and customer feedback database.

I put "accelerometer design" into a Google search and found somebody called Willcoxon who makes industrial grade accelerometers. I bet their units are guaranteed to be reliable but probably not cheap.

 

[marellasunny]

Home electronics that use cheap accelerometers accurately typically have more than one chip which produces an average. Even game controllers tend to have this setup.

Thank you for the reply. No this is not a critical application like aircraft systems. Although, if accelerometer did fail in my system, errors would accumulate over a period of few days and system positioning would become highly inaccurate. No safety issue though.
The reason 2 cheap accelerometers are used in home electronics (instead of 1 ) is to make sure device still functions despite 1 accelerometer fail?

I found a good accelerometer IC 'ADXL103'. But, I would still like to explore your idea of using 2 accelerometers on the I2C bus as it provides failure redundancy.

 

[Baluncore]

If you fit two accelerometers where one might be expected to fail at some time, then you need some way to tell when one has failed. You would need to read the two accelerometers at the same time, then compute the sum and difference. Half of the sum gives the average, while the difference detects a failure.

Any failure should flag a repair alarm, but it would be good if the software could reliably fall back to the one that still works until the repair is done. If the accelerometers are identical then the module with both should be replaced when either one fails as the other is probably from the same batch with the same history and life expectancy.

If it was a critical application, then three accelerometers with independent processors would be used to vote on which accelerometer was reliable.