I wonder how much error logging for purpose of maintenance is used routinely for complex systems by experienced users?

It depends on how catastrophical the consequences of a failure are. In some sensoring applications the extensive use of plausibility checks is essential. Assume for instance, that throttle position of car engine is 'told' by throttle potentiometer to be 'maximum open'. Further assume, that this is not real, but caused by a broken cable. Would you give full power to the engine??
Well, fortunately the pressure in the inlet channel is measured too, and by the help of this (and other!) additional data a plausibility check can be made, whether this 'full gas' condition is real or not...

Another example:
Space shuttle contains more than 1 million different sensors and more than 5 different computers (I hope my knowledge is still up to date). Estimate the percentage of plausibility checks which are made each second by these 5 computers...

It's the plausibility checks and the further going interpretations of collected data in order to valuate the actual state of project or application which gives embedded systems 'intelligence'.

In the future multi microcontroller systems will be used to control safety relevant applications, just in order to increase redundance. But it's difficult to keep the multi microcontroller systems independant from each other, so that a failure of the one is not negatively affecting the others, but at the same time give the proper functioning ones the possibility to cancel or correct the failing ones...


Kai