About Reliability

All engineered systems follow a recognized pattern of failure known as the 'bathtub curve'
.


The timescale for this curve is typically 10-20 years for electronics and 1-5 years for mechanical systems depending on stresses and the purpose of the system. A car, for example, might be expected to show signs of wear after, say 60,000 miles which at an average of 30 mph is only 2000 hours or about 3 months of continuous use. A machine in a production environment might be expected to work 3 shifts a day for, say 2 years before showing signs of wear which would be about 12,000 hours.

In the early part of a system's life things often go wrong. This accounts for the high failure rate in the early history and it reduces rapidly as time passes. This period is often referred to as infant mortality. In almost every case such failures are a direct result of manufacturing defects, for example a poor connection or a screw not tight enough or a part incorrectly made or fitted. No-one is immune although the motor industry has it almost beat. In small scale production as in the production of robots the answer is to cycle the system for as long as possible before delivery. Any defective component will fail during this period and not after it is in use by the customer. ST robots are run day and night for a week before delivery, which is not much but the curve is very steep at this stage. There is still the possibility of infant mortality after delivery but this is very much reduced and is covered by the warranty period. Infant mortality post delivery is usually quantified as a percentage i.e. percentage of units delivered which will go wrong during the warranty period.

Next comes a period of steady but low rate of failure which is usually quantified by MTBF - mean time between failures. Quite often this phase overlaps with the infant mortality phase. During this period also are failures due to design weaknesses or software issues. For example if a robot or other instrument changes calibration the 'fix' is easy but it is nevertheless a failure. The MTBF of an ST robot is approximately 10,000 hours.

Finally things begin to wear out (senility). Failure rates begin to gradually increase. The system can continue to be used but will require frequent maintenance and replacement parts. ST robots are expected to work for about 30,000 hours continuous light duty use.

I am reminded of a (supposedly true) story about someone who took a new Rolls Royce across Europe. In the middle of nowhere the half-shaft broke and he called the factory. They sent an engineer with a replacement by helicopter. When he got back to UK he called the factory to enquire if there was any charge for the exceptional service and was told: 'Sir, Rolls Royce half shafts do not break'.