A "NEW" PHILOSOPHY OF STRUCTURAL RELIABILITY, FAIL SAFE VERSUS SAFE LIFE

As an indication of the advancing state of the art in helicopter structural technology, the draft RFP for the U.S. Army's Light Experimental Helicopter, (LHX), requires an unprecedented reliability against catastrophic failure of 0.999999 at a life of 10,000 hours.* This requirement is equivalent to an expected failure of less than one in the life of the fleet. Attempting to meet this reliability requirement with a safe-life design approach will impose major penalties in weight and cost relative to current stress-analysis procedures. Also, some possibility for catastrophic failure remains. The weight penalty associated with the safe-life design approach comes about because of the need to use a more conservative S-N curve. That is, in contrast with current stress-analysis procedures which usually use a working S-N curve 3 standard deviations (commonly defined as 3 sigma) below the mean S-N curve, the safe-life approach to achieving a reliability of 0.999999 would require a working S-N curve of approximately 5 sigma below the mean S-N curve. On the other hand, a fail-safe, redundant design approach with four structural elements is likely to be much lighter because it can use a working S-N curve only approximately 1 sigma below the mean S-N curve. However, not all portions of all components can be designed to be fail safe. Accordingly, a design approach is described which combines fail-safe features with enhanced safe-life features and gives promise of minimizing or eliminating both weight and cost penalties and of reducing the probability of catastrophic failure. It is recommended that such a design approach be considered for all future helicopter developments. It is further recommended that government agencies sponsor design studies of multiply-redundant helicopter components to explore the feasability of achieving this fail-safe characteristic. The advantages of the fail-safe approach over the safe-life approach exist over a wide range of coefficients of variation and at least up to a confidence level of 80%.