How the Royal Australian Air Force used RelCode to assist with the determination of a new c130 propeller valve housing maintenance policy
by Peter McLennan
Wing Commander
Royal Australian Air Force
The RAAF has used the RELCODE software to evaluate the reliability characteristics of Lockheed C130 propeller valve housings. These housings are an extremely complex mechanism of gears, springs, cams and dashpots. The motivation for this analysis was a combination of low reliability, expensive maintenance, and shortages of serviceable spare items (all of which are, of course, related). Unfortunately, this work was conducted during a period of considerable organisational and procedural change within the RAAF, which has made it difficult to reconstruct the exact details of the analysis. The following information is based on the best available data and the recollections of several of the people involved.
Initial statistical analysis in the early 1990s suggested that the overall failure pattern of these items was essentially random, and had not been greatly affected by earlier variations in the overhaul interval of the housing. Accordingly, an 'age exploration' program was established, which involved progressively increasing the overhaul interval until signs of increasing unreliability became evident. The overhaul interval was increased from 5,000 hours in use to 5,500 hours at that time, and the intention was to subsequently review the results of this change to ascertain whether a further increase may be desirable.
A more rigorous RELCODE analysis was conducted in the mid-90s, when the proposal to extend the interval by another 500 hours was considered. The overall failure deterioration pattern now seemed to exhibit signs of wear out, with Weibull shape parameter (ß) values over 2. However, the Weibull curve did not accurately fit the data, particularly at low values of time since overhaul, which pointed to the possibility of a multi-modal failure pattern. Accordingly, the failure data were subdivided into separate failure modes to the extent possible, and these data sets were then analysed separately. This revealed the existence of failure patterns with significantly different characteristics to the overall pattern, including an electrical failure mode with a pronounced infant mortality ('burn in') failure pattern (ß = 0.46).
An inspection of maintenance records indicated that the majority of components were not surviving to the overhaul life of 5,500 hours, and those that did survive that long were usually found to be in good condition. This further confirmed the suspicion that strong infant mortality was present, which was not being taken into account in the proposal to increase the overhaul interval. Such failures could not be prevented by existing maintenance procedures, and so should not have been considered when determining the maintenance interval (in fact, revised procedures are required to deal with infant mortality, which is usually associated with quality problems in maintenance or parts). In addition, the inclusion of this data in the overall failure pattern was also potentially masking stronger wear-out characteristics of other failure modes; accordingly, the proposal to further extend the maintenance interval was not approved.
This episode reveals several lessons about the correct means of determining maintenance requirements. Conclusions based only on MTBF would fail to detect the effects described above. Moreover, just because the failure probability density function appears to be randomly (or evenly) distributed does not mean that the failure pattern is random (constant hazard rate), with no wear-out occurring. Even a superficial Weibull analysis can be misleading if multiple failure modes are present. However, a proper analysis, facilitated by RELCODE, is able to reveal the existence of failure modes of different kinds, leading to appropriate means to deal with what is actually happening.
Without the analysis described above, the RAAF would not have become aware of the quality problems resulting in infant mortality, and would probably have continued to lengthen the interval between overhauls with consequent further reductions in reliability, availability and economy.