Method for detection and confirmation of multiple failure modes with numerous survivor data

Modern aerospace systems are often subject to multiple modes of failure. Unless the failure modes have radically different statistical characteristics, it can be very difficult to recognize in failure data that multiple failure modes are in play. When using graphical inferential methods for reliability analysis, the potential for multiple failure modes may be suggested by a bent regression line. Subsequent analysis of these modes is usually performed by segregating the data to separate sets to be processed separately for each failure mode, usually segregated about the bend in the regression line. This data segregation process raises a number of questions, especially when numerous survivor data are to be processed. The primary issue with a survivor datum in this case is that it reflects that the system has not yet failed, in any mode. Assigning a survivor datum to any particular failure mode data set for separate processing appears completely arbitrary, and may produce spurious results that may then result in unnecessary costs or unacceptable risks in preventative maintenance.