Mining multi-attribute sequential pattern in onboard failure logging

Onboard Failure Logging (OBFL) is an advanced feature of hardware system. It records failure-related data which is useful for failure analysis process and system reliability improvement. OBFL records are event sequences with multi-attributes. There are lots of algorithms proposed for sequential pattern mining, whereas not much effort has been made to use attribute held by events. However such attributes are critical for failure pattern detecting in failure analysis process. In this paper, we point out the problem of mining multi-attribute sequential pattern in OBFL dataset and propose a new algorithm, called MA-PrefixSpan, to solve it. Finally, we design the OBFL Analysis System to generate the real world OBFL datasets and apply MA-PrefixSpan to mine the failure pattern. The results show that the algorithm can effectively locate the multi-attribute failure patterns which are correlated with failure trends.