A system and method for fault diagnosis includes receiving information defining a relationship between failure modes and diagnostic trouble codes and extracting diagnostic trouble code data, including set times, frequency data and diagnostic trouble code sequence information for a plurality of diagnostic trouble codes relating to a plurality of failure modes. The system and method further include constructing a Markov chain using the diagnostic trouble code data for each of the plurality of failure modes, training the Markov chain to learn a set of state parameters using the diagnostic trouble code data, and computing a likelihood of a diagnostic trouble code sequence for each of the plurality of failure modes using the trained Markov chains.