Computationally efficient tiered inference for multiple fault diagnosis

A computer-implemented method for tiered inference multiple fault diagnosis, comprising:

• A. using a computer processor, partitioning a hypothesis space into tiers, wherein a hypothesis identifies potential component faults in a production system;
• B. performing diagnosis on the lowest available tier until the lowest available tier is completely ruled out by the observation data;
• C. moving onto the next incremental tier;
• D. based on the current tier of diagnosis, partitioning the production system into a set of subsystems likely to have single faults;
• E. if the tier is ruled out, repeat actions C and D, otherwise move to a step F;
• F. applying a single-fault diagnosis algorithm to each subsystem in the new set of subsystems to identify any probable faults within the subsystem;
• and if the single-fault algorithm determines that each subsystem contains a probable single fault, then perform the following:
• F(1). identify any hypotheses within the determined set of hypotheses that identify the same subsystem faults as the single-fault diagnosis algorithm;
• and
• F(2). store the identified hypotheses into computer memory;
• and
• G. if the observation data suggest that a subsystem has more than one fault, store the invalidated single-fault diagnostic data in memory and remove any hypothesis assuming the single-fault subsystem from the hypotheses space.

A computer-implemented method for partitioning a system into subsystems, comprising:

• providing to a computer memory:
• a set of hypotheses corresponding to a set of components in a production system;
• a probability for each hypothesis that the hypothesis contains a fault;
• and
• a number j of expected faults;
• or each component in the system, calculating a correlation coefficient between the each component and every other component in the system;
• using a computer processor, selecting the j components having the highest autocorrelation values as seed values for a set of j subsystems;
• assigning each of the remaining modules to one of the subsystems based on the calculated correlation coefficients;
• and
• storing the subsystems to computer memory.

A computer-based system for tiered inference multiple fault diagnosis, comprising:

• (i) an observation module configured to:
• receive observation data from a production system;
• and
• update probabilities for a hypothesis space based on the received observation data, where a hypothesis identifies potential component faults in the production system;
• (ii) a dynamic partitioning module configured to:
• partition the hypothesis space into tiers;
• and
• partition the production systems into a set of subsystems where each subsystem is likely to contain at most a single fault;
• (iii) a single fault diagnosis module configured to:
• apply a single-fault diagnosis algorithm to each subsystem in the set of subsystems to identify any probable faults within the subsystem;
• and
• if the single-fault algorithm determines that each subsystem contains a probable single fault, then perform the following:
• identify any hypotheses within the determined set of hypotheses that identify the same subsystem faults as the single-fault diagnosis algorithm;
• and
• store the identified hypotheses into computer memory.