Estimation of a diagnosis of a physical system by predicting fault indicators by machine learning

1. A method for establishing a diagnosis of a physical system comprising acquiring measurements (m 1 , m 2 , . . . , m s ) provided by sensors, providing said measurements at an input of a set of predictive models each generating a fault indicator, then determining a diagnosis of said physical system from said fault indicators and a signature matrix associating diagnostics and fault indicators, the set of predictive models having been trained using a training set consisting in associating sets of measurements and tags representing a desired diagnosis corresponding to an associated malfunction or to a nominal case, each predictive model of said set of predictive models having been trained, with said training set, in a supervised and iterative manner, to generate a desired fault indicator determined from a tag associated with one of said sets of measurements and said signature matrix.

2. The method of claim 1 , wherein said signature matrix is determined by determining a graph to depict the variables of the physical system and relationships between the variables, determining minimally overdetermined sub-graphs from said graph, then determining said signature matrix from the minimally overdetermined sub-graphs.

3. The method of claim 2 , wherein the determination of the minimally overdetermined sub-graphs is obtained by a Dulmage-Mendelsohn decomposition.

4. The method of claim 3 , wherein several predictive models are trained for a same fault indicator, and comprising selecting one predictive model from among said several predictive models based on defined criteria.

5. A diagnostic system for diagnosing a physical system comprising means for acquiring measurements provided by sensors, for providing said measurements at an input of a set of predictive models each generating a fault indicator, then for determining a diagnosis of said physical system from said fault indicators and a signature matrix associating diagnostics and fault indicators, the set of predictive models having been trained using a training set consisting in associating sets of measurements and tags representing a desired diagnosis corresponding to an associated malfunction or to a nominal case, each predictive model of said set of predictive models having been trained, with said training set, in a supervised and iterative manner, to generate a desired fault indicator determined from a tag associated with one of said sets of measurements and said signature matrix.

6. A system comprising:

a computer configured to cause performance of a diagnostic system according to claim 5 ;

a physical system configured to be diagnosed by said diagnostic system; and

sensors configured to provide measurements of said physical system to said diagnostic system.

7. A non-transitory computer-readable storage medium having a computer program product comprising instructions stored thereon which, when executed by a computer, lead said computer to implement the method according to claim 1 .

8. A non-transitory computer-readable storage medium having a computer program product comprising instructions stored thereon which, when executed by a computer, lead said computer to implement the method according to claim 2 .

9. A non-transitory computer-readable storage medium having a computer program product comprising instructions stored thereon which, when executed by a computer, lead said computer to implement the method according to claim 3 .

10. A non-transitory computer-readable storage medium having a computer program product comprising instructions stored thereon which, when executed by a computer, lead said computer to implement the method according to claim 4 .