Monitoring system for an engine test bench

1. A self-adaptive monitoring system adapted for monitoring a test bench and at least one aircraft engine component installed on the test bench and mechanically coupled to the test bench via a shaft, comprising:

sensors provided on the test bench and provided on the engine component, the sensors configured to acquire measurements recorded at high frequency and measurements recorded at low frequency; and

a processor configured to

acquire time signal packets corresponding to measurements of endogenic and exogenic parameters specific to a combination of the test bench and the at least one aircraft engine component including a shaft line corresponding to a mechanical coupling between the test bench and the at least one aircraft engine component, in which the at least one aircraft engine component is driven in rotation by the shaft that is turned by the bench, at successive instants;

construct an endogenic indicator vector and an associated exogenic indicator vector at each instant of the successive instants, using time signal packets earlier than the instant;

identify a context class for the exogenic indicator vector;

calculate a risk probability of the endogenic indicator vector conditioned by the identified context class for the associated exogenic indicator vector using at least one anomaly detector, to produce a diagnostic of a state of the combination of the test bench and the at least one aircraft engine component; and

issue a warning or tripping the monitoring system if an anomaly is confirmed.

2. A system according to claim 1 , wherein the processor is further configured to:

construct a set of context classes starting from a sequence of initial exogenic indicator vectors during a learning phase; and

update the set of context classes during an execution phase, starting from new exogenic indicator vector inputs.

3. A system according to claim 2 , wherein the processor is further configured to update the set of context classes by checking if a new detection of an exogenic indicator vector belongs to a previously constructed context class, and recording the exogenic indicator vector of the new detection in a database if there is no context class that corresponds to the new detection until an appropriate number of similar exogenic indicator vectors have been detected to form a new context class.

4. A system according to claim 2 , wherein the processor is further configured to update the set of context classes, by verifying if new detections of exogenic indicator vectors belong to previously constructed context classes and recording at least some of the new detections in the corresponding context classes.

5. A system according to claim 2 , wherein the processor is further configured to identify the context class of an exogenic indicator vector by calculating a match value of the exogenic indicator vector relative to each context class.

6. A system according to claim 2 , wherein the processor is further configured to construct the set of context classes using a likelihood maximization criterion.

7. A system according to claim 2 , wherein the processor is further configured to select an appropriate number of context classes based on an optimization criterion applied to the exogenic indicator vectors.

8. A system according to claim 1 , further comprising:

buffer memories to buffer at least one packet of time signals earlier than the instant, for each endogenic or exogenic parameter; and

the processor is further configured to

smooth each of the time signal packets according to at least one scale to form curves representative of the packets;

re-sample the representative curves; and

compress the re-sampled curves to construct the endogenic or exogenic indicator vector.

9. A system according to claim 1 , wherein the processor is further configured to calculate a quality value of the risk probability.

10. A system according to claim 1 , wherein the anomaly detector implements a normal behavior model and generates a normality measurement by a likelihood calculation.

11. A system according to claim 1 , wherein the anomaly detector implements a bearing damage detection model.

12. A system according to claim 1 , wherein the anomaly detector implements an intermittent events detection model.

13. A system according to claim 1 , further comprising supervision means in which each anomaly detector is encapsulated, the supervision means being configured to start each anomaly detector, to procure input data for each anomaly detector, to receive output messages from each anomaly detector, and to manage instances corresponding to parameter settings and calibration choices for each anomaly detector.

14. A management system comprising:

a control system connected to an engine test bench, the control system configured to control the test bench and to record data output from the test bench and at least one component of an aircraft engine in storage means; and

a monitoring system according to claim 1 , the monitoring system being connected to the test bench through the control system that sends the data to the monitoring system output from the test bench and engine component combination.

15. A method of self-adaptively monitoring a test bench and at least one aircraft engine component installed on the test bench and mechanically coupled to the test bench via a shaft, comprising:

providing sensors on the test bench and on the engine component, the sensors being configured to acquire measurements recorded at high frequency and measurements recorded at low frequency;

acquiring time signal packets at successive instants corresponding to endogenic and exogenic parameter measurements specific to a combination of the test bench and the at least one aircraft engine component including a shaft line corresponding to a mechanical coupling between the test bench and the at least one aircraft engine component, in which the at least one aircraft engine component is driven in rotation by the shaft that is turned by the bench;

at each instant of the successive instants, using packets of time signals earlier than the instant to construct an endogenic indicator vector and an associated exogenic indicator vector;

identifying a context class for the exogenic indicator vector;

using at least one anomaly detector to calculate a risk probability of the endogenic indicator vector conditioned by the identified context class for the associated exogenic indicator vector to produce a diagnostic of a state of the combination of the test bench and the at least one aircraft engine component; and

issuing a warning or tripping if an anomaly is confirmed.

16. A non-transitory computer readable medium comprising executable code instructions for implementation of the method according to claim 15 when executed by a computer.