초록 ▼

A real-time on board engine component performance tracking system for deriving high fidelity engine models capable of providing a reference level from which performance changes can be trended with reduced operating cost, increased safety, and increased reliability while minimizing memory requirement

A real-time on board engine component performance tracking system for deriving high fidelity engine models capable of providing a reference level from which performance changes can be trended with reduced operating cost, increased safety, and increased reliability while minimizing memory requirements and maximizing speed.

대표청구항 ▼

What is claimed is: 1. A method of training a multi-layer network used in conjunction with a physics-based engine model to define a hybrid model for modeling a gas turbine engine used for flight comprising: inputting an input signal to the gas turbine engine for operational control and to the physi

What is claimed is: 1. A method of training a multi-layer network used in conjunction with a physics-based engine model to define a hybrid model for modeling a gas turbine engine used for flight comprising: inputting an input signal to the gas turbine engine for operational control and to the physics-based engine model, wherein the input signal comprises a plurality of parameters; combining output parameters from the gas turbine engine with corresponding parameters estimated by the physics-based engine model to form a residual for each parameter, the gas turbine output parameters and physics-based engine model estimated parameters are responsive to the input signal parameters; deriving a mean and a standard deviation for each parameter of a first predetermined number of the input signals; deriving a mean and a standard deviation for residuals responsive to the first predetermined number of the input signals; associating a mean Mach and a mean altitude, wherein the Mach and altitude are input signal parameters, with one of a plurality of flight envelope cells, wherein if a parameter of the input signal or a responsive residual parameter is included in a pre-existing statistical distribution, the pre-existing distribution is used for training the multi-layer network, and if the mean Mach and the mean altitude are not included in a pre-existing statistical distribution, a new statistical distribution is created comprising the mean and standard deviation for each parameter of the first predetermined number of the input signals, the mean and standard deviation for residuals responsive to the first predetermined number of the input signals, and the first predetermined number value. 2. The method according to claim 1 further comprising calculating distances from the mean Mach and mean altitude to a plurality of flight envelope cell centers. 3. The method according to claim 2 further comprising selecting a cell that is the shortest distance to the mean Mach and mean altitude. 4. The method according to claim 3 further comprising comparing each independent input signal parameter from the input signal with pre-existing statistical distributions within the associated cell, wherein if an independent parameter of the input signal is included in a pre-existing statistical distribution, the pre-existing distribution is used for training the multi-layer network, and if an independent parameter of the input signal is not included in a pre-existing statistical distribution, a new statistical distribution is created for each input signal parameter from the input signal. 5. The method according to claim 4 wherein the statistical distribution is a Gaussian mixture model. 6. The method according to claim 5 further comprising normalizing the mean for each parameter of the first predetermined number of the input signals and the mean for each parameter of the residuals responsive to the first predetermined number of input signals. 7. The method according to claim 6 further comprising calculating a stability measurement that determines whether the input signal parameters exhibit steady state or transient behavior. 8. The method according to claim 7 further comprising examining a change of rate (slope) of at least one parameter of the input signal over a second predetermined number of input signals, wherein if the change of rate is greater than a predetermined threshold, the input signal is exhibiting transient behavior. 9. The method according to claim 8 wherein if the input signal is exhibiting steady state behavior, the first predetermined number is set at a steady state value. 10. The method according to claim 9 wherein if the input signal is exhibiting transient behavior, the first predetermined number is set at a transient value. 11. The method according to claim 10 further comprising stopping creating the statistical distribution when the first predetermined number is greater than a preset limit. 12. The method according to claim 10 further comprising calculating a statistical measure (Mahalanobis distance) correlating the mean for each parameter of the first predetermined number of the input signals with an initial average, and stopping creating the statistical distribution when the statistical measure is greater than a third predetermined number. 13. The method according to claim 11 further comprising storing the statistical distribution in a statistical distribution data store. 14. The method according to claim 13 further comprising training the multi-layer network using the statistical distributions in the statistical distribution data store. 15. A system for training a multi-layer network used in conjunction with a physics-based engine model to define a hybrid model for modeling a gas turbine engine used for flight comprising: means for inputting an input signal to the gas turbine engine for operational control and to the physics-based engine model, wherein the input signal comprises a plurality of parameters; means for combining output parameters from the gas turbine engine with corresponding parameters estimated by the physics-based engine model to form a residual for each parameter, the gas turbine output parameters and physics-based engine model estimated parameters are responsive to the input signal parameters; means for deriving a mean and a standard deviation for each parameter of a first predetermined number of the input signals; means for deriving a mean and a standard deviation for residuals responsive to the first predetermined number of the input signals; means for associating a mean Mach and a mean altitude, wherein the Mach and altitude are input signal parameters, with one of a plurality of flight envelope cells, wherein if a parameter of the input signal or a responsive residual parameter is included in a pre-existing statistical distribution, the pre-existing distribution is used for training the multi-layer network, and if the mean Mach and the mean altitude are not included in a pre-existing statistical distribution, a new statistical distribution is created comprising the mean and standard deviation for each parameter of the first predetermined number of the input signals, the mean and standard deviation for residuals responsive to the first predetermined number of the input signals, and the first predetermined number value. 16. The system according to claim 15 further comprising means for calculating distances from the mean Mach and mean altitude to a plurality of flight envelope cell centers. 17. The system according to claim 16 further comprising means for selecting a cell that is the shortest distance to the mean Mach and mean altitude. 18. The system according to claim 17 further comprising means for comparing each independent input signal parameter from the input signal with pre-existing statistical distributions within the associated cell, wherein if an independent parameter of the input signal is included in a pre-existing statistical distribution, the pre-existing distribution is used for training the multi-layer network, and if an independent parameter of the input signal is not included in a pre-existing statistical distribution, a new statistical distribution is created for each input signal parameter from the input signal. 19. The system according to claim 18 wherein the statistical distribution is a Gaussian mixture model. 20. The system according to claim 19 further comprising means for normalizing the mean for each parameter of the first predetermined number of the input signals and the mean for each parameter of the residuals responsive to the first predetermined number of input signals. 21. The system according to claim 20 further comprising means for calculating a stability measurement that determines whether the input signal parameters exhibit steady state or transient behavior. 22. The system according to claim 21 further comprising means for examining a change of rate (slope) of at least one parameter of the input signal over a second predetermined number of input signals, wherein if the change of rate is greater than a predetermined threshold, the input signal is exhibiting transient behavior. 23. The system according to claim 22 wherein if the input signal is exhibiting steady state behavior, the first predetermined number is set at a steady state value. 24. The system according to claim 23 wherein if the input signal is exhibiting transient behavior, the first predetermined number is set at a transient value. 25. The system according to claim 24 further comprising means for stopping creating the statistical distribution when the first predetermined number is greater than a preset limit. 26. The system according to claim 24 further comprising means for calculating a statistical measure (Mahalanobis distance) correlating the mean for each parameter of the first predetermined number of the input signals with an initial average, and stopping creating the statistical distribution when the statistical measure is greater than a third predetermined number. 27. The system according to claim 25 further comprising means for storing the statistical distribution in a statistical distribution data store. 28. The system according to claim 27 further comprising means for training the multi-layer network using the statistical distributions in the statistical distribution data store.