Methods and systems for airfoil prognostics using physics-informed machine learning
Methods and systems for airfoil prognostics using physics-informed machine learning introduce perturbations to a nominal airfoil geometry, simulating potential faults. The perturbed airfoil geometries are processed by physics-informed neural network (PINN) models, trained on geometric and physics-based features, to predict lift coefficients, thereby predicting the health status of the airfoil. A fault probing strategy is applied to assess health status under mission-specific requirements, predicting a remaining useful life and confidence bounds for each mission-specific requirement. Efficient airfoil health prediction, including precise fault simulation, allow for proactive maintenance planning based on the predicted health status of the airfoil.
1 . A computing system, comprising:
a processor; and
a memory storing executable instructions that, when executed by the processor, cause the processor to execute:
an impending fault simulator configured to:
receive input of a nominal airfoil geometry of an airfoil,
simulate impending faults in the airfoil by perturbing the nominal airfoil geometry to generate perturbed airfoil geometries, and
input the perturbed airfoil geometries into one or more physics-informed neural network (PINN) models to generate respective predicted airfoil lift coefficients for the perturbed airfoil geometries; and
a health status evaluator configured to generate a predicted future health status for the airfoil under one or more operational conditions based on the predicted airfoil lift coefficients for the perturbed airfoil geometries, thereby leveraging physics-informed machine learning for efficiently generating airfoil prognostics, wherein
the health status evaluator further includes an extreme operation condition identifier configured to use a fault probing strategy to search for plausible fault probing conditions, and use the plausible fault probing conditions to simulate impending faults and generate the predicted future health status for the airfoil under the one or more operational conditions;
the health status evaluator is configured to generate the predicted future health status for the airfoil for each of a plurality of mission-specific requirements; and
the extreme operation condition identifier is configured to generate the predicted future health status for the airfoil for each of the plurality of mission-specific requirements.
2 . The computing system of claim 1 , wherein the processor further executes a model trainer configured to train the one or more PINN models to learn lift coefficients of the airfoil using training input data comprising geometric features and physics-based features of the airfoil.
3 . The computing system of claim 2 , wherein the physics-based features include at least one selected from the group of angles of attack, Reynolds numbers, simulated lift coefficients, and simulated pressure drag coefficients.
4 . The computing system of claim 2 , wherein the model trainer randomly initializes weights and biases of the one or more PINN models, so as to apply a unique configuration of weights and biases to each of the one or more PINN models.
5 . The computing system of claim 4 , wherein the model trainer uses the training input data and the randomly initialized weights and biases to train the one or more PINN models via a loss function measuring an error between the predicted airfoil lift coefficients and true lift coefficients, the loss function being minimized over iterative updates of the weights and biases.
6 . The computing system of claim 5 , wherein the true lift coefficients are provided from at least one selected from the group of physics experiments, flight tests, wind tunnel experiments, and high-fidelity simulators.
7 . The computing system of claim 1 , wherein the impending fault simulator is configured to generate the perturbed airfoil geometries by using at least one mechanism selected from the group of a random surface deformation mechanism, a leading-edge surface erosion mechanism, a leading-edge surface corrosion mechanism, a trailing-edge surface erosion mechanism, a trailing-edge surface corrosion mechanism, and a combination thereof.
8 . The computing system of claim 1 , wherein the impending fault simulator is configured to quantify uncertainty regions of the predicted airfoil lift coefficients under different operational dimensions.
9 . The computing system of claim 1 , wherein the predicted future health status for the airfoil includes a remaining useful life for each of the plurality of mission-specific requirements, the predicted future health status for each of the plurality of mission-specific requirements having confidence bounds.
10 . The computing system of claim 9 , wherein the predicted future health status for the airfoil further includes one of a plurality of health status categories.
11 . A method, comprising:
receiving input of a nominal airfoil geometry of an airfoil,
simulating impending faults in the airfoil by perturbing the nominal airfoil geometry to generate perturbed airfoil geometries, and
inputting the perturbed airfoil geometries into one or more physics-informed neural network (PINN) models to generate respective predicted airfoil lift coefficients for the perturbed airfoil geometries; and
generating a predicted future health status for the airfoil under one or more operational conditions for each of a plurality of mission-specific requirements based on the predicted airfoil lift coefficients for the perturbed airfoil geometries, thereby leveraging the one or more PINN models to efficiently generate the predicted future health status for the airfoil, wherein
the predicted future health status for the airfoil under the one or more operational conditions is generated by using a fault probing strategy to search for plausible fault probing conditions for each of the plurality of mission-specific requirements, and use the plausible fault probing conditions to simulate impending faults.
12 . The method of claim 11 , further comprising training the one or more PINN models to learn lift coefficients of the airfoil using training input data comprising geometric features and physics-based features of the airfoil.
13 . The method of claim 12 , wherein the physics-based features include at least one selected from the group of angles of attack, Reynolds numbers, simulated lift coefficients, and simulated pressure drag coefficients.
14 . The method of claim 12 , wherein the training includes randomly initializing weights and biases of the one or more PINN models, so as to apply a unique configuration of weights and biases to each of the one or more PINN models.
15 . The method of claim 14 , wherein the training includes using the training input data and the randomly initialized weights and biases to train the one or more PINN models via a loss function measuring an error between the predicted airfoil lift coefficients and true lift coefficients, the loss function being minimized over iterative updates of the weights and biases.
16 . The method of claim 15 , wherein the true lift coefficients are provided from at least one selected from the group of physics experiments, flight tests, wind tunnel experiments, and high-fidelity simulators.
17 . The method of claim 11 , wherein the perturbed airfoil geometries are generated by using at least one mechanism selected from the group of a random surface deformation mechanism, a leading-edge surface erosion mechanism, a leading-edge surface corrosion mechanism, a trailing-edge surface erosion mechanism, a trailing-edge surface corrosion mechanism, and a combination thereof.
18 . The method of claim 11 , wherein uncertainty regions of the predicted airfoil lift coefficients are quantified under different operational dimensions.
19 . The method of claim 11 , wherein the predicted future health status for the airfoil includes a remaining useful life for each of the plurality of mission-specific requirements, the predicted future health status for each of the plurality of mission- specific requirements having confidence bounds.
20 . The method of claim 19 , wherein the predicted future health status for the airfoil further includes one of a plurality of health status categories.