Methods and systems for fault diagnosis
Methods may comprise: identifying a fault indicator associated with a physical system; collecting first data related to a state of the physical system; applying a surrogate model to the first data to produce a plurality of potential fault modes; applying an optimization algorithm to the plurality of potential fault modes using a similarity metric to produce an input and a plurality of outputs, wherein each of the plurality of outputs corresponds to one of the plurality of potential fault modes, wherein the input provides differentiation between each of the plurality of outputs; applying the input to the physical system; collecting second data from physical system in response to applying the input; identifying a true mode of the physical system based on a comparison of the second data and the plurality of outputs; and diagnosing a fault of the physical system based on the true mode.
1. A method for diagnosing a fault in a physical system, the method comprising:
identifying a fault indicator associated with the physical system;
collecting first data related to a state of the physical system;
applying a surrogate model to the first data to produce a plurality of potential fault modes, wherein the surrogate model is a recurrent neural network (RNN) trained by:
inputting a training data set, the training data set generated from an augmented system model and a first set of one or more separate training modes generated by the RNN;
comparing, to the training data set, a second set of one or more separate training modes, the second set of one or more separate training modes generated by the RNN based on the training data set; and
based on the comparing, adjusting one or more weights of the RNN;
applying an optimization algorithm to the plurality of potential fault modes using a similarity metric to produce an input and a plurality of outputs, wherein each of the plurality of outputs corresponds to one of the plurality of potential fault modes, wherein the input provides differentiation between each of the plurality of outputs; applying the input to the physical system;
collecting second data from physical system in response to applying the input;
identifying a true mode of the physical system based on a comparison of the second data and the plurality of outputs; and
diagnosing the fault of the physical system based on the true mode.
2. The method of claim 1 , wherein the method is performed in real-time.
3. The method of claim 1 further comprising:
changing an operating parameter of the system in response to the fault.
4. The method of claim 1 further comprising:
applying a remedial action to the system in response to the fault.
5. The method of claim 1 further comprising:
displaying the fault.
6. The method of claim 1 , wherein the optimization algorithm is a gradient-free optimization algorithm.
7. The method of claim 1 , wherein the surrogate model is a neural network trained based on third data produced by the augmented system model that is a physics-based model capable of modeling nominal modes and faulty modes of the physical system.
8. The method of claim 1 , wherein the physical system comprises a mechanical system, an electrical system, and/or a thermal system.
9. The method of claim 1 , wherein the similarity metric comprises a cosine similarity metric and/or an L2 similarity metric.
10. A computing system for diagnosing a fault in a physical system, the computing system comprising:
a processor;
a memory coupled to the processor; and
instructions provided to the memory, wherein the instructions are executable by the processor to cause the system to perform the method of claim 1 .
11. A method for diagnosing a fault in a system, the method comprising:
identifying a fault indicator associated with the physical system;
collecting first data related to a state of the physical system;
applying a surrogate model to the first data to produce a plurality of potential fault modes, wherein the surrogate model is a recurrent neural network (RNN) trained by:
inputting a training data set, the training data set generated from an augmented system model and a first set of one or more separate training modes generated by the RNN;
comparing, to the training data set, a second set of one or more separate training modes, the second set of one or more separate training modes generated by the RNN based on the training data set; and
based on the comparing, adjusting one or more weights of the RNN;
applying an optimization algorithm to the plurality of potential fault modes using a similarity metric to produce a plurality of inputs and a plurality of outputs for each of the plurality of inputs, wherein each of the plurality of outputs for each of the plurality of inputs corresponds to one of the plurality of potential fault modes, wherein at least two of the plurality of inputs produce a different output for one of the plurality of potential fault modes;
applying the plurality of inputs to the physical system;
collecting second data from physical system in response to applying the plurality of inputs;
identifying a true mode of the physical system based on a comparison of the second data and the plurality of outputs for each of the plurality of inputs; and diagnosing the fault of the physical system based on the true mode.
12. The method of claim 11 further comprising:
changing an operating parameter of the system in response to the fault.
13. The method of claim 11 further comprising:
applying a remedial action to the system in response to the fault.
14. The method of claim 11 further comprising:
displaying the fault.
15. The method of claim 11 , wherein the optimization algorithm is a gradient-free optimization algorithm.
16. The method of claim 11 , wherein the surrogate model is a neural network trained based on third data produced by the augmented system model that is a physics-based model capable of modeling nominal modes and faulty modes of the physical system.
17. A system comprising:
a processor;
a memory coupled to the processor; and
instructions provided to the memory, wherein the instructions are executable by the processor to cause the system to perform the method of claim 11 .
18. A method for generating a surrogate model, the method comprising:
applying a fault augmentation to a physics-based model of a physical system using physics-based fault modes to yield an augmented system model;
generating training data by applying a plurality of inputs to the augmented system model, the inputs including a first set of one or more separate training modes generated by a recurrent neural network (RNN); and
training the surrogate model comprising differential equations with the training data, wherein the surrogate model is the RNN trained by:
inputting the training data;
comparing, to the training data, a second set of one or more separate training modes, the second set of one or more separate training modes generated by the RNN based on the training data; and
based on the comparing, adjusting one or more weights of the RNN.