Model based system monitoring
According to an aspect, a method includes configuring a predictive system model with a plurality of component models to correspond with a modeled state of a mechanical system. A set of perturbation inputs is applied to one or more of the component models to produce a plurality of prediction results. The prediction results are compared to a targeted mode of the mechanical system. A set of configuration parameters of the mechanical system is captured in combination with one or more of the perturbation inputs that most closely results in the prediction results matching the targeted mode of the mechanical system.
1. A method comprising:
configuring, in a processing system of a mechanical system, a predictive system model with a plurality of component models to correspond with a modeled state of a mechanical system, wherein the predictive system model is a reduced order model derived from one or more higher fidelity models of the mechanical system by removing higher order terms above second-order effects;
applying a set of perturbation inputs to one or more of the component models to produce a plurality of prediction results;
comparing the prediction results to a targeted mode of the mechanical system, the targeted mode defining a target/goal state to be achieved;
capturing, in a memory system of the processing system, a set of configuration parameters of the mechanical system in combination with one or more of the perturbation inputs that most closely results in the prediction results matching the targeted mode of the mechanical system, wherein associations between fault conditions of the mechanical system and the set of configuration parameters and perturbation inputs are captured in failure mode definitions to reproduce one or more faults; and
using the failure mode definitions to predict and detect one or more failure modes of the mechanical system in real-time through trends of sensor data relative to an observed history comprising prediction and detection of one or more parameters changing at a faster rate than expected as defined in the failure mode definitions.
2. The method of claim 1 , wherein the modeled state of the mechanical system is a degraded condition of the mechanical system.
3. The method of claim 1 , wherein the targeted mode corresponds to an observed condition of the mechanical system from sensor data.
4. The method of claim 1 , wherein the predictive system model defines mechanical and fluid interactions between the component models.
5. The method of claim 1 , further comprising:
performing fault trend detection to observe one or more progressive sequences ofthe configuration parameters and perturbation inputs that result in each of the one or more faults.
6. The method of claim 1 , further comprising:
using a neural network to interpolate conditions between the failure mode definitions.
7. The method of claim 1 , further comprising:
capturing a history of the prediction results; and
comparing observed faults of the mechanical system to the prediction results to train the predictive system model.
8. A processing system operably coupled to a sensor system of a mechanical system, the processing system comprising:
a memory system; and
one or more processors configured to execute a plurality of instructions comprising:
a predictive system model comprising a plurality of component models configured to correspond with a modeled state of the mechanical system, wherein the predictive system model is a reduced order model derived from one or more higher fidelity models of the mechanical system by removing higher order terms above second-order effects, and the predictive system model uses the failure mode definitions to predict and detect one or more failure modes of the mechanical system in real-time through trends of sensor data relative to an observed history comprising prediction and detection of one or more parameters changing at a faster rate than expected as defined in the failure mode definitions; and
a model exerciser operable to apply a set of perturbation inputs to one or more of the component models to produce a plurality of prediction results, compare the prediction results to a targeted mode of the mechanical system, and capture in the memory system a set of configuration parameters of the mechanical system in combination with one or more ofthe perturbation inputs that most closely results in the prediction results matching the targeted mode of the mechanical system, wherein the targeted mode defines a target/goal state to be achieved, and wherein associations between fault conditions of the mechanical system and the set of configuration parameters and perturbation inputs are captured in the failure mode definitions to reproduce one or more faults.
9. The processing system of claim 8 , wherein the modeled state of the mechanical system is a degraded condition of the mechanical system.
10. The processing system of claim 8 , wherein the targeted mode corresponds to an observed condition of the mechanical system from sensor data.
11. The processing system of claim 8 , wherein the predictive system model defines mechanical and fluid interactions between the component models.
12. The processing system of claim 8 , further comprising fault trend detection to observe one or more progressive sequences of the configuration parameters and perturbation inputs that result in each of the one or more faults.
13. The processing system of claim 8 , wherein a neural network is used to interpolate conditions between the failure mode definitions.
14. The processing system of claim 8 , wherein the processing system is further operable to capture a history of the prediction results in the memory system and compare observed faults of the mechanical system to the prediction results to train the predictive system model.