Failure prediction using gradient-based sensor identification
Methods and systems for predicting failure in a cyber-physical system include determining a prediction index based on a comparison of input time series, from respective sensors in a cyber-physical system, to failure precursors. A failure precursor is detected in the input time series, responsive to a comparison of the prediction index to a threshold. A subset of the sensors associated with the failure precursor is determined, based on a gradient of the prediction index. A corrective action is performed responsive to the determined subset of sensors.
1. A method for predicting failure in a cyber-physical system, comprising:
determining a prediction index as a Euclidean distance between time series vectors based on a comparison of a plurality of input time series, from respective sensors in a cyber-physical system, to each of a plurality of failure precursors;
detecting a failure precursor in the plurality of input time series, responsive to a comparison of the prediction index to a threshold;
determining a subset of the sensors associated with the failure precursor based on a gradient of the prediction index with respect to the plurality of input time series; and
performing a corrective action responsive to the determined subset of sensors.
2. The method of claim 1 , further comprising converting the plurality of input time series to an input vector representation.
3. The method of claim 2 , wherein determining the prediction index comprises comparing the input vector representation to respective vector representations of each of the plurality of failure precursors.
4. The method of claim 3 , wherein the prediction index is a degree of similarity between the input vector representation and a vector representation of at least one of the plurality of failure precursors.
5. The method of claim 2 , wherein determining a subset of the sensors associated with the failure precursor comprises determining a score for each sensor.
6. The method of claim 5 , wherein determining the score for each sensor comprises calculating a gradient for distances of the prediction index using a representative vector of a past failure.
7. The method of claim 6 , wherein the representative vector of the past failure has a smallest distance from the input vector representation.
8. The method of claim 6 , wherein the score for each sensor is determined using back propagation in a neural network.
9. The method of claim 5 , wherein a score for a sensor that has a constant value output is set to zero, or a negative value.
10. The method of claim 1 , wherein the corrective action comprises changing at least one operational parameter in the cyber-physical system.
11. A system for predicting failure in a cyber-physical system, comprising:
a neural network configured to determine a prediction index as a Euclidean distance between time series vectors based on a comparison of a plurality of input time series, from respective sensors in a cyber-physical system, to each of a plurality of failure precursors, and to detect a failure precursor in the plurality of input time series, responsive to a comparison of the prediction index to a threshold;
a failure preventer configured to determine a subset of the sensors associated with the failure precursor based on a gradient of the prediction index with respect to the plurality of input time series, and to perform a corrective action responsive to the determined subset of sensors.
12. The system of claim 11 , wherein the neural network is further configured to convert the plurality of input time series to an input vector representation.
13. The system of claim 12 , wherein the neural network is further configured to compare the input vector representation to respective vector representations of each of the plurality of failure precursors.
14. The system of claim 13 , wherein the prediction index is a degree of similarity between the input vector representation and a vector representation of at least one of the plurality of failure precursors.
15. The system of claim 12 , wherein the failure preventer is further configured to determine a score for each sensor.
16. The system of claim 15 , wherein the failure preventer is further configured to calculate a gradient for distances of the prediction index using a representative vector of a past failure.
17. The system of claim 16 , wherein the representative vector of the past failure has a smallest distance from the input vector representation.
18. The system of claim 16 , wherein the score for each sensor is determined using back propagation in the neural network.
19. The system of claim 15 , wherein a score for a sensor that has a constant value output is set to zero, or a negative value.
20. The system of claim 11 , wherein the corrective action comprises changing at least one operational parameter in the cyber-physical system.