System and method for modeling multilabel classification and ranking
View Patent ↗The present invention provides methods and apparatus for determining and utilizing detection models, such as models for machine condition monitoring. Specifically, the present invention provides a method for identifying and prioritizing labeled data. The model allows a monitored system to be associated with a calibrated and ordered set of states. Further, in machine condition monitoring, the machine condition is associated with the entire set of states in a particular order with a relevance zero-point. That is, a ranked set of calibrated data describing machine conditions is augmented with an annotation indicating a cut-off between relevant and non-relevant data.
1. A method for creating a model comprising:
receiving a set of labels;
ranking the set of labels;
partitioning, by a processor, the ranked set of labels into a first subset of labels and a second subset of labels; and,
inserting, by a processor, a zero-point label between the first subset of labels and the second subset of labels, wherein the zero-point label represents a split point between the first subset of labels and the second subset of labels.
2. The method of claim 1 wherein ranking the set of labels is accomplished by pairwise classification.
3. The method of claim 1 wherein ranking the set of labels is accomplished by constraint classification.
4. The method of claim 1 wherein the first subset of labels are relevant and the second subset of labels are non-relevant.
5. The method of claim 1 wherein the first subset of labels is an empty set.
6. The method of claim 1 wherein the second subset of labels is an empty set.
7. The method of claim 1 wherein the set of labels is a set of annotated sensor data.
8. An apparatus for creating a model comprising:
means for receiving a set of labels;
means for ranking the set of labels;
means for partitioning the ranked set of labels into a first subset of labels and a second subset of labels; and,
means for inserting a zero-point label between the first subset of labels and the second subset of labels, wherein the zero-point label represents a split point between the first subset of labels and the second subset of labels.
9. The apparatus of claim 8 wherein the means for ranking the set of labels comprises means for pairwise classification of the set of labels.
10. The apparatus of claim 8 wherein the means for ranking the set of labels comprises means for constraint classification of the set of labels.
11. The apparatus of claim 8 wherein the first subset of labels are relevant and the second subset of labels are non-relevant.
12. The apparatus of claim 8 wherein the first subset of labels is an empty set.
13. The apparatus of claim 8 wherein the second subset of labels is an empty set.
14. The apparatus of claim 8 wherein the set of labels is a set of annotated sensor data.
15. A computer program product for creating a model comprising:
a non-transitory medium readable by a computer, the computer readable medium having computer program code adapted to:
receive a set of labels;
rank the set of labels;
receive partition information for the ranked set of labels, wherein the partition information partitions the ranked set of labels into a first subset of labels and a second subset of labels;
receive zero-point information for the ranked set of labels, wherein the zero-point information inserts a zero-point label between the first subset of labels and the second subset of labels, wherein the zero-point label represents a split point between the first subset of labels and the second subset of labels; and,
produce a model based on the received partition information and the received zero-point information.
16. The computer program product of claim 15 wherein the computer program code is further adapted to:
partition the ranked set of labels into a first subset of labels and a second subset of labels; and,
insert a zero-point between the first subset of labels and the second subset of labels.
17. The computer program product of claim 15 wherein the computer program code is further adapted to rank the set of labels by pairwise classification.
18. The computer program product of claim 15 wherein the computer program code is further adapted to rank the set of labels by constraint classification.
19. The computer program product of claim 15 wherein the first subset of labels are relevant and the second subset of labels are non-relevant.
20. A method of machine condition monitoring comprising:
monitoring a plurality of sensors adapted to detect machine condition information;
evaluating the machine condition information by employing a calibrated label ranking model predicting a zero-point label between a first subset of labels and a second subset of labels, wherein the zero-point label represents a split point between the first subset of labels and the second subset of labels; and,
outputting an evaluated machine condition from the calibrated label ranking model based on the machine condition information.
21. The method of claim 20 wherein the first subset of labels are relevant and the second subset of labels are non-relevant.
22. The method of claim 20 wherein the first subset of labels is an empty set.
23. The method of claim 20 wherein the second subset of labels is an empty set.
24. The method of claim 20 wherein employing a calibrated label ranking model comprises:
determining a fault condition for each of the plurality of monitored sensors;
applying the fault conditions for each of the plurality of monitored sensors to the calibrated ranking model predicting a zero-point between a first subset of labels and a second subset of labels; and,
determining if the first subset of the labels is consistent with a first subset of the fault conditions.
25. The method of claim 20 further comprising:
generating the calibrated label ranking model based on a set of training data prior to monitoring the plurality of sensors.
26. A system for machine condition monitoring comprising:
a plurality of machine sensors adapted to detect machine condition information; and,
an evaluator adapted to:
receive machine condition information from the plurality of machine sensors;
employ a calibrated label ranking model predicting a zero-point label between a first subset of labels and a second subset of labels, wherein the zero-point label represents a split point between the first subset of labels and the second subset of labels; and,
output an evaluated machine condition from the calibrated label ranking model based on the machine condition information.
27. The system of claim 26 wherein the first subset of labels are relevant and the second subset of labels are non-relevant.
28. The system of claim 26 wherein the first subset of labels is an empty set.
29. The system of claim 26 wherein the second subset of labels is an empty set.
30. The system of claim 26 wherein the evaluator is further adapted to:
determine a fault condition for each of the plurality of monitored sensors;
apply the fault conditions for each of the plurality of monitored sensors to the calibrated ranking model predicting a zero-point between a first subset of labels and a second subset of labels; and,
determine if the first subset of the labels is consistent with a first subset of the fault conditions.
31. The system of claim 26 wherein the evaluator is further adapted to:
apply a set of training data to the calibrated label ranking model.