Method and system for ascertaining anomalies in electric motors
First Claim
1. A computer-based method for ascertaining anomalies in an electric motor, comprising the steps of:
- (a) obtaining a set of samples of supply current for a motor known to be in a normal condition;
(b) computing and processing a set of Fast Fourier Transforms (FFT'"'"'s) based on waveforms of said supply current for said motor known to be in a normal condition so as to create a feature set of input vectors;
(c) calculating all possible input vector assignments for clusters 1 through n using Ward'"'"'s method, where n is the number of p-dimensional input vectors, in accordance with the following steps;
Step 1;
initializing,Step 2;
setting k, the number of groups considered, equal to n, the number of input vectors,Step 3;
setting the "best value" Z[pk13 1, qk-- 1, k-1] equal to some initial large value and putting i equal to the smallest active identification number,Step 4;
setting j equal to the first active identification number greater than i,Step 5;
computing Z[i,j,k-1] associated with the hypothesized union of sets i and j,Step 6;
determining whether Z[i,j,k-1] is better than the best value Z[pk-- 1, qk-- 1, k-1] up to this comparison;
if the answer yes, this is followed byStep 7, replacing the old value of Z[pk-- 1, qk-- 1, k-1] by Z[i,j,k-1] and making pk-- 1=i and qk-- 1=j and proceeding to Step 8;
if the answer is no, this is followed byStep 8, ascertaining whether j is equal to the last active identification number;
if the answer is no, thenStep 9, set j equal to the next higher active identification number, and return to Step 5;
if the answer is yes, proceed toStep 10, determine whether i is egual to the next higher active identification number;
if not proceed to Step 11;
if yes, proceed to Step 12,Step 11, set i equal to the next higher active identification number (Step
11) and return to Step 4;
Step 12, find a best union of 2 sets and identify it by the identification numbers pk-- 1 and qk-- 1, the value associated with their union being Z[pk-- 1, qk-- 1, k-i], then toStep 13, identify the new union by the number pk-- 1 and make the identification number qk-- 1 inactive, Step 14, calculate trace values for 2 clusters that were merged and for the new cluster which was formed by the merge,Step 15, compute lambda-- r and E-- r for this step in the clustering procedure,Step 16, determine whether k, the number of groups under consideration, is equal to 2;
if not, proceed to Step 17;
if the answer is yes, proceed to Step 18,Step 17, set k equal to k-1 (Step
17) and return to step 2;
Step 18, calculate the approximate weight of evidence for all steps from 1 to n-1;
(d) computing the Approximate Weight of Evidence (AWE) for said cluster counts 1 through n;
(e) selecting that count associated with a maximum AWE and designating this count as s;
(f) for each cluster 1 through s, finding the member vector farthest from the cluster'"'"'s centroid and defining this as the cluster'"'"'s radius;
(g) reading in a single input sample for a motor under supervision;
(h) computing and processing an FFT based on said input sample for newly generating a feature vector; and
(i) checking whether said newly generated feature vector is inside any of said clusters 1 through s as defined for each cluster by said radius; and
(j) if not, outputting a warning signal.
0 Assignments
0 Petitions
Accused Products
Abstract
A computer-based system and method for ascertaining anomalies in an electric motor, includes computing and processing a set of Fast Fourier Transforms (FFT'"'"'s) of the supply current waveforms for a motor known to be in a normal condition to create a feature set of input vectors. The input vector assignments for all possible cluster groupings 1 through n are derived using Ward'"'"'s method. The method includes computing the Approximate Weight of Evidence (AWE) for each cluster grouping then selecting that count associated with a maximum AWE and designating this count as s. For each cluster in each grouping 1 through s, the method finds the member vector farthest from the cluster'"'"'s centroid and defines this as the cluster'"'"'s radius. A single input sample for a motor under supervision is read in, following by computing and processing an FFT based on the input sample for a newly generated feature vector and then checking whether the newly generated feature vector is inside any of the clusters 1 through s as defined for each cluster by its respective radius, and if not, outputting a warning signal.
-
Citations
12 Claims
-
1. A computer-based method for ascertaining anomalies in an electric motor, comprising the steps of:
-
(a) obtaining a set of samples of supply current for a motor known to be in a normal condition; (b) computing and processing a set of Fast Fourier Transforms (FFT'"'"'s) based on waveforms of said supply current for said motor known to be in a normal condition so as to create a feature set of input vectors; (c) calculating all possible input vector assignments for clusters 1 through n using Ward'"'"'s method, where n is the number of p-dimensional input vectors, in accordance with the following steps; Step 1;
initializing,Step 2;
setting k, the number of groups considered, equal to n, the number of input vectors,Step 3;
setting the "best value" Z[pk13 1, qk-- 1, k-1] equal to some initial large value and putting i equal to the smallest active identification number,Step 4;
setting j equal to the first active identification number greater than i,Step 5;
computing Z[i,j,k-1] associated with the hypothesized union of sets i and j,Step 6;
determining whether Z[i,j,k-1] is better than the best value Z[pk-- 1, qk-- 1, k-1] up to this comparison;
if the answer yes, this is followed byStep 7, replacing the old value of Z[pk-- 1, qk-- 1, k-1] by Z[i,j,k-1] and making pk-- 1=i and qk-- 1=j and proceeding to Step 8;
if the answer is no, this is followed byStep 8, ascertaining whether j is equal to the last active identification number;
if the answer is no, thenStep 9, set j equal to the next higher active identification number, and return to Step 5;
if the answer is yes, proceed toStep 10, determine whether i is egual to the next higher active identification number;
if not proceed to Step 11;
if yes, proceed to Step 12,Step 11, set i equal to the next higher active identification number (Step
11) and return to Step 4;Step 12, find a best union of 2 sets and identify it by the identification numbers pk-- 1 and qk-- 1, the value associated with their union being Z[pk-- 1, qk-- 1, k-i], then to Step 13, identify the new union by the number pk-- 1 and make the identification number qk-- 1 inactive, Step 14, calculate trace values for 2 clusters that were merged and for the new cluster which was formed by the merge, Step 15, compute lambda-- r and E-- r for this step in the clustering procedure, Step 16, determine whether k, the number of groups under consideration, is equal to 2;
if not, proceed to Step 17;
if the answer is yes, proceed to Step 18,Step 17, set k equal to k-1 (Step
17) and return to step 2;Step 18, calculate the approximate weight of evidence for all steps from 1 to n-1; (d) computing the Approximate Weight of Evidence (AWE) for said cluster counts 1 through n; (e) selecting that count associated with a maximum AWE and designating this count as s; (f) for each cluster 1 through s, finding the member vector farthest from the cluster'"'"'s centroid and defining this as the cluster'"'"'s radius; (g) reading in a single input sample for a motor under supervision; (h) computing and processing an FFT based on said input sample for newly generating a feature vector; and (i) checking whether said newly generated feature vector is inside any of said clusters 1 through s as defined for each cluster by said radius; and (j) if not, outputting a warning signal.
-
-
2. A computer-based method for ascertaining anomalies in an electric motor, comprising the steps of:
-
(a) sampling current flowing in a winding of a motor known to be not defective; (b) deriving from said current a set of input vectors comprising selected frequency component values obtained by taking a Fast Fourier Transform (FFT) of said current; (c) processing said set of vectors in accordance with a clustering program for finding vector assignments for all possible clusters from 1 to n, where n is the number of said input vectors; (d) selecting a number s of clusters out of said all possible clusters in accordance with a Bayesian approach that uses the log likelihoods to estimate the probability for the number of clusters; (e) assigning an effective cross-section for each of said clusters; (f) testing a new data vector from a motor to be monitored; (g) if said data vector falls within said effective cross-section of any of said s clusters, then labelling said data vector as not indicative of a motor problem; and (h) if said data vector falls outside of said effective cross-section of every one of said s clusters, then labelling said data vector as indicative of a motor problem. - View Dependent Claims (3, 4, 5, 6)
-
-
7. A computer-based method for ascertaining anomalies in an electric motor, comprising the steps of:
-
(a) sampling current flowing in a winding of a motor known to be not defective; (b) deriving from said current a set of input vectors comprising selected frequency component values obtained by taking a Fast Fourier Transform (FFT) of said current; (c) processing said set of vectors in accordance with Ward'"'"'s clustering method for finding vector assignments for all possible clusters from 1 to n, where n is the number of said input vectors, in accordance with the following steps; Step 1;
initializing,Step 2;
setting k, the number of groups considered, equal to n, the number of input vectors,Step 3;
setting the "best value" Z[pk-- 1, qk-- 1, k-1] equal to some initial large value and putting i equal to the smallest active identification number,Step 4;
setting j equal to the first active identification number greater than i,Step 5;
computing Z[i,j,k-1] associated with the hypothesized union of sets i and j,Step 6;
determining whether Z[i,j,k-1] is better than the best value Z[pk-- 1, qk-- 1, k-1] up to this comparison;
if the answer is yes, this is followed byStep 7, replacing the old value of Z[pk-- 1, qk-- 1, k-1] by Z[i,j,k-1] and making pk 1=i and qk 1=j and proceeding to Step 8;
if the answer is no, this is followed byStep 8, ascertaining whether j is equal to the last active identification number;
if the answer is no, thenStep 9, set j equal to the next higher active identification number, and return to Step 5;
if the answer is yes, proceed toStep 10, determine whether i is equal to the next higher active identification number;
if not proceed to Step 11;
if yes, proceed to Step 12,Step 11, set i equal to the next higher active identification number (Step
11) and return to Step 4;Step 12, find a best union of 2 sets and identify it by the identification numbers pk-- 1 and qk-- 1, the value associated with their union being Z[pk-- 1, qk-- 1, k-1], then to Step 13, identify the new union by the number pk-- 1 and make the identification number qk-- 1 inactive, Step 14, calculate trace values for 2 clusters that were merged and for the new cluster which was formed by the merge, Step 15, compute lambda-- r and E-- r for this step in the clustering procedure, Step 16, determine whether k, the number of groups under consideration, is equal to 2;
if not, proceed to step 17;
if the answer is yes, proceed to Step 18,Step 17, set k equal to k-1 (Step
17) and return to step 2;Step 18, calculate the approximate weight of evidence for all steps from 1 to n-1; (d) selecting a number s of clusters out of said all possible clusters by using an iterative procedure of Approximate Weight of Evidence (AWE); (e) assigning an effective radius for each of said s clusters, said radius being determined by the farthest member of each of said s clusters from the respective centroid of that cluster; (f) testing a new data vector from a motor to be monitored; (g) if said data vector falls within said effective radius of any of said s clusters, then labelling said data vector as not indicative of a motor problem; and (h) if said data vector falls outside of said effective radius of every one of said s clusters, then labelling said data vector as indicative of a motor problem.
-
-
8. A computer-based method for ascertaining anomalies in an electric motor, comprising the steps of:
-
training a system utilizing a set of Fast Fourier Transforms (FFT'"'"'s) of samples of supply current for a motor known to be in a normal condition so as to form a number s of clusters of input vector assignments, where s is a cluster count having a maximum Approximate Weight of Evidence (AWE) for each cluster 1 through s, finding the member vector farthest from the cluster'"'"'s centroid and identifying this as the cluster'"'"'s radius; reading in a single input sample for a motor under supervision; computing and processing an FFT based on said input sample for newly generating a feature vector; checking whether said newly generated feature vector is inside any of said clusters 1 through s as defined by said radius; and if not, then outputting a warning signal.
-
-
9. A computer-based system for ascertaining anomalies in an electric motor, comprising:
-
means for sampling current flowing in a winding of a motor known to be not defective; means for deriving from said current a set of input vectors comprising selected frequency component values obtained by taking a Fast Fourier Transform (FFT) of said current; means for processing said set of vectors in accordance with a clustering program for finding vector assignments for all possible clusters from 1 to n, where n is the number of said input vectors, means for selecting a number s of clusters out of said all possible clusters in accordance with a Bayesian approach that uses the log likelihoods to estimate the probability for the number of clusters; means for assigning an effective cross-section for each of said clusters; and means for testing a new data vector from a motor to be monitored and for labelling said data vector as indicative of a motor problem if said data vector falls outside of said effective cross-section of every one of said s clusters.
-
-
10. A computer-based system for ascertaining anomalies in an electric motor, comprising:
-
means for sampling current flowing in a winding of a motor known to be not defective; means for deriving from said current a set of input vectors comprising selected frequency component values obtained by taking a Fast Fourier Transform (FFT) of said current; means for processing said set of vectors in accordance with Ward'"'"'s clustering method for finding vector assignments for all possible clusters from 1 to n, where n is the number of said input vectors; means for selecting a number s of clusters out of said all possible clusters by using an iterative procedure of Approximate Weight of Evidence (AWE); means for assigning an effective radius for each of said s clusters, said radius being determined by the farthest member of each of said s clusters from the respective centroid of that cluster; and means for testing a new data vector from a motor to be monitored and for labelling said data vector as indicative of a motor problem if said data vector falls outside of said effective radius of every one of said s clusters.
-
-
11. A computer-based method for ascertaining anomalies in an electric motor, comprising:
- a trained system utilizing a set of Fast Fourier Transforms (FFT'"'"'s) of samples of supply current for a motor known to be in a normal condition so as to form a number s of clusters of input vector assignments, where s is a cluster count having a maximum Approximate Weight of Evidence (AWE);
for each cluster 1 through s, means for finding the member vector farthest from the cluster'"'"'s centroid and identifying this as the cluster'"'"'s radius; means for reading in a single input sample for a motor under supervision; means for computing and processing an FFT based on said input sample for a newly generated feature vector; and means for checking whether said newly generated feature vector is inside any of said clusters 1 through s as defined by said radius and if not, for then outputting a warning signal.
- a trained system utilizing a set of Fast Fourier Transforms (FFT'"'"'s) of samples of supply current for a motor known to be in a normal condition so as to form a number s of clusters of input vector assignments, where s is a cluster count having a maximum Approximate Weight of Evidence (AWE);
-
12. A computer-based method for ascertaining anomalies in an electrical device, said device tending to exhibit a representative current having characteristics indicative of anomalous performance or condition, said method comprising the steps of:
-
(a) obtaining a set of samples of said representative current for a device of the type under consideration known to be in a normal condition; (b) computing and processing a set of Fast Fourier Transforms (FFT'"'"'s) based on waveforms of said representative current for said device known to be in a normal condition so as to create a feature set of input vectors; (c) calculating all possible input vector assignments for clusters 1 through n using Ward'"'"'s method, where n is the number of p-dimensional input vectors; (d) computing the Approximate Weight of Evidence (AWE) for said cluster counts 1 through n; (e) selecting that count associated with a maximum AWE and designating this count as s; (f) for each cluster 1 through s, finding the member vector farthest from the cluster'"'"'s centroid and defining this as the cluster'"'"'s radius; (g) reading in a single input sample for a device under supervision; (h) computing and processing an FFT based on said input sample for a newly generating a feature vector; and (i) checking whether said newly generated feature vector is inside any of said clusters 1 through s as defined for each cluster by said radius; and (j) if not, outputting a warning signal.
-
Specification