| CPC G05B 19/41885 (2013.01) [G05B 19/4183 (2013.01); G05B 19/4188 (2013.01); G06F 18/2113 (2023.01); G06F 18/2155 (2023.01); G06F 18/23 (2023.01)] | 13 Claims |
|
1. A method comprising:
performing a manufacturing process that comprises steps of:
manufacturing at least one ball bearing;
manufacturing a metal ring;
assembling the at least one ball bearing with the metal ring to manufacture a wheel bearing;
at a particular step of the manufacturing process attaching the wheel bearing to a test point to generate component waveforms;
identifying, at a computing device, a test response parameter of a wheel bearing;
receiving, at the computing device, a first plurality of component waveforms associated with the test response parameter from the test point, wherein each waveform of the plurality of waveforms comprises a plurality of datapoints;
generating, at the computing device, a model;
training, at the computing device and on the first plurality of component waveforms, the model, thereby generating one or more parameters associated with the model,
wherein training the model further comprises generating a reduced feature space associated with the first plurality of component waveforms,
wherein the generating the reduced feature space further comprises:
applying a Z-standardization to the first plurality of component waveforms, thereby producing a result; and
passing the result to a linear principal component analysis;
clustering the data points in the reduced feature space via a density-based spatial clustering of applications with noise algorithm, wherein the clustering comprises:
identifying one or more core-points, wherein the one or more core points are characterized by containing a minimum number of data points within a radius of neighborhood, wherein the generating the one or more parameters associated with the model further comprises:
determining the minimum number of data points as two times a number of principal components of the linear principal component analysis; and
determining the radius of neighborhood as the elbow point in a k-distance graph, wherein k is the minimum number of data points minus one;
assigning one or more data points to a cluster, wherein a cluster is based on a core point, wherein the training the model further comprises generating cluster labels for each cluster and storing the generated cluster labels;
receiving a second plurality of component waveforms associated with the test response parameter from the test point;
accessing the trained model;
indicating, using the trained model, whether any of the waveforms of the second plurality of component waveforms comprises an anomaly; and, for each indicated waveform:
reviewing the indicated waveform;
for each reviewed waveform not comprising an anomaly:
labelling the waveform; and
based at least in part on at least one of the indicated waveforms comprising an anomaly:
returning the wheel bearing to one of the steps of the manufacturing process preceding the particular step.
|