US 12,314,047 B2
Learning method and system for determining prediction horizon for machinery
Cahit Bagdelen, Walldorf (DE); and Atreju Tauschinsky, Walldorf (DE)
Assigned to SAP SE, Walldorf (DE)
Filed by SAP SE, Walldorf (DE)
Filed on Apr. 28, 2023, as Appl. No. 18/308,798.
Application 18/308,798 is a continuation of application No. 17/524,195, filed on Nov. 11, 2021, granted, now 11,681,284.
Claims priority of provisional application 63/229,404, filed on Aug. 4, 2021.
Prior Publication US 2023/0259119 A1, Aug. 17, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G05B 23/02 (2006.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01)
CPC G05B 23/0283 (2013.01) [G05B 23/0224 (2013.01); G06F 18/214 (2023.01); G06F 18/217 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for training a failure prediction model, the method comprising:
dividing run-to-failure sequences into sets of a negative and a positive sub-sequences of data associated with a machinery asset, each set associated with each feature of a plurality of features associated with a failure event, wherein each of the run-to-failure sequences include an occurrence of the failure event, and wherein each run-to-failure sequence is divided in two based on a plurality of potential cut-off points;
determining relative frequency distributions of the positive and the negative sub-sequences for each of the plurality of features and for each of the plurality of potential cut-off points;
calculating distances between the determined relative frequency distributions of the positive and the negative sub-sequences for the plurality of features for each of the plurality of potential cut-off points; and
iteratively training a failure prediction model by using a classification algorithm to iteratively extract sets of features from the plurality of features, wherein the iteratively extracted sets of features include features that are subsets to one another and are determined at a corresponding iteration as relevant for predicting a potential prediction horizon for the machinery asset, wherein the failure prediction model is for use in determining a prediction horizon value for the machinery asset.