US 11,681,282 B2
Systems and methods for determining relationships between defects
Andrew Poh, San Francisco, CA (US); Andre Frederico Cavalheiro Menck, New York, NY (US); Arion Sprague, San Francisco, CA (US); Benjamin Grabham, London (GB); Benjamin Lee, London (GB); Bianca Rahill-Marier, New York, NY (US); Gregoire Omont, London (GB); Jim Inoue, Kirkland, WA (US); Jonah Scheinerman, Ann Arbor, MI (US); Maciej Albin, London (GB); Myles Scolnick, Englewood, CO (US); Paul Gribelyuk, Jersey City, NJ (US); Steven Fackler, Menlo Park, CA (US); Tam-Sanh Nguyen, Olney, MD (US); Thomas Powell, London (GB); and William Seaton, New York, NY (US)
Assigned to Palantir Technologies Inc., Denver, CO (US)
Filed by Palantir Technologies, Inc., Palo Alto, CA (US)
Filed on Apr. 10, 2020, as Appl. No. 16/845,797.
Application 16/845,797 is a continuation of application No. 15/385,664, filed on Dec. 20, 2016, granted, now 10,620,618.
Prior Publication US 2020/0241518 A1, Jul. 30, 2020
Int. Cl. G05B 23/02 (2006.01); G06N 20/00 (2019.01); G06Q 10/0639 (2023.01); H01L 21/66 (2006.01); G06F 16/2457 (2019.01); G06F 11/07 (2006.01); G06F 16/35 (2019.01); G06F 16/28 (2019.01)
CPC G05B 23/0235 (2013.01) [G06N 20/00 (2019.01); G06Q 10/06395 (2013.01); G06F 11/079 (2013.01); G06F 16/24578 (2019.01); G06F 16/285 (2019.01); G06F 16/35 (2019.01); H01L 22/20 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
at least one processor; and
at least one memory storing machine-readable instructions, wherein the at least one processor is configured to access the at least one memory and execute the machine-readable instructions to cause the system to:
train a machine learning model to determine feature weights used to determine relationships between defect data objects that represent defects in a manufacturing or industrial process, wherein the feature weights are determined to maximize an issue quality score among related defect data objects and a feature weight of the feature weights comprises an attribute of the manufacturing or the industrial process;
obtain a first defect data object and a second defect data object from a database to determine a defect similarity between the first defect data object and the second defect data object, the first defect data object comprising first unstructured data and the second defect data object comprising second unstructured data;
extract frequencies of terms from the first defect data object and the second defect data object;
determine a comparison metric with which to determine the defect similarity based at least in part on the extracted frequencies;
output, using the trained machine learning model, the defect similarity between the first defect data object and the second defect data object;
determine, using the trained machine learning model, that the first defect data object and the second defect data object are related based on the comparison metric satisfying a relatedness level;
generate and store an issue data object comprising the first defect data object and the second defect data object in the database, the issue data object indicative of a diagnosis of a defect common to the first defect data object and the second defect data object;
and
determine additional defect data objects in which respective comparison value scores indicative of a degree to which additional defect data objects correspond to or are related to the issue data object exceed a threshold comparison value score; and
retrain the machine learning model based on the additional defect data objects.