IP Library › Granted Patent US 12,737,384
Granted Patent B2
US 12,737,384 · App. 18/950,675 · Granted Sep 15, 2026

Embedding-based fault resolution

Inventors: Christoph Fritzsche (Speyer, DE); Michael Philipp (Wiesloch, DE); Fabienne Spaethe (Ittlingen, DE)
Assignee: SAP SE
G06F16/284G06F16/2455
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,737,384
App. No.
18/950,675
Granted
Sep 15, 2026
Kind
B2
Abstract

A system and method include identification of a future fault period and a future fault amount associated with a component, determination of production characteristics of the component, generation of a multi-dimensional vector representing the future fault period, the future fault amount and the production characteristics, searching of a vector database of multi-dimensional vectors to identify a second multi-dimensional vector similar to the first multi-dimensional vector and first actions corresponding to the second multi-dimensional vector, and automatic execution of the first actions to resolve the future fault period and the future fault amount associated with the component.

Claims (101)

1 . A system comprising:

a memory storing program code; and

at least one processing unit to execute the program code to cause the system to:

identify a fault comprising a fault period and a fault amount associated with a component, the fault amount comprising a shortage in a supply of the component;

transmit a prompt to an embedding model, the prompt comprising dimensional values representing the fault;

receive, from the embedding model, a multi-dimensional vector for the fault, the multi-dimensional vector representing the fault period and the fault amount;

search a vector database of multi-dimensional vectors, calculate a measurement of similarity between the multi-dimensional vector and a plurality of candidate multi-dimensional vectors of the vector database, and identify a second multi-dimensional vector of the plurality of candidate multi-dimensional vectors similar to the first multi-dimensional vector, based on the measurement of similarity, and identify first user actions corresponding to the second multi-dimensional vector;

present the first user actions to a user on a user interface;

in response to resolution of the fault period, detect a plurality of second user actions associated with the resolution of the fault period and the fault amount associated with the component; and

store a record in the vector database associating the multi-dimensional vector with data representing the plurality of second user actions.

2 . The system of claim 1 , wherein generation of the multi-dimensional vector comprises:

generation of text comprising the fault period, the fault amount, and the component; and

input of the text to an embedding model.

3 . The system of claim 1 , wherein storage of the record in the vector database comprises:

determination that the plurality of user actions are different from the first user actions; and

storage of the record in the vector database in response to the determination that the plurality of user actions are different from the first user actions.

4 . The system of claim 1 , the at least one processing unit to execute the program code to cause the system to:

identify a second fault period and a second fault amount associated with the first component;

generate a third multi-dimensional vector representing the second fault period and the second fault amount;

search the vector database of multi-dimensional vectors to identify the second multi-dimensional vector as similar to the third multi-dimensional vector and the first user actions corresponding to the second multi-dimensional vector;

present the first user actions;

detect a second plurality of user actions associated with resolution of the second fault period and the second fault amount associated with the component;

determine that the second plurality of user actions are similar to the first user actions; and

in response to the determination that the second plurality of user actions are similar to the first user actions, do not store a record in the vector database associating the third multi-dimensional vector with data representing the second plurality of user actions.

5 . The system of claim 1 , the at least one processing unit to execute the program code to cause the system to:

identify a second fault period and a second fault amount associated with a second component;

generate a third multi-dimensional vector representing the second fault period and the second fault amount;

search the vector database of multi-dimensional vectors to identify a fourth multi-dimensional vector as similar to the third multi-dimensional vector and second user actions corresponding to the fourth multi-dimensional vector;

present the second user actions;

detect a second plurality of user actions associated with resolution of the second fault period and the second fault amount associated with the second component;

determine that the second plurality of user actions are different from the second user actions; and

in response to the determination that the second plurality of user actions are different from the second user actions, store a second record in the vector database associating the third multi-dimensional vector with data representing the second plurality of user actions.

6 . The system of claim 5 , wherein generation of the multi-dimensional vector comprises:

generation of text comprising the fault period, the fault amount and the component;

input of the text to an embedding model, and

wherein generation of the third multi-dimensional vector comprises:

generation of second text comprising the second fault period, the second fault amount, and the second component; and

input of the second text to the embedding model.

7 . A method comprising:

identifying a fault comprising a fault period and a fault amount associated with a component, the fault amount comprising a shortage in a supply of the component;

transmitting a prompt to an embedding model, the prompt comprising dimensional values representing the fault;

receiving, from the embedding model, a multi-dimensional vector for the fault, the multi-dimensional vector representing the fault period and the fault amount;

searching a vector database of multi-dimensional vectors, calculating a measurement of similarity between the multi-dimensional vector and a plurality of candidate multi-dimensional vectors of the vector database, and identifying a second multi-dimensional vector of the plurality of candidate multi-dimensional vectors similar to the first multi-dimensional vector, based on the measurement of similarity, and identifying first actions corresponding to the second multi-dimensional vector;

presenting the first user actions to a user on a user interface; and

automatically executing the first actions to resolve the fault period and the fault amount associated with the component.

8 . The method of claim 7 , wherein generating the multi-dimensional vector comprises:

generating text comprising the fault period, the fault amount, and the component; and

inputting the text to an embedding model.

9 . The method of claim 7 , wherein automatically executing the first actions comprises:

determining that a similarity between the first multi-dimensional vector and the second multi-dimensional vector is greater than a threshold; and

automatically executing the first actions in response to determining that the similarity between the first multi-dimensional vector and the second multi-dimensional vector is greater than the threshold.

10 . The method of claim 9 , further comprising:

identifying a second fault period and a second fault amount associated with the first component;

generating a third multi-dimensional vector representing the second fault period and the second fault amount;

searching the vector database of multi-dimensional vectors to identify the second multi-dimensional vector as similar to the third multi-dimensional vector and the first user actions corresponding to the second multi-dimensional vector;

determining that a similarity between the third multi-dimensional vector and the second multi-dimensional vector is less than a threshold;

in response to determining that the similarity between the third multi-dimensional vector and the second multi-dimensional vector is less than the threshold, presenting the first user actions;

detecting a second plurality of user actions associated with resolution of the second fault period and the second fault amount associated with the component;

determining that the second plurality of user actions are not similar to the first actions; and

in response to the determination that the second plurality of user actions are not similar to the first actions, storing a record in the vector database associating the third multi-dimensional vector with data representing the second plurality of user actions.

11 . The method of claim 7 , further comprising:

identifying a second fault period and a second fault amount associated with the first component;

generating a third multi-dimensional vector representing the second fault period and the second fault amount;

searching the vector database of multi-dimensional vectors to identify the second multi-dimensional vector as similar to the third multi-dimensional vector and the first actions corresponding to the second multi-dimensional vector;

presenting the first actions;

detecting a second plurality of user actions associated with resolution of the second fault period and the second fault amount associated with the component;

determining that the second plurality of user actions are not similar to the first actions; and

in response to the determination that the second plurality of user actions are not similar to the first actions, storing a record in the vector database associating the third multi-dimensional vector with data representing the second plurality of user actions.

12 . The method of claim 7 , further comprising:

identifying a second fault period and a second fault amount associated with a second component;

generating a third multi-dimensional vector representing the second fault period and the second fault amount;

searching the vector database of multi-dimensional vectors to identify a fourth multi-dimensional vector as similar to the third multi-dimensional vector and second actions corresponding to the fourth multi-dimensional vector;

automatically executing the second actions to resolve the fault period and the fault amount associated with the component.

13 . The method of claim 12 , wherein generating the multi-dimensional vector comprises:

generating text comprising the fault period, the fault amount, and the component;

inputting the text to an embedding model, and

wherein generating the third multi-dimensional vector comprises:

generating second text comprising the second fault period, the second fault amount, and the second component; and

inputting the second text to the embedding model.

14 . One or more non-transitory computer-readable recording media storing program code, the program code executable by at least one processing unit of a computing system to:

identify a fault comprising a fault period and a fault amount associated with a component, the fault amount comprising a shortage in a supply of the component;

transmit a prompt to an embedding model, the prompt comprising dimensional values representing the fault;

receive, from the embedding model, a multi-dimensional vector for the fault, the multi-dimensional vector representing the fault period and the fault amount;

search a vector database of multi-dimensional vectors, calculate a measurement of similarity between the multi-dimensional vector and a plurality of candidate multi-dimensional vectors of the vector database, and identify a second multi-dimensional vector of the plurality of candidate multi-dimensional vectors similar to the first multi-dimensional vector, based on the measurement of similarity, and identify first user actions corresponding to the second multi-dimensional vector;

present the first user actions to a user on a user interface;

in response to resolution of the fault period, detect a plurality of second user actions associated with the resolution of the fault period and the fault amount associated with the component; and

store a record in the vector database associating the multi-dimensional vector with data representing the plurality of second user actions.

15 . The one or more non-transitory computer-readable recording media of claim 14 , wherein generation of the multi-dimensional vector comprises:

generation of text comprising the fault period, the fault amount, and the component; and

input of the text to an embedding model.

16 . The one or more non-transitory computer-readable recording media of claim 14 , wherein storage of the record in the vector database comprises:

determination that the plurality of user actions are different from the first user actions; and

storage of the record in the vector database in response to the determination that the plurality of user actions are different from the first user actions.

17 . The one or more non-transitory computer-readable recording media of claim 14 , the program code executable by at least one processing unit of a computing system to:

identify a second fault period and a second fault amount associated with a second component;

generate a third multi-dimensional vector representing the second fault period and the second fault amount;

search the vector database of multi-dimensional vectors to identify a fourth multi-dimensional vector as similar to the third multi-dimensional vector and second user actions corresponding to the fourth multi-dimensional vector;

present the second user actions;

detect a second plurality of user actions associated with resolution of the second fault period and the second fault amount associated with the second component;

determine that the second plurality of user actions are different from the second user actions; and

in response to the determination that the second plurality of user actions are different from the second user actions, store a second record in the vector database associating the third multi-dimensional vector with data representing the second plurality of user actions.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 18, 2024
From: FRITZSCHE, CHRISTOPH; PHILIPP, MICHAEL; SPAETHE, FABIENNE
To: SAP SE
Reel/Frame 069302/0455 →
Continuity (1)
Related Publication 20260140967A1 · May 21, 2026
References Cited (29)
US 6650779B2 · Vachtesvanos · 2003 [cited by examiner]
US 9400490B2 · Turolla · 2016 [cited by examiner]
US 9818242B2 · Volponi · 2017 [cited by examiner]
US 10331817B1 · Tertois · 2019 [cited by examiner]
US 11704318B1 · Keivanloo · 2023 [cited by examiner]
US 12055998B2 · Settle · 2024 [cited by examiner]
US 12242496B2 · Kroll · 2025 [cited by examiner]
US 12299397B2 · Turkkan · 2025 [cited by examiner]
US 12443635B1 · Wolff · 2025 [cited by examiner]
US 20030225466A1 · Yulevitch · 2003 [cited by examiner]
US 20040039499A1 · Felke · 2004 [cited by examiner]
US 20050114743A1 · Moorhouse · 2005 [cited by examiner]
US 20080010531A1 · Hendler · 2008 [cited by examiner]
US 20080307269A1 · Bennett · 2008 [cited by examiner]
US 20120101622A1 · Yun · 2012 [cited by examiner]
US 20130144866A1 · Jerzak · 2013 [cited by examiner]
US 20150219530A1 · Li · 2015 [cited by examiner]
US 20170075946A1 · Bossa · 2017 [cited by examiner]
US 20190171622A1 · Zong · 2019 [cited by examiner]
US 20190171644A1 · Xu · 2019 [cited by examiner]
US 20190392533A1 · Thurner · 2019 [cited by examiner]
US 20200233391A1 · Ma · 2020 [cited by examiner]
US 20200272923A1 · Banerjee · 2020 [cited by examiner]
US 20210303381A1 · Baldassarre · 2021 [cited by examiner]
US 20230160942A1 · Liu · 2023 [cited by examiner]
US 20230316711A1 · Unuma · 2023 [cited by examiner]
US 20240078444A1 · Fang · 2024 [cited by examiner]
US 20250004431A1 · Wang · 2025 [cited by examiner]
US 20250355748A1 · Wang · 2025 [cited by examiner]