DE102020202092A1 - Computer-implemented method for identifying defects and / or defect patterns in objects to be tested - Google Patents

Abstract

Computer-implemented method (100) for identifying errors and / or error patterns in objects to be checked, the objects to be checked in particular comprising control devices and / or components thereof, the method (100) comprising the steps of: creating (110) a knowledge graph (KG ) based on test data (D) of the objects to be checked; deriving (120) rules (R) from the knowledge graph (KG).

Description

State of the art

The disclosure relates to a computer-implemented method for identifying defects and / or defect patterns in objects to be tested.

In product development, the objects to be developed are subjected to various tests and inspections during development and / or at the end of development. The object to be tested, engl. Device under test (DUT) is, for example, an electromechanical object, in particular a control device and / or components thereof. For example, such a test can be a test for checking electromagnetic compatibility (EMC).

The evaluation of the test data, in particular the identification of errors and / or error patterns in the test data, is very time-consuming and can hardly be carried out manually due to the large amount of test data, among other things.

Disclosure of the invention

• Erstellen eines Knowledge Graphs basierend auf Testdaten der zu prüfenden Objekte; Ableiten von Regeln aus dem Knowledge Graph.

The disclosure relates to a computer-implemented method for identifying errors and / or error patterns in objects to be checked, the objects to be checked in particular comprising control devices and / or components thereof, the method comprising the steps:

• Creation of a knowledge graph based on test data of the objects to be tested; Deriving rules from the knowledge graph.

It is therefore proposed to store the test data in a structured manner in a knowledge graph and to derive rules based on the knowledge graph. Using these rules, errors or error patterns can then be identified.

In knowledge-based systems, a knowledge graph is understood to be a structured storage of knowledge in the form of a knowledge graph. Knowledge graphs encompass entities and show relationships between entities.

The derivation of rules from a knowledge graph is described, for example, in

Mohamed H. Gad-Elrab, Daria Stepanova, Jacopo Urbani, Gerhard Weikum. 2016 Exception-Enriched Rule Learning from Knowledge Graphs. International Semantic Web Conference (1) 2016: 234-251 , and

Hai Dang Tran, Daria Stepanova, Mohamed H. Gad-Elrab, Francesca A. Lisi, Gerhard Weikum. 2019 . Towards Nonmonotonic Relational Learning from Knowledge Graphs. ILP 2016: 94-107.

According to a preferred embodiment, the test data include test results and / or properties of the objects to be tested. The test results also include, for example, specifications of the tests carried out. The properties of the objects to be checked include, for example, information about the type and properties of the objects to be checked. The test data can be in text form, for example. The test data can advantageously be supplemented with specialist knowledge and / or expert knowledge from specific areas.

The creation of the knowledge graph can be done manually, for example. In the case of extensive test data in particular, it can prove to be advantageous if the creation of the knowledge graph takes place semi-automatically or automatically.

According to a preferred embodiment it is provided that the method further comprises: creating an embedding model of the knowledge graph. This proves to be advantageous since the knowledge graph created based on the test data can initially be incomplete and can be completed by the embedding model. An embedding model known from the prior art is, for example, TransE disclosed in Antoine Bordes, Nicolas Usunier, Alberto Garcia-Durän. 2013 . Translating Embeddings for Modeling Multi-relational Data. In proceedings of NIPS, pages 2787-2795, 2013 . Further models are described in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 . Rule learning from knowledge graphs guided by embedding models. International Semantic Web Conference (1), 2018: 72-90 .

According to a preferred embodiment it is provided that the method further comprises: evaluating rules based on the embedding model. Using the embedding model, the rules derived from the knowledge graph, in particular a quality of the rules, can be evaluated.

According to a preferred embodiment it is provided that the step of deriving rules comprises an iterative process for constructing rules. The iterative process is described, for example, in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 . Rule learning from knowledge graphs guided by embedding models. International Semantic Web Conference (1), 2018: 72-90 .

According to a preferred embodiment it is provided that the method further comprises: outputting final rules.

According to a preferred embodiment, it is provided that final rules include errors and / or error patterns in objects to be checked. On the basis of these errors or error patterns, it can then be advantageously explained why certain tests or checks have led to a certain result for certain objects. Advantageously, by outputting final rules, reasons for the failure of certain tests or checks on the relevant objects can be recognized. Advantageously, one or more cause-effect relationships can also be recognized in the final rules.

Further preferred embodiments relate to a device comprising at least one processor, the processor being designed to carry out the method according to the embodiments.

Further preferred embodiments relate to a use of a method according to the embodiment and / or a device according to the embodiments for evaluating test data in order to identify errors and / or error patterns in the products to be tested.

Such a use is suitable, for example, in the area of product development and / or in the area of materials science. The test data include, for example, existing test data from tests that were obtained during product development and / or final testing of an object, in particular an electromechanical object, in particular during an EMC test.

Advantageously, errors and / or error patterns in the objects to be tested and / or around common errors and / or error patterns in products and / or in product families can be identified.

• 1 eine schematische Übersicht eines Verfahrens gemäß einer Ausführungsform in einem Blockdiagramm;
• 2 eine schematische Darstellung von Schritten eines Verfahrens gemäß einer Ausführungsform in einem Flussdiagramm, und
• 3 eine schematische Darstellung von Schritten eines Verfahrens gemäß einer weiteren Ausführungsform in einem Flussdiagramm, und
• 4 eine beispielhafte Darstellung eines Knowledge Graphs.

Further features, possible applications and advantages of the invention emerge from the following description of exemplary embodiments of the invention, which are shown in the figures of the drawing. All of the features described or shown form the subject matter of the invention individually or in any combination, regardless of how they are summarized in the patent claims or their reference, and regardless of their wording or representation in the description or in the drawing.

• 1 a schematic overview of a method according to an embodiment in a block diagram;
• 2 a schematic representation of steps of a method according to an embodiment in a flowchart, and
• 3 a schematic representation of steps of a method according to a further embodiment in a flowchart, and
• 4th an exemplary representation of a knowledge graph.

1 shows a schematic representation of steps of a method 100 for identifying errors and / or error patterns in objects to be checked, the objects to be checked in particular comprising control devices and / or components thereof in a block diagram. Also includes 1 a schematic representation of data, in particular test data.

• einen Schritt 110 zum Erstellen eines Knowledge Graphs KG basierend auf Testdaten D der zu prüfenden Objekte,
• und einen Schritt 120 zum Ableiten von Regeln R aus dem Knowledge Graph KG.

The procedure 100 includes the following steps:

• one step 110 to create a knowledge graph KG based on test data D. the objects to be checked,
• and one step 120 for deriving rules R. from the Knowledge Graph KG .

According to the procedure 100 so are the test data D. in a knowledge graph KG to be saved in a structured manner and based on the Knowledge Graph KG rules R. derived. Using these rules R. errors or error patterns can then be identified.

The Knowledge Graph KG includes a structured storage of the test data D. in the form of a knowledge graph. Knowledge graphs generally include entities and show relationships between entities.

The derivation 120 of rules from a knowledge graph is described, for example, in Mohamed H. Gad-Elrab, Daria Stepanova, Jacopo Urbani, and Gerhard Weikum. 2016 Exception-Enriched Rule Learning from Knowledge Graphs. International Semantic Web Conference (1) 2016: 234-251 , and Hai Dang Tran, Daria Stepanova, Mohamed H. Gad-Elrab, Francesca A. Lisi, Gerhard Weikum. 2019. Towards Nonmonotonic Relational Learning from Knowledge Graphs. ILP 2016: 94-107.

According to a preferred embodiment, the test data comprise D. Test results and / or properties of the objects to be tested. The test results also include, for example, specifications of the tests carried out. the Properties of the objects to be checked include, for example, information about the type and properties of the objects to be checked. The test data D. can be in text form, for example. The test data can D. advantageously supplemented with specialist knowledge, expert knowledge from specific areas.

Creating 110 of the Knowledge Graph KG can be done manually, for example. Especially with extensive test data D. it can prove beneficial when creating 110 of the Knowledge Graph KG semi-automatically or automatically.

According to a preferred embodiment it is provided that the method has a further step 130 to create an embedding model M. of the Knowledge Graph KG includes. This proves to be beneficial because of the test data D. based knowledge graph KG may be incomplete at first, and by the embedding model M. can be completed. An embedding model M known from the prior art is, for example, TransE disclosed in Antoine Bordes, Nicolas Usunier, Alberto Garcia-Durän. 2013. Translating Embeddings for Modeling Multi-relational Data. In proceedings of NIPS, pages 2787-2795, 2013 . Further models are described in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 . Rule learning from knowledge graphs guided by embedding models. International Semantic Web Conference (1), 2018: 72-90 .

According to one embodiment, the method further comprises a step 140 to evaluate rules R. based on the embedding model M. . This is in 1 schematically by the double arrow 140 shown. Using the embedding model M. can use the knowledge graph KG derived rules R. , in particular a quality of the rules R. , be rated.

According to a preferred embodiment it is provided that the step of deriving 120 rules R comprises an iterative process for constructing rules R. Such an iterative process is described, for example, in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 . Rule learning from knowledge graphs guided by embedding models. International Semantic Web Conference (1), 2018: 72-90 .

According to a preferred embodiment it is provided that the method 100 one step further 150 for outputting final rules includes Rf.

According to a preferred embodiment it is provided that final rules Rf include errors and / or error patterns in objects to be checked. On the basis of these errors or error patterns, it can then be advantageously explained why certain tests or checks have led to a certain result for certain objects. Advantageously, by spending 150 final rules Rf recognize reasons for the failure of certain tests or checks on the objects in question. One or more cause-effect relationships can advantageously also be recognized in the final rules Rf.

Further preferred embodiments relate to the use of a method 100 according to the embodiment for evaluating test data in order to identify defects and / or defect patterns in the products to be tested.

Such a use is suitable, for example, in the area of product development and / or in the area of materials science. The test data D. include, for example, existing test data D. from tests that were obtained during product development and / or the final test of an, in particular electromechanical, object, in particular during an EMC test.

Advantageously, errors and / or error patterns in the objects to be tested and / or around common errors and / or error patterns in products and / or in product families can be identified.

2 shows the steps 110 and 120 of the procedure 100 in a schematic representation in a flow chart.

3 shows the steps 110 , 120 , 130 , 140 , 150 of the procedure 100 in a schematic representation in a flow chart. The step 130 to create the embedding model M. can also be before the step 120 for deriving rules R. are executed. It is also conceivable that the step 120 for deriving rules R. and / or the step 140 to evaluate rules R. and / or the step 150 for outputting final rules Rf for a respective R. is executed separately.

4th shows an exemplary representation of a knowledge graph KG . The Knowledge Graph KG is for example according to step 120 of the procedure 100 based on test data D. created. In 4th only a small excerpt is shown as an example.

The Knowledge Graph KG comprises entities according to the illustrated embodiment E1 until E10 and relationships between the entities E1 until E10 . Relationships are represented by the arrows R12 , R32 , R34 , R51 , R65 , R63 , R67 , R610 , R98 , R109 . The Knowledge Graph KG is based on test data, for example D. from tests of an ECU. entity E5 is a first product according to the illustrated embodiment 1 . entity E5 is of type Entity E1 . The relationship between the entities E5 and E1 is in this case "is type". "Type". The entity E1 is for example an engine control unit. Engine control unit. The entity E2 is for example a DCDC converter. entity E2 is part of Entity, for example E1 . The entity E1 thus shows entity E2 on. The relationship R12 between entity E1 and E2 is in this case "have""Haspart". entity E3 a DCDC converter is in condition 1 . entity E3 for example, is a copy of the entity E2 . In this case the relationship is R32 between entity E3 and entity E2 "Is a copy of""Copyof". entity E6 is a copy of, R65, the entity E5 . entity E6 is a product 1 in a first state, engl. Product1lnState1, where the state is given by the relationship R67 "Has condition", engl. "Has State", between the entities E6 and E7 is shown, and E7 the state of engine operation. "Engine running". entity E7 is through a first test, entity E10 , failed. This is done through the relationship R610 "Failed", engl. "Failed" between the entities E6 and E10 shown. entity E10 is of the type, relationship, according to the illustrated embodiment R109 , Entity E9 , where entity E9 for example an "antenna polarization". "Antenna polarization" is. entity E9 is of the type, relationship, according to the illustrated embodiment R98 , Entity E8 , where entity E8 for example, an "interference emission". "Radiated emission" is.

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Non-patent literature cited

• Mohamed H. Gad-Elrab, Daria Stepanova, Jacopo Urbani, Gerhard Weikum. 2016 Exception-Enriched Rule Learning from Knowledge Graphs. International Semantic Web Conference (1) 2016: 234-251 [0008]
• Hai Dang Tran, Daria Stepanova, Mohamed H. Gad-Elrab, Francesca A. Lisi, Gerhard Weikum. 2019 [0009]
• An embedding model known from the prior art is, for example, TransE disclosed in Antoine Bordes, Nicolas Usunier, Alberto Garcia-Durän. 2013 [0012]
• Translating Embeddings for Modeling Multi-relational Data. In proceedings of NIPS, pages 2787-2795, 2013 [0012]
• Further models are described in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 [0012, 0029]
• Rule learning from knowledge graphs guided by embedding models. International Semantic Web Conference (1), 2018: 72-90 [0012, 0014, 0029, 0031]
• According to a preferred embodiment it is provided that the step of deriving rules comprises an iterative process for constructing rules. The iterative process is described, for example, in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 [0014]
• The derivation 120 of rules from a knowledge graph is described, for example, in Mohamed H. Gad-Elrab, Daria Stepanova, Jacopo Urbani, and Gerhard Weikum. 2016 Exception-Enriched Rule Learning from Knowledge Graphs. International Semantic Web Conference (1) 2016: 234-251 [0026]
• An embedding model M known from the prior art is, for example, TransE disclosed in Antoine Bordes, Nicolas Usunier, Alberto Garcia-Durän. 2013. Translating Embeddings for Modeling Multi-relational Data. In proceedings of NIPS, pages 2787-2795, 2013 [0029]
• According to a preferred embodiment it is provided that the step of deriving 120 rules R comprises an iterative process for constructing rules R. Such an iterative process is described, for example, in Vinh Thinh Ho, Daria Stepanova, Mohamed Gad-Elrab, Evgeny Kharlamov, Gerhard Weikum. 2018 [0031]

Claims (10)

Computer-implemented method (100) for identifying errors and / or error patterns in objects to be checked, the objects to be checked in particular comprising control devices and / or components thereof, the method (100) comprising the steps: Creation (110) of a knowledge graph (KG) based on test data (D) of the objects to be checked; Deriving (120) rules (R) from the knowledge graph (KG). Method (100) according to Claim 1 , the test data (D) comprising test results and / or properties of the objects to be tested. Method (100) according to at least one of the preceding claims, wherein the creation (110) of the knowledge graph (KG) takes place semi-automatically or automatically. The method (100) according to at least one of the preceding claims, wherein the method (100) further comprises: creating (130) an embedding model (M) of the knowledge graph (KG). The method (100) according to at least one of the preceding claims, wherein the method (100) further comprises: evaluating (140) rules (R) based on the embedding model (M). The method (100) according to at least one of the preceding claims, wherein the step of deriving (120) rules (R) comprises an iterative process for constructing rules (R). The method (100) according to at least one of the preceding claims, wherein the method (100) further comprises: outputting (150) final rules (Rf). The method (100) according to at least one of the preceding claims, wherein the final rules (Rf) include errors and / or error patterns in objects to be checked. Apparatus comprising at least one processor, wherein the processor is designed, the method (100) according to at least one of Claims 1 until 8th to execute. Using a method (100) according to one of the Claims 1 until 8th and / or a device according to Claim 9 for evaluating test data (D) in order to identify defects and / or defect patterns in the products to be tested.

Images