EP4489924A1 - Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produkten - Google Patents
Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produktenInfo
- Publication number
- EP4489924A1 EP4489924A1 EP23719669.6A EP23719669A EP4489924A1 EP 4489924 A1 EP4489924 A1 EP 4489924A1 EP 23719669 A EP23719669 A EP 23719669A EP 4489924 A1 EP4489924 A1 EP 4489924A1
- Authority
- EP
- European Patent Office
- Prior art keywords
- product
- products
- repositioning
- auto
- encoder
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/34—Sorting according to other particular properties
- B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
Definitions
- the invention relates to a computer-implemented method, a sorting system and a client-server system for repositioning defective products.
- the invention relates to a computer program, an electronically readable data carrier and a data carrier signal.
- AI Artificial intelligence algorithms
- the aim is not only to establish conformity with a reference product, but also to determine the degree of deviation from this “gold standard”.
- the object according to the invention is achieved by a method of the type mentioned at the beginning, the following steps being carried out: a) providing a first product at a first position and at least a second product at at least a second position on a shelf, such as a moving conveyor belt, b ) Detecting properties and position of the first product as first sensor data and of the at least one second product as at least second sensor data by at least one sensor means and providing the first and the at least second sensor data to an analysis device with a processor and a memory, c) Providing a reference model based on machine learning, which has predefined properties of a product with which the first and the at least one second product have the highest possible degree of agreement and is generated and trained accordingly, d) Determining a first parameter with regard to the degree of agreement of the first sensor data with the reference model using a first auto-encoder, e) determining at least one second parameter with regard to the degree of agreement of the at least second sensor data with the reference model using at least a second Auto-encoders
- a defective product is understood to be a product for which a corresponding parameter is recorded in the course of a quality inspection that lies outside a specified range of values.
- a parameter can be a parameter for a product that characterizes the product itself, such as dimensions or shape of the product, or number of product parts that form the product, and the like.
- An intended target value with a range of values for maximum permissible tolerances can be specified for the parameter or characteristic variable.
- the product shape can be described, for example, by the course of an outer contour, the course of inner contours of product parts or adjacent product parts, as well as by straight courses, radii of curvature and/or proportions of product parts to one another.
- a property of the product can, for example, be formed by a material property of the product, such as the color or the elasticity of the product.
- One or more properties can each be determined by corresponding sensors, such as an optical camera sensor or a mechanical pen, which is guided to the product and records a penetration depth or a retaining force with regard to the product elasticity.
- product properties can be determined using non-intrusive sensors, such as ultrasonic or radar sensors for determining the moisture content or metal content of a product.
- the position or orientation of the product on the conveyor belt can form parameters that are detected, for example, by an optical camera sensor.
- a sensor can record several properties or product parameters.
- a subsequently controlled robot can therefore select and pick products that are sorted according to the degree of deviation from the gold standard.
- products with a high deviation should be selected with a higher priority, and only when the sorting robot is available are the products with a lower deviation sorted out.
- a comparison means determining the degree of agreement between sensor data and the reference model.
- a sorting order is thus achieved in which one of the two criteria is achieved: • All defective products in a specific product batch have been sorted out into a designated container.
- This best-effort approach has the advantage of lowering the performance requirements of the robotics solution, making automated visual quality control more accessible even for smaller systems.
- step g) for the first and at least one second characteristic each provides a numerical value, for example the reconstruction error from the auto-encoder.
- the numerical values obtained in this way can now be ranked based on their amount and output as a list, which uses the ranked numerical values to represent a prioritization of products that are subsequently to be repositioned on the shelf or sorted out.
- an auto-encoder is particularly suitable for a comparison with a reference model because it can be implemented easily and at high speed.
- the manipulation device can transport or sort out products from their positions on the shelf where they were provided to alternative positions.
- the shelf is covered by a working area of the manipulation device.
- the shelf can only be available as a work area for a limited time if subsequent further work steps prevent or restrict access to the shelf.
- the shelf in the form of a conveyor belt can move the provided products to an adjacent location, such as a container away from the shelf, which is also located in the working area of the manipulation device. The speed of movement of the conveyor belt determines the time availability of the products on the shelf in the working area of the manipulation device.
- the work area can also be restricted if the products are further processed directly on the shelf, in that another processing device covers the products and the shelf is therefore not accessible as a work area.
- step f) The process can be repeated in step f), whereby step a) can be skipped if there are further products on the shelf that have not yet been examined.
- the first auto-encoder and the at least one second auto-encoder are formed by a common auto-encoder.
- An autoencoder is an artificial neural network that is used to learn efficient coding.
- the goal of an autoencoder is to learn a compressed representation, the so-called "encoding" for a set of data and thus also to extract essential features. This means that it can be used for dimension reduction.
- the autoencoder uses three or more layers: an input layer: in face recognition, for example, the neurons could map the pixels of a photograph. • some significantly smaller layers (compared to the input layer and output layer) that form the encoding.
- the autoencoder proves to be a particularly good implementation in the system.
- the first characteristic is formed by the reconstruction error of the first auto-encoder and the at least one second characteristic is formed by the reconstruction error of the at least one second auto-encoder.
- the reference model in step c) is generated and provided using federated learning by a connected, distributed client-server system.
- a newly added client therefore does not need to be trained specifically, but can rely on a global model through federated learning.
- steps d) to f) are carried out by the analysis device.
- a sorting system of the type mentioned at the outset comprising a shelf for providing a first product at a first position and at least a second product at at least a second position, at least one sensor means for detecting properties and position of the first product and the at least one second product, an analysis device with a processor and a memory and a manipulation device for repositioning products, and the system is set up to carry out the method according to the invention.
- the object according to the invention is solved by a client-server system of the type mentioned at the beginning, comprising a sorting system according to the invention, which is located as a client at an edge, and the client-server system uses federated learning to create a global model based on machine learning generated, trained and provided to the connected sorting system as a local model for use in the method according to the invention.
- the object according to the invention is solved by a computer program comprising commands which, when executed by a computer, cause it to carry out the method according to the invention.
- the object according to the invention is achieved by an electronically readable data carrier with readable control information stored thereon, which includes at least the computer program according to the invention and is designed in such a way that it carries out the method according to the invention when the data carrier is used in a computing device.
- Fig. 1 an exemplary embodiment of a sorting system according to the invention
- FIG. 2 an exemplary embodiment of a flow chart of the method according to the invention
- Fig. 3 an exemplary embodiment of a client-server system according to the invention.
- Fig. 1 shows a sorting system according to the invention.
- a shelf 1 for providing products at respective positions is formed by a conveyor belt with products transported thereon.
- the conveyor belt moves the products placed thereon at a predefined speed la in a transport direction.
- a sensor means 2 in the form of a camera system detects the products, possibly with unfavorable impairment of the field of vision, and generates the respective sensor data SD.
- An analysis device 3 in particular with an auto-encoder, carries out a computer-implemented recognition algorithm for visual quality control for the sensor data provided.
- a process based on artificial intelligence is carried out and deviations from a reference model are determined, in particular based on the “reconstruction error” of an auto-encoder.
- the reference model is created, trained and saved in a computing device with valid and permissible product data (gold standard).
- the model can either be local or based on a global model, which ches over several, distributed models of further computing devices are generated according to the principle of “federated learning” and made available to the said computing device. This allows the local effort for providing the said local model to be reduced and the accuracy can be improved.
- Corresponding data sets are thus formed from the sensor data SD using the respective reconstruction errors.
- the first data sets Pok designate products whose respective parameter with regard to the degree of agreement of the sensor data corresponds sufficiently to the reference model, i.e. lies within a predetermined permissible value range.
- Second data sets Pnok denote products whose respective parameter with regard to the degree of agreement of the sensor data does not correspond sufficiently with the reference model, i.e. lies outside a predetermined permissible value range.
- the products identified as non-compliant with the gold standard, i.e. the model, are then prioritized.
- the “reconstruction error” of an auto-encoder is used in particular, which can be used as a parameter in a comparison between sensor data and a reference model.
- a subsequently arranged manipulation device 4 in the form of a robot is controlled in order to sort non-conforming products, which are selected according to a predetermined value range for product conformity, from the conveyor belt and into a new, predefined one Remove position, for example in a container intended for this purpose.
- the sorting system for repositioning defective products includes a tray 1 for providing a first product PI at a first position and at least a second product P2 at at least a second position.
- the sorting system includes at least one sensor means 2 for detecting properties and position of the first product PI and the at least one second product P2.
- the sorting system also includes an analysis device 3 with a processor and a memory and a manipulation device 4 for repositioning products.
- the manipulation device 4 can have a work area 5, which limits the predefined area in which repositioning on the work area 1, i.e. the conveyor belt, is carried out and which, together with the movement speed la of the conveyor belt, forms the basis for transfer times of repositioning.
- Fig. 2 the method according to the invention is shown as a flow chart: a) providing a first product PI at a first position and a second product P2 at a second position on a shelf 1, b) detecting properties and position of the first product PI as a first sensor Data and the second product P2 as second sensor data through a sensor means 2 and providing the first and second sensor data to an analysis device 3 with a processor and a memory, c) providing a reference model based on machine learning , which predefined properties of a product with which the first and second products Pl, P2 has the highest possible degree of agreement and is generated and trained accordingly, d) determining a first parameter with regard to the degree of agreement of the first sensor data with the reference model using an auto-encoder, e) determining at least a second parameter with regard to the degree of agreement of the at least second sensor data with the reference model using the auto-encoder, f) determining whether both the first parameter and the at least one second parameter lie outside a predetermined range, and if
- the reference model in step c) can be created and provided by a distributed client-server system using federated learning.
- Fig. 3 shows an exemplary embodiment of a client-server system according to the invention, comprising a sorting system according to the invention, which is located as a client at an edge.
- the client-server system creates and trains a global model GM based on machine learning using federated learning.
- the global model GM is provided to connected sorting systems Kl, K2 as the respective local model LM1, LM2 for use in the method according to the invention.
Landscapes
- Sorting Of Articles (AREA)
Abstract
Description
Claims
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/EP2022/060029 WO2023198290A1 (de) | 2022-04-14 | 2022-04-14 | Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produkten |
| PCT/EP2023/059428 WO2023198700A1 (de) | 2022-04-14 | 2023-04-11 | Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produkten |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| EP4489924A1 true EP4489924A1 (de) | 2025-01-15 |
Family
ID=81648684
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| EP23719669.6A Pending EP4489924A1 (de) | 2022-04-14 | 2023-04-11 | Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produkten |
Country Status (2)
| Country | Link |
|---|---|
| EP (1) | EP4489924A1 (de) |
| WO (2) | WO2023198290A1 (de) |
-
2022
- 2022-04-14 WO PCT/EP2022/060029 patent/WO2023198290A1/de not_active Ceased
-
2023
- 2023-04-11 WO PCT/EP2023/059428 patent/WO2023198700A1/de not_active Ceased
- 2023-04-11 EP EP23719669.6A patent/EP4489924A1/de active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023198700A1 (de) | 2023-10-19 |
| WO2023198290A1 (de) | 2023-10-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023194009A1 (de) | Verfahren zur prädiktion von trajektorien von objekten | |
| DE102016100134B4 (de) | Verfahren und Vorrichtung zum Untersuchen eines Objekts unter Verwendung von maschinellem Sehen | |
| DE69714534T2 (de) | Verfahren zur Handhabung von Blechteilen in einem Arbeitsraum mit einer Werkzeugmaschine und einem Roboter | |
| DE102020207613A1 (de) | Verfahren zur Bewertung einer Schnittkante eines Körpers | |
| EP4489924A1 (de) | Computer-implementiertes verfahren und systeme zur umpositionierung von fehlerhaften produkten | |
| EP3576013A1 (de) | Abschätzen eines verlaufs eines schienenpfads | |
| EP3557490A1 (de) | Trainingsverfahren | |
| DE102015012344A1 (de) | Verfahren zum Kalibrieren einer Kamera | |
| DE102018110063A1 (de) | Industrielle fertigungsstätte und verfahren zur automatisierten verbuchung von manuellen tätigkeiten | |
| DE102023004171A1 (de) | Verfahren zur automatischen Zuordnung von Labels und/oder Objekten zu Sensordaten einer Fahrzeugumfeld-Sensorik | |
| EP4296970A1 (de) | Computerimplementiertes verfahren und system zum erzeugen einer virtuellen fahrzeugumgebung | |
| DE102020208765A1 (de) | Bildklassifikator mit variablen rezeptiven Feldern in Faltungsschichten | |
| DE10013137B4 (de) | Verfahren zur bildgesteuerten Prüfung und Bearbeitung von Produkten | |
| DE102025104541B3 (de) | Computerimplementiertes Verfahren zur Erzeugung von 3D-Punkten im Raum | |
| DE102024122058B3 (de) | Verfahren und Vorrichtung zum Erkennen zumindest einer Instanz eines Objektes während eines Arbeitsablaufs in einer Arbeitsumgebung und System zum Steuern eines Arbeitsablaufs in einer Arbeitsumgebung | |
| DE3332508C2 (de) | ||
| DE102011110132B4 (de) | Verfahren zur automatischen Identifizierung fehlender oder nicht gelesener Ladungsgüter einer Ladeinheit | |
| DE4136830C2 (de) | Datenverarbeitungseinrichtung und Verfahren zur Mustererkennung | |
| DE102022209409A1 (de) | Anker zur Linienerkennung | |
| WO2023227201A1 (de) | Computer-implementiertes verfahren und system zur steuerung der herstellung eines produkts | |
| EP4372618A1 (de) | Vorrichtung und verfahren zum klassifizieren von objekten mittels maschinellem lernen | |
| DE102024201292A1 (de) | Verfahren und Vorrichtung zum Trainieren eines maschinellen Lernmodells | |
| DE102023210661A1 (de) | Verfahren zum Bereitstellen von gelabelten Bilddaten zur automatischen optischen Inspektion | |
| DE102024209264A1 (de) | Konvolutionales neuronales Netzwerk und Verfahren zum Betreiben des konvolutionalen neuronalen Netzwerks | |
| DE102022204882A1 (de) | Verfahren und Vorrichtung zum Bereitstellen von Betriebsdaten zum Betreiben mindestens einer Assistenzeinrichtung eines Fahrzeugs |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: UNKNOWN |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE |
|
| PUAI | Public reference made under article 153(3) epc to a published international application that has entered the european phase |
Free format text: ORIGINAL CODE: 0009012 |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE |
|
| 17P | Request for examination filed |
Effective date: 20241007 |
|
| AK | Designated contracting states |
Kind code of ref document: A1 Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC ME MK MT NL NO PL PT RO RS SE SI SK SM TR |
|
| STAA | Information on the status of an ep patent application or granted ep patent |
Free format text: STATUS: EXAMINATION IS IN PROGRESS |
|
| 17Q | First examination report despatched |
Effective date: 20250618 |
|
| DAV | Request for validation of the european patent (deleted) | ||
| DAX | Request for extension of the european patent (deleted) |