WO2024052264A1

WO2024052264A1 - Examination of components for faults by means of a component analyzer

Info

Publication number: WO2024052264A1
Application number: PCT/EP2023/074145
Authority: WO
Inventors: Tobias Masiak; Georg Schneider; Patrick Trampert; Felix Schmidt
Original assignee: Zf Friedrichshafen Ag
Priority date: 2022-09-06
Filing date: 2023-09-04
Publication date: 2024-03-14
Also published as: DE102022209247A1

Abstract

The invention relates to a method for examining components for faults by means of a component analyzer, the method comprising an AI routine (K1) for object analysis and an AI routine (K2) for recognizing and classifying defects which are known to the component analyzer from data sets, and/or an AI routine (K3) for recognizing anomalies that are not known to the component analyzer from data sets, wherein the AI routines are supplied with sensor data or processed sensor data regarding a component to be examined, as input data, and generate, as output data, annotated data regarding the component to be examined, containing a result of the routine in question, and wherein at least one AI routine is trained using annotated output data from at least one of the other AI routines.

Description

Examining components for defects using a component analyzer

FIELD OF THE INVENTION

The present invention relates to a method for examining components for defects using a component analyzer and a component analyzer.

SUMMARY OF THE INVENTION

Accordingly it is provided:

- a method for examining components for defects using a component analyzer, the method comprising a Kl routine for object analysis and a Kl routine for detecting and classifying defects that are known to the component analyzer from data sets and/or a Kl routine for Detection of anomalies that are not known to the component analyzer from data sets, the Kl routines being fed with sensor data or processed sensor data relating to a component to be examined as input data and annotated data relating to the component to be examined as output data containing a result of the respective routine generate, wherein at least one Kl routine is trained with annotated output data of at least one of the other Kl routines; as well as

- a component analyzer for examining components for defects with a holder in which the component is positioned during the examination; at least one sensor, in particular an optical sensor, to generate sensor data relating to a component to be examined; a control device to control the sensor such that it automatically generates the sensor data; an interface to a Kl module with at least one Kl routine for object analysis, at least one Kl routine for detecting and classifying defects that are known to the Kl module from data sets, and at least one Kl routine for detecting anomalies are not known to the Kl module from data sets, whereby the Kl routines of the Kl module Modules interact with each other; and with means for sorting examined components according to their examination results.

A sensor, also known as a detector, (measurement or measuring) transducer or (measuring) probe, is a technical component that detects certain physical, chemical properties or states, e.g. B. temperature, humidity, pressure, speed, brightness, acceleration, pH value, ionic strength, electrochemical potential and / or the material nature of its environment can be recorded qualitatively or quantitatively as a measurement variable. These variables are recorded using physical or chemical effects and converted as sensor data into an electrical signal that can be further processed. Sensors include camera, lidar, radar, TOF and other sensor technologies, such as acoustic sensors.

Computer program products typically include a sequence of instructions that, when the program is loaded, causes the hardware to perform a specific procedure that results in a specific result.

Machine learning or artificial intelligence (Kl) is a generic term for the “artificial” generation of knowledge from experience: an artificial system learns from examples and can generalize them after the learning phase has ended. To do this, machine learning algorithms build a model that is based on training data. This means that the examples are not simply memorized, but patterns and regularities are recognized in the learning data. The system can also assess unknown data (learning transfer) or fail to learn unknown data (overfitting).

In artificial intelligence or machine learning, the type and power of knowledge representation play an important role. A distinction is made between symbolic approaches, in which the knowledge - both the examples and the induced rules - is explicitly represented, and non-symbolic approaches, such as neural networks, which are "trained" to behave in a predictable manner, but which do not provide any insight into it allow learned solutions; Knowledge is implicitly represented here. A Kl routine is a computer program section that is implemented using artificial intelligence.

Object analysis refers to determining the properties of a component. This includes, for example, object classification, i.e. the division of components into different groups/classes according to their type and/or properties, the measurement of an object, or object recognition, i.e. the classification of components according to their type and/or properties.

If defects are known to a component analyzer or a Kl routine, it can be assumed that the Kl routine has already processed sensor data regarding comparable defects in the past, or that the Kl routine has already been trained for this.

Training manifests itself in a quantitative assessment of the distribution of training data. An AI uses the training data to learn what distribution of sensor data is to be expected in the future.

Anomalies are defects that are not known to a Kl routine. These can, for example, be defects that occur rarely or are caused by a change in the manufacturing process. Anomaly detection can be achieved using statistical methods in conjunction with artificial intelligence by first determining a distribution of sensor data. If the distribution deviates from the expected distribution by more than a predetermined standard deviation factor, it may be useful to detect an anomaly.

As part of anomaly detection, the distribution of expected training data, e.g. regarding error-free components or components with known errors, can be determined as a reference and/or patterns are determined in the data based on historical data, which can be classified at least into the normal classes through appropriate annotations and optionally classified as abnormal. One way to train a Kl is error feedback. In the early training phase, a class makes mistakes. To correct the error, the triggers of the deviations (errors) between the generated assignment (actual output data) and the solution assignment (target output data) are traced back and controlling factors are changed in such a way that the assignment errors become smaller.

In this application, training data is data pairs consisting of input data that is to be processed by the KL and target output data that is to be determined by the KL. During training, the Kl is adjusted based on a comparison of target output data with the actual output data determined by the Kl, which results in a training effect.

An examination result can, for example, be the classification of a component into quality classes, such as good/not good, or a probability for the quality of a component. A result space can contain any number of classes

The components of a component analyzer include hardware and/or software modules, comprising hardware and/or software modules for regulating and/or controlling industrial manufacturing processes. The hardware modules include electronic units, integrated circuits, embedded systems, microcontrollers, multiprocessor systems-on-chip, central processors and/or hardware accelerators, for example graphics processors, data storage units and connectivity elements, for example WLAN modules, RFID modules, Bluetooth modules, NFC modules. According to one aspect of the invention, the Kl modules are executed as functional software in the cloud infrastructure.

The commands of the computer programs according to the invention include machine instructions, source text or object code written in assembly language, an object-oriented programming language, for example C++, or in a procedural programming language, for example C. According to one aspect of the invention, the computer programs are hardware-independent application programs that, for example, via a data carrier or a data carrier signal is provided using Software Over The Air technology. The basic idea of the invention is a component analyzer or a method for examining components using several Kl routines, each of which is specialized for individual special tasks. There are at least one Kl routine for object analysis, at least one Kl routine for detecting and classifying defects that are known to the component analyzer from data sets, and / or at least one Kl routine for detecting anomalies that are not known to the component analyzer from data sets are.

The basic idea of the invention is also the interaction between different Kl modules. Accordingly, the invention comprises at least two Kl modules

It is intended that the Kl routines interact with each other, which means that the Kl routines learn from each other. For example, it can be provided that if the Kl routine detects a previously unknown anomaly, this deviation will also be recognized in the future by a Kl routine for detecting and classifying defects. Furthermore, it is provided that the Kl routines can process output data from other Kl routines. For example, it can be provided that a result of a Kl routine for object analysis is further processed by a Kl routine for detecting and classifying defects.

Furthermore, it is provided that the Kl routines are fed with sensor data or processed sensor data relating to a component to be examined and generate annotated output data for these sensor data with conclusions that the respective Kl routine has determined.

Advantageous refinements and further developments result from the further subclaims and from the description with reference to the figures in the drawing.

According to a preferred development of the invention, the annotated data is at least partially checked by a human editor. For this purpose, the component analyzer has a human-machine interface. It can be provided that a further Kl routine is provided for determining the reliability of output data and that the annotated data is presented to a human processor for checking, provided that the reliability determined for the annotated output data falls below a predetermined threshold value.

Alternatively, it can be provided that the output data of the Kl routine for detecting anomalies is generally checked by a human processor, provided that the output data contains recognized anomalies. This means that newly occurring defects can be checked and production coordinators can be informed about new defects that have arisen. The human editor can also decide whether other Kl routines are trained with the respective annotated source data.

According to a preferred development of the invention, a first examination result is generated by means of at least a first Kl routine regarding the quality of a component, while the component analyzer has access to the component and the component is prepared by the component analyzer for sorting according to quality classes or is sorted according to quality class.

This means that the component analyzer decides on the quality of a component within the clock frequency. The component analyzer can therefore also ensure or at least prepare for the sorting out of rejects.

It is also expedient if a second examination result is generated using at least a second Kl routine regarding the quality of a component if the component was sorted into a positive quality class based on the first examination result, the second examination result being generated while the component analyzer does not has access to the component.

This makes it possible to carry out further examinations of the sensor data after a component is no longer in the component analyzer. So you can ensure that the clock frequency is not slowed down to wait for the results of the Kl routines. Instead, it can be planned to carry out examinations that require a longer period of time while a component is being transported or further processed.

It is also expedient if a batch containing components that were sorted into a positive quality class based on the first test result is retrieved if a second test result corresponding to a negative quality class was generated for at least one component of the batch.

Provision can be made to discard the entire batch or to examine which component of the batch generated a negative second test result.

According to a preferred development of the invention, optical sensor data relating to a component to be examined are transformed to a predetermined size and at least one Kl routine is fed with the transformed sensor data.

Image data with a fixed size is often easier to process by a Kl routine. In the past, such changes to image data to a fixed size were often not considered, as this often results in the loss of interpretability for human editors, as a transformation to a fixed size causes distortions, for example. However, experiments have shown that Kl routines can deal with distortions and distortions do not reduce the reliability of the results of a Kl routine.

According to a preferred development of the invention, the Kl routine for object analysis is fed with optical sensor data and the fed Kl routine generates image sections relating to relevant features of a component as output data. It is intended that the output data (annotated image sections) be annotated with annotations by the Kl routine for object analysis.

The annotations can, for example, contain information about which component it is. Kl routines, which are downstream in the examination process of the Kl routines for object analysis, can use this result and, for example, assess the relevance of abnormalities in sensor data based on the object analysis.

It is also useful if at least one Kl routine for detecting and classifying defects and/or at least one Kl routine for detecting anomalies is fed with annotated image sections.

This can also ensure that a component analyzer examines a large number of components that are different in nature, without the need for measures to change the component analyzer.

According to a preferred development of the invention, at least two, in particular a plurality, Kl routines for detecting and classifying defects are provided, each of the Kl routines for detecting and classifying defects of a predetermined type and/or specialized in a predetermined location of defects is. For example, one Kl routine can be specialized for the detection of scratches and another Kl routine for the detection of breakage damage. Alternatively or additionally, another Kl routine may inspect a coating, a weld, or other local features in a specific local area of the component.

A computer program product according to a method of an embodiment of the invention carries out the steps of a method according to the preceding description when the computer program product runs on a computer. When the program in question is used on a computer, the computer program product causes an effect, namely the interaction of several Kl routines, which improves the reliability of the detection of rejects in a component analyzer.

CONTENTS OF THE DRAWINGS The present invention is explained in more detail below using the exemplary embodiments shown in the schematic figures of the drawings. It shows:

Figure 1 is a schematic block diagram of an embodiment of the invention;

Figure 2 shows a schematic principle sketch of an embodiment of the invention.

The accompanying drawings are intended to provide further understanding of embodiments of the invention. They illustrate embodiments and, in connection with the description, serve to explain principles and concepts of the invention. Other embodiments and many of the advantages mentioned arise with regard to the drawings. The elements of the drawings are not necessarily shown to scale to one another.

In the figures of the drawings, identical, functionally identical and identically effective elements, features and components are each provided with the same reference numerals, unless stated otherwise.

DESCRIPTION OF EMBODIMENTS

Figure 1 shows a schematic block diagram according to a method for examining components for errors using a component analyzer. The method includes a Kl routine for object analysis K1, as well as a Kl routine for detecting and classifying defects K2 and a Kl routine for detecting anomalies K3. The Kl routines K2 and K3 are downstream of the Kl routine K1.

The Kl module with the Kl routines K1, K2, K3 is fed with sensor data S relating to a component to be examined and from this generates output data containing annotated data relating to the component to be examined. Furthermore, the Kl routines K2 and K3 are fed with the annotated output data of the Kl routine K1.

It is also provided that at least one Kl routine is trained with annotated output data of at least one other Kl routine. The training can take place during ongoing operations or on an event-related basis.

Figure 2 shows a schematic block diagram of a component analyzer for examining components for errors. The component analyzer comprises a receptacle 1 in which the component is positioned during the examination, at least one sensor 2 to generate sensor data relating to a component to be examined, a control device 3 to control the sensor in such a way that it automatically generates the sensor data, an interface 4 to a Kl module 5 with at least one Kl routine K1 for object analysis, at least one Kl routine K2 for detecting and classifying defects, and at least one Kl routine K3 for detecting anomalies. Furthermore, the component analyzer has means 6 for sorting examined components according to their examination results.

reference symbol

K1 Kl routine

K2 Kl routine

K3 Kl routine

S sensor data

A Initial data

1 shot

2 sensors

3 control unit

4 interface

5 Kl module

6 means

10 component analyzer

Claims

Patent claims

1 . Method for examining components for defects using a component analyzer, the method comprising a Kl routine (K1) for object analysis and a Kl routine (K2) for detecting and classifying defects that are known to the component analyzer from data sets and/or a Kl -Routine (K3) for detecting anomalies that are not known to the component analyzer from data sets, the Kl routines being fed with sensor data or processed sensor data relating to a component to be examined as input data and annotated data relating to the component to be examined as output data containing a result of the respective routine, wherein at least one Kl routine is trained with annotated output data of at least one of the other Kl routines.

2. The method according to claim 1, wherein the annotated data is at least partially checked by a human editor.

3. Method according to one of the preceding claims, wherein a first examination result is obtained by means of at least a first

Kl routine regarding the quality of a component is generated while the component analyzer has access to the component, and the component is prepared by the component analyzer for sorting according to quality classes or is sorted according to quality classes.

4. The method according to claim 3, wherein a second examination result is determined by means of at least a second classification Routine regarding the quality of a component is generated when the component has been sorted into a positive quality class based on the first examination result, with the second examination result being generated while the component analyzer has no access to the component.

5. The method according to claim 4, wherein a batch containing components that were sorted into a positive quality class based on the first test result is retrieved if a second test result corresponding to a negative quality class was generated for at least one component of the batch.

6. The method according to any one of the preceding claims, wherein optical sensor data relating to a component to be examined are transformed to a predetermined size, and at least one Kl routine is fed with transformed sensor data.

7. The method according to any one of the preceding claims, wherein the Kl routine for object analysis is fed with optical sensor data and image sections relating to relevant features of a component are generated as output data, the output data, hereinafter referred to as annotated image sections, from the Kl routine for object analysis be annotated with annotations.

8. The method according to claim 7, wherein at least one Kl routine for detecting and classifying defects and / or at least one Kl routine for detecting anomalies are fed with annotated image sections.

9. The method according to any one of the preceding claims, wherein at least two, in particular a plurality, Kl routines are provided for the detection and classification of defects, each of the Kl routines for the detection and classification of defects of a predetermined type, predetermined cause and / or specializes in a predetermined location of defects.

10. Component analyzer (10) for examining components for errors

- a receptacle (1) in which the component is positioned during the examination;

- at least one sensor (2), in particular an optical sensor, in order to generate sensor data relating to a component to be examined;

- a control device (3) to control the sensor in such a way that it automatically generates the sensor data;

- an interface (4) to a Kl module (5) with at least one Kl routine (K1) for object analysis, at least one Kl routine (K2) for detecting and classifying defects that are known to the Kl module from data sets , and at least one Kl routine (K3) for detecting anomalies that are not known to the Kl module from data sets, the Kl routines of the Kl module interacting with one another; and with

- Means (6) for sorting examined components according to their examination results.

11. Computer program product with program code means to carry out the method according to one of claims 1 -9.