阅读说明：本技术 用于识别构件的损伤的方法和系统 (Method and system for identifying damage to a component ) 是由 D·沃夫 J·切法 J·阿帕奥 于 2018-11-20 设计创作，主要内容包括：本发明涉及一种用于识别构件的损伤的方法,该方法具有：提供(I)参考状态下构件的区段的第一记录内容；获取(II)正常使用之后构件的区段的后续记录内容；以及将后续记录内容与第一记录内容进行比较(III),以确定构件是否受损伤。在确定构件损伤的情况下,该方法进一步包括：将后续记录内容与损伤数据库的数据进行对比(IV),以确定损伤的类型；以及输出(V)所确定的损伤类型。(The invention relates to a method for identifying damage to a component, comprising: providing (I) a first recorded content of a section of the member in a reference state; obtaining (II) subsequent recorded content of a section of the component after normal use; and comparing (III) the subsequent recorded content with the first recorded content to determine whether the component is damaged. In the case where damage to the component is determined, the method further comprises: comparing (IV) the subsequent recorded content with data of a damage database to determine the type of damage; and outputting (V) the determined type of injury.)

1. A method for identifying a damage of a component (10), the method having:

providing (I) a first recorded content (14.0) of a section of the component (10) in a reference state;

acquiring (II) subsequent recorded content (14.1; 14.2; 14.3) of the section of the component (10) after normal use; and

comparing (III) the subsequent recorded content (14.1; 14.2; 14.3) with the first recorded content (14.0) to determine whether the component (10) is damaged;

wherein in case of a determination of a damage of the component (10), the method further comprises:

comparing (IV) the subsequent recorded content (14.1; 14.2; 14.3) with data of a damage database to determine a damage type; and

outputting (V) the determined type of injury.

2. Method according to claim 1, characterized in that the data of the first recorded content (14.0), the subsequent recorded content (14.1; 14.2; 14.3) and the damage database are photo-type recorded content.

3. Method according to one of the preceding claims, characterized in that the comparison of the first recorded content (14.0) with the subsequent recorded content (14.1; 14.2; 14.3) is performed by means of a correlation analysis.

4. Method according to one of the preceding claims, characterized in that the comparison of the subsequently recorded content (14.1; 14.2; 14.3) with the data of the damage database comprises a feature detection and matching method.

5. The method of claim 4, further comprising:

determining at least one region of interest in the subsequent recorded content (14.3) and at least one region of interest in the image data of the damage database; and

the determined regions of interest are compared by means of the feature detection and matching method.

6. Method according to one of the preceding claims, characterized in that the comparison of the subsequently recorded content (14.1; 14.2; 14.3) with the data of the damage database is embedded in a machine learning Method (ML) and the method further comprises:

(VI) evaluating and correcting, by the user, the outputted damage type; and

storing the correction to train the machine learning method.

7. Method according to one of the preceding claims, characterized in that in the case of a damage determination and/or a damage type determination, a feedback to a subsequent run of the component (10) is automatically prompted.

8. A device (1) for identifying a damage of a component (10), the device having:

a first interface for providing a first recorded content (14.0) of a section of the component (10) in a reference state;

a second interface for providing subsequent recorded content (14.1; 14.2; 14.3) of the section of the component (10) after normal use;

a third interface for providing a damage database with data of various damages;

a fourth interface for outputting the determined type of impairment;

a computer device configured to: comparing the first recorded content (14.0) with the subsequent recorded content (14.1; 14.2; 14.3) in order to determine whether the component (10) is damaged based on the comparison; in the event of a determination of damage to the component (10), comparing the subsequent recorded content (14.1; 14.2; 14.3) with data of the damage database to determine the type of damage; and outputting the determined damage type by the fourth interface.

9. The device (1) according to claim 8, the device further having:

a reference memory (3) for storing a first recorded content (14.0) of a section of the component (10), the reference memory being connected to the first interface;

a recorded content memory (4) for storing subsequent recorded content (14.1; 14.2; 14.3) of the section of the component (10) after normal use, the recorded content memory being connected to the second interface; and

a database memory (8) for storing the damage database, the database memory being connected to the third interface.

10. The device (1) according to claim 8 or 9, the device further having:

an object detection device (2) for acquiring subsequent recorded content (14.1; 14.2; 14.3) of a section of the component (10) after normal use;

wherein the computer device is configured to store subsequent recorded content (14.1; 14.2; 14.3) acquired by means of the object detection device (2) on the recorded content memory (4).

11. Device (1) according to claim 10, characterized in that the data of the first recorded content (14.0), the subsequent recorded content (14.1; 14.2; 14.3) and the damage database are photo-type recorded content and the object detection means (2) are a camera (2) producing an image.

12. Device (1) according to one of the claims 8 to 11, wherein the computer means have comparing means (5) for comparing the first recorded content (14.0) with the subsequent recorded content (14.1; 14.2; 14.3) and separate comparing means (9) for comparing the subsequent recorded content (14.1; 14.2; 14.3) with the data of the damage database.

13. The apparatus according to claims 10 and 12 or according to claims 11 and 12, characterized in that the reference memory (3), the recorded content memory (4) and the comparison means (5) are arranged in the object detection means (2).

14. The device (1) according to claims 10 and 12, according to claims 11 and 12, or according to claim 13, the device further having:

a server (6) and a network (7) configured to enable the object detection device (2) and the server (7) to communicate with each other via the network (7);

wherein the database memory (8) and the comparison means (9) are arranged on the server (6).

15. A transmission having a component (10) and a device (1) according to one of claims 8 to 14, wherein the device is configured for identifying a damage of the component (10).

Technical Field

The invention relates to a method and a system for identifying damage to a component. The invention comprises comparing the recorded content of the component in the reference state with the recorded content of the component after normal use.

Background

Inspection of components is currently typically performed on site by an expert. This is particularly time consuming and expensive in the case of larger stationary facilities (e.g., wind turbines) or larger mobile machinery (e.g., boats, excavators, dumpers, etc.). Furthermore, if the damage occurs rapidly and occurs between two inspection intervals, then higher subsequent costs may result.

EP 2921840 a1 relates to a test unit and a method for quantitative analysis of the wear pattern of a gear, wherein images of the gear at different points in time are compared with each other.

Disclosure of Invention

A method for identifying damage to a component includes: a first recorded content of a section of the member in a reference state is provided. The component may be a gear, a bearing, a turbine blade, a shift fork, an actuator, a cylinder, a motion screw or the like of the transmission. For example, the component is part of a wind power plant, an aircraft, a passenger car (PKW) or other technical installation. The components in the technical device have, in particular, technical functions which can lead to damage to the components during operation of the device. The damage to the component may be material fatigue (i.e. a slowly progressive damage process) and/or violent damage (i.e. damage due to short static loads) and/or other forms of damage. The section of the component can, for example, relate to a surface for force transmission.

A group of a plurality of members (e.g., a gear and an adjoining housing) may be imaged in the first recorded content. Alternatively, only a single component or only a section of a single component may be imaged in the first recorded content.

The reference condition may relate to a condition of the component prior to delivery, prior to installation and/or prior to normal commissioning. The component can be in particular intact or have damage within a tolerable degree in the reference state. In particular, it is conceivable to produce the first recorded content in the reference state in an inline test (EOL) and therefore immediately after the production and assembly of the component. However, the first recorded content in the reference state can also be generated in the area of commissioning (for example after the installation of the wind energy installation has been completed).

The method further comprises the following steps: subsequent recorded content of the section of the component after normal use is acquired. In particular, the subsequent recorded content can be generated at a point in time when the component is to be checked for damage, for example at a fixed check point in time. Furthermore, subsequent recordings of the component may be repeatedly captured periodically and/or acquired when the recording is compatible with the operational flow. Normal use of a component is understood to be for the function for which the component is constructed and/or designed.

Further, the method comprises: the subsequent recorded content is compared to the first recorded content to determine whether the component is damaged. When comparing the recorded contents, differences and/or commonalities of the recorded contents can be searched. The comparison may be computer-supported, i.e. performed by an electronic data processing means, such as a microcontroller. For comparison, the recorded content to be compared may be in digital form. In the determination, it can be checked whether these recorded contents exceed a defined differentiation threshold and therefore whether the component is considered damaged. This threshold may be changeable.

If a damage is determined when comparing the first recorded content with the subsequent recorded content, these subsequent recorded contents are compared with the data of the damage database in a next step in order to derive the damage type. The damage database can be categorized according to the damage category and have any number of different damage data, in particular damage data in different manifestations and severity. The data in the damage database relates in particular to optoelectronic data of different damage types. For example, the damage database may have images, such as images of Pitting (Pitting), tooth breakage, greying (micropitting), erosion and/or wear of the gear, in particular in the form of representations of different intensities. The impairment database may relate to a digital data set.

In addition, the impairment database may have metadata. The metadata may be: transmission type, application, component type, defect type, severity, location description, etc. Depending on the metadata, different applications (e.g. wind and maritime) can be covered in parallel in this way. Thus, the impairment database may contain application-specific and cross-application data sets.

Further, the method comprises: outputting a damage type, which has been derived from a comparison of the subsequently recorded content with data of a damage database. The output damage type is technically characterized in that by means of the output further damage or even destruction of the component and thus potential injury to surrounding persons can be prevented. The output may be in the form of a digital and/or analog report and/or by means of a display device.

The methods of the present invention may be performed non-centrally, centrally and/or decentralised.

The method of the invention can realize the damage type of the component in a simple mode with high accuracy. By means of the two-stage approach of the method, only the recorded contents of the damage of the component which has been determined beforehand are compared with the damage database. The precise preselected range of recorded contents that are subjected to a more complex and error-prone comparison method to determine the type of damage can be determined by a preceding comparison step. Thus, the two-stage approach achieves higher accuracy in identifying and diagnosing component damage.

The data of the first recorded content, the subsequent recorded content and the impairment database may relate to photo-type recorded content, in particular digital photos. Alternatively, however, the recorded content may also relate to other data which are captured by means of sensors for electromagnetic waves (for example radio waves, microwaves, radar, UV light, IR light, visible light and/or gamma rays).

The comparison of the first recorded content with the subsequent recorded content may be achieved by means of a correlation analysis. In particular, gray values and/or color values of the respective recorded contents can be correlated with one another. A simple and efficient comparison can thus be achieved with a high degree of accuracy. To improve accuracy, the correlation analysis may be performed in a lighting standardized manner. This means that the illumination intensities respectively present in the recorded contents to be compared are removed during the correlation.

Furthermore, comparing the subsequent recording with the data of the damage database may comprise a Feature-Detection and Matching method (Feature-Detection end-Matching method). By means of the method, characteristic features in the recorded content to be compared are identified (feature detection). These characteristic features may relate, for example, to regions of the recorded content in which there is a high gradient in the signal magnitude. For example, in the case of photo-type recorded content, a high gradient in color values and/or gray values may be a characteristic feature. Thus, the characteristic feature may be an edge, for example. Feature detection may be implemented by means of SURF (Speeded Up Robust Feature), MSER (maximum steady state limit region), BRISK (Binary Robust extensible key), and/or other algorithms. The respective algorithm to be used may be selected depending on the data sets to be compared, so that the algorithm achieving the highest accuracy in the respective lesion type is therefore used. Thus, for example, the SURF algorithm may be used for comparison of subsequently recorded content with a first damage type of the damage database, and the BRISK algorithm may be used for comparison of subsequently recorded content with a second damage type of the damage database. The accuracy of determining the type of lesion may be improved by providing a feature detection and matching method. In addition to this, subsequent recorded content can be compared to non-currently applied (anwendungsfremden) lesion recorded content stored in the lesion database because the feature detection and matching method is not rotation and scale invariant. Thus, for example, also lesion image parts originating from literature and other applications can be used.

The method may further comprise: at least one Region of Interest (Region of Interest) in the subsequent recording and/or at least one Region of Interest (Region of Interest) in the image data of the impairment database is determined. The regions of interest are unique in the same type or different image data content. The determined area of interest may first be checked, preferably and/or by means of other feature detection and matching methods. If, for example, a plurality of components are imaged in the data of the subsequent recording and/or damage database, a component-specific damage comparison can be carried out by determining the region of interest. If, for example, the gear and the housing are imaged on subsequent recordings, the housing can be compared only with the housing damage data of the damage database and the gear only with the gear damage data by means of the determination of the region of interest. The result is improved efficiency and high accuracy of lesion type determination.

Further, the aforementioned industrial image processing method forming an image processing method without learning ability may be embedded in the machine learning method. The machine learning methods may involve, for example, neural networks, support vector machines, Convolutional Neural Networks (CNNs), and/or other machine learning methods. The machine learning method can be trained in particular by the aforementioned industrial image processing method for identifying component damage. To this end, the method may comprise: the damage type output by the industrial image processing method is evaluated and, if necessary, the output damage type is subjected to the required correction by the user. The user may be an expert who manually checks the outputted damage type. The method may further have: the correction is stored to train the machine learning method. By inspection and correction by the user, the machine learning method can learn new associations between recorded image material and the impairment database and thus improve the accuracy of the impairment determination.

It is also possible to automatically initiate a feedback to the subsequent operation of the component in the event of a damage determination and/or a damage type determination. For example, in case of a determined damage of a component and/or a determined type of damage of a component, the power and/or the rotational speed of the installation may be reduced or the installation may be completely shut down.

Furthermore, the invention relates to a device for identifying a damage of a component. The device comprises a first interface for providing a first recorded content of a section of the member in a reference state and a second interface for providing a subsequent recorded content of the section of the member after normal use. Furthermore, the device comprises a third interface for providing a damage database with data of various damages and a fourth interface for outputting the determined damage type. The interface may relate to an input/output device which may, for example, be designed for transmitting and receiving data streams in a packetized or unpackaged form. These interfaces may be designed as separate devices. Alternatively, some or all of these interfaces may also be implemented by means of a unique device (e.g. a bus). For an understanding of other features, reference is made to the method-related embodiments above.

Further, the apparatus includes a computer device configured to compare the first recorded content with subsequent recorded content to determine whether the component is damaged based on the comparison. The computer arrangement may have various computer units, which may be arranged at different locations. Alternatively, the computer device can also have only one single computer unit. The computer means of the apparatus is further configured to: if the damage of the component is determined, comparing the subsequent recorded content with the data of the damage database to determine the damage type; and outputting the determined damage type through a fourth interface. A device of a computer device for a certain function is understood to be a specific programming of the device for the corresponding function.

The apparatus may further have: a reference memory for storing a first recording content of a section of the component, a recording content memory for storing a subsequent recording content of the section of the component after normal use, and a database memory for storing a damage database. The reference memory may be connected to the first interface, the recorded content memory may be connected to the second interface, and the database memory may be connected to the third interface. The various memories of the device may be provided on a single memory member or on different, locally separated memory members. The memory may relate to an electronic memory.

Furthermore, the device may have object detection means for acquiring subsequent recorded content of the section of the component after normal use. The computer means may be configured to store subsequent recorded content acquired by means of the object detection means on the recorded content memory. The object detection device may be portable or stationary.

The first recorded content and the subsequent recorded content may be recorded with the object detection device in the same relative positioning with respect to the section of the member. This embodiment has the advantage that the two recordings can be compared with high accuracy and efficiency, since the recordings image sections of the component with the same zoom and viewing angle. Sometimes, however, these recordings can also be generated from different perspectives.

The object detection means may relate to a camera that produces an image. The object detection device can be designed, for example, as a camera device, with the aid of which a single image or a short series of single images can be recorded. Alternatively or additionally, the camera producing the images may also be implemented as a motion image camera, e.g. a video camera, to capture a continuous sequence of images. Instead of the embodiments of the object detection device described above, other image-generating sensors for detecting electromagnetic waves are also conceivable, for example sensors for detecting radio waves, microwaves, radar, UV light, IR light, visible light and/or gamma rays.

The computer means may have comparing means for comparing the first recorded content with the subsequent recorded content and separate comparing means for comparing the subsequent recorded content with the data of the damage database. The comparison means of this embodiment is separate, i.e. provided as a separate unit from the comparison means. However, the two units may be networked to each other.

Further, the reference memory, the recorded content memory, and the comparison means may be provided in the object detection means. It is thereby achieved that a comparison of the subsequently recorded content with the recorded content in the reference state is performed at the position of the component. Thus, the comparison can be performed without a network for data transmission, which in turn enables particularly rapid comparison operations.

The apparatus may further have a server and a network by which the object detection device and the server can communicate with each other. The server may relate to a computer that may communicate with different clients, such as object detection devices. Various object detection devices may each be designed to identify component damage, for example, in accordance with embodiments of the present invention. The server may be a central unit and, for example, have or provide ECUs and/or cloud services. The server may be separated from the object detection means on the spot, e.g. arranged outside the calling, listening and/or speaking range. The network may relate to a wireless and/or wired network. The object detection device and the server may communicate with each other via, for example, a Global Area Network (GAN), such as the internet. Alternatively or additionally, it is conceivable that the object detection device communicates with the server via a LAN, WLAN, bluetooth or the like.

Furthermore, the database memory and the comparing means may be provided on the server. It is thereby possible to provide the damage database centrally on a server for a large number of clients, for example object detection devices. Centralized maintenance of the damage database may also be achieved. The potentially costly comparison operation can thus likewise be performed by a centralized, high-performance computer for a large number of clients.

The device may have a user interface for the purpose of maintaining a database of injuries on the database memory by the user. In particular, (data) maintenance is understood to mean: updating, supplementing and optimizing data that damages the database or deleting unusable data sets. The new damage image may be added to the database, for example, via a user interface. This may be achieved by means of software modules or programming interfaces, such as Web services and/or clients SW. Furthermore, a function can be provided in the user interface which feeds back to the user which images/image properties (e.g. size, resolution, brightness and similar characteristics) have led to a successful lesion identification, in particular a lesion identification specific to the type of defect. Therefore, the quality of the image data in the damage database can be improved. In addition, automation extensions are provided which automatically record images resulting from the operation of the method, which have been confirmed by experts or are located above a certain reliability threshold (confidence level) as a result of the method, for example, into a damage database.

Furthermore, the invention relates to a transmission having a component and a device according to one of the preceding embodiments, which device is configured for identifying a damage of the component. The transmission may be a transmission of a wind power plant.

Drawings

FIG. 1 illustrates a system for identifying a component damage according to one embodiment of the present invention.

FIG. 2 shows a flow chart for describing a method of identifying a damage to a component according to one embodiment of the invention.

FIG. 3 illustrates exemplary results of a correlation analysis performed within the scope of the embodiments presented in FIG. 2.

Fig. 4 shows exemplary results of feature detection and matching analysis performed within the scope of the embodiments presented in fig. 2.

FIG. 5 illustrates a block diagram of a machine learning method that may be trained by the method illustrated in FIG. 2, according to an embodiment of the present invention.

Detailed Description

Embodiments of the present invention are described below with reference to the drawings.

Fig. 1 shows an embodiment of a device 1 for identifying damage to a component 10 according to the invention. The system 1 has a camera type camera 2 and a server 6. The camera 2 and the server 6 are connected to each other via a network 7, so that communication between the camera 2 and the server 6 can be realized. In the present embodiment, the network 7 is the internet.

The apparatus 1 of the present embodiment further comprises a member 10. In the present exemplary embodiment, component 10 relates to a gear, more specifically to a spur gear of a wind power plant.

The camera 2 of the device 1 is arranged such that a section of the component 10, from which a damage of the component 10 can be seen, can be detected with the camera. Furthermore, embodiments are also conceivable in which the cameras 2 are arranged in a movable manner, in particular in a pivotable manner, so that different components 10 or different sections of one component 10 can be detected with only one camera 2. In a refinement, the camera 2 may have a cleaning device in order to remove dirt due to lubricant and wear, so that high-quality recorded content can be produced.

The camera 2 has a reference memory 3 for storing a first recorded content 14.0 of a section of the component 10 in a reference state. The camera 2 further comprises a recorded content memory 4 for storing subsequent recorded content 14.1, 14.2, 14.3 of the section of the member 10 after normal use. Furthermore, the camera 2 comprises a comparison device 5 with which the recorded content 14.0 in the reference state can be compared with the subsequent recorded content 14.1, 14.2, 14.3 by means of a correlation analysis and it can be determined whether the component 10 is damaged. The comparison device 5 relates to a correspondingly programmed controller or ECU having a microcontroller.

Furthermore, the device 1 comprises a database memory 8 for storing a lesion database with image data of various lesions. Similarly, the apparatus 1 comprises a contrast device 9 configured to: in the case of a lesion determined by the comparison means 5, the subsequently recorded content 14.1, 14.2, 14.3 is compared with the image data of the database memory 8 in order to determine the type of lesion and the severity of the lesion.

The device further comprises a user interface 11 for maintaining a database of damages. With regard to the design of this user interface 11 and the impairment database, reference is made to the above embodiments.

In this embodiment, the reference memory 3, the recorded content memory 4, and the comparison device 5 are provided in the camera 2. The database memory 8, the comparing means 9 and the user interface 11 are arranged in the server 6. However, the present embodiment is not limited to this design. Alternatively, it is also possible for the reference memory 3, the recorded content memory 4, the comparison device 5, the database memory 8, the comparison device 9 and the user interface 11 to be provided in the camera 2. Any other distribution of the individual components between the camera 2 and the server 6 is also conceivable. Furthermore, it is also possible to provide the individual components redundantly, in particular both on the camera 2 and on the server 6. For example, the reference memory 3 and the recorded content memory 4 may be provided on the camera 2, and the additional reference memory and the additional recorded content memory may be provided on the server 6. This is meaningful for example for archiving purposes.

The following describes the procedure of the method for identifying damage to a component 10 according to an embodiment of the invention with the aid of the flow chart shown in fig. 2. Embodiments of the method described hereinafter are described with reference to the system 1 shown in fig. 1. However, the method according to this embodiment may also be performed on other systems.

In a first step I of the method according to the present exemplary embodiment, a first recorded content 14.0 of the component 10 is acquired with the camera 2 in a different operating position in the reference state. The acquired recorded contents are stored in the reference memory 3 visible in fig. 1 together with the respectively associated operating position of the component 10. Alternatively or additionally, it is conceivable that the recorded content of the component 10 in the reference state is not acquired by the camera 2, but rather is provided on the reference memory 3 by the manufacturer of the component 10 or in another way.

In a second step II, after normal use of the component 10, the subsequent recorded content 14.1, 14.2, 14.3 of the component 10 in a certain operating position is acquired with the camera 2 and stored on the recorded content memory 4.

In a next step III, the subsequent recorded contents 14.1, 14.2, 14.3 of the component 10 are compared by the comparison means 5 with the first recorded content 14.0 in the reference state which was recorded at the same operating position. In this embodiment, the comparison is carried out by means of an illumination-normalized correlation analysis of the gray values and/or color values.

Fig. 3 shows a comparison of the first recorded content 14.0 of the component 10 in the reference state, recorded at time t0, with the subsequent recorded content 14.1, 14.2, 14.3 of the component 10, recorded at the same location and recorded after normal use of the component 10 at times t1, t2 and t 3. If the correlation between the recorded contents is determined to be below the threshold, the component 10 is damaged. For example, if the correlation is below 99.9%, 95%, 90%, or 82%, then a lesion may be determined.

At a time point t0, these recorded contents coincide, so that the correlation is 100%. Over time, the correlation between the first recorded content and the subsequent recorded content 14.1, 14.2, 14.3 decreases. Therefore, the correlation between the subsequent recorded content 14.3 recorded at the time point t3 and the reference recorded content 14.0 is smaller than the correlation between the reference recorded content 14.0 and the subsequent recorded content 14.1 captured at the time point t 1. This is because, in the example shown, the degree of component damage increases over time, which is reflected in a greater deviation from the recorded content to be compared. This greater deviation in turn leads to a lower relevance of the recorded content.

If the damage of the component 10 is determined by comparing the subsequent recorded content 14.1, 14.2, 14.3 with the recorded content 14.0 in the reference state in step III, the subsequent recorded content 14.1, 14.2, 14.3 in the recorded content memory 4 is compared with the data of the aforementioned damage database on the database memory 8 in a next step IV using the comparison means 9. This comparison is used to determine the type of injury. If in the comparison step III it is determined that the component 10 is not damaged, steps II and III are re-executed at a later point in time until damage is determined.

In this embodiment, the comparison step IV for determining the type of lesion is carried out by means of feature detection and matching analysis between the recorded content in the recorded content memory 4 and the image data of the lesion database on the database memory 8. The results of the feature detection and matching analysis of the subsequent recorded content 14.3 in the recorded content memory 4 with the non-currently applied marred recorded content 15 are exemplarily presented in fig. 4. In this case, this results, for example, in that the damage record 15 not currently used shows a gear wheel arrangement with straight toothing, whereas the subsequent record 14.3 shows a gear wheel with inclined toothing.

Before the feature detection and the matching analysis are applied, in the present embodiment, a correlation between the subsequently recorded content 14.1, 14.2, 14.3 and the image data of the damage database is also carried out within the scope of the comparison step IV. Depending on the threshold value, the correlation is first tested with the damage category that matches this deviation or has the greatest probability, in order to provide a computationally and therefore time-and cost-effective comparison. For this purpose, it is also advantageous to carry out a component-specific damage search in the image data of the damage database, since it is represented in the individual image data which components are known and documented where they are located. It is thus possible to compare further preselected ranges which compare the image data of the damage database with the subsequently recorded content 14.1, 14.2, 14.3 by means of feature detection and matching analysis.

In step V of the method of the present embodiment, the determined type of damage is output. The output has metadata such as the location of the lesion and the point in time at which the lesion occurred, in addition to the lesion category and lesion severity. Alternatively or additionally, feedback to the operation of the component 10 may be automatically caused based on the output.

According to the present embodiment, the comparison means 9 is further configured to: the machine learning Method (ML) is performed in addition to the industrial image processing method (IB method) described in connection with fig. 2-4.

As shown in fig. 5, the ML method 16 is trained by the IB method 17 in the present embodiment. IB method 16 relates to the method previously described in connection with fig. 2-4, wherein for determining the type of damage subsequent recorded content 14.1, 14.2, 14.3 in the recorded content memory 4 is compared with reference recorded content in the reference memory 3 and with image data on the database memory 8.

In optional step VI, the method of this embodiment further comprises verifying, by expert 19, the type of damage 18 output by IB method 17, along with the aforementioned metadata 20. From the output of the IB method 18 confirmed and denied by the expert 19, the ML method 16 can learn a new association between the recorded content recorded in the application and the damage database.

Here, the exchange of the training results of the machine learning method can also be carried out in a product-specific and/or application-specific and/or generic manner, for example, depending on the error mechanism.

List of reference numerals

System for identifying damage to a component

2 Camera

3 reference memory

4 recording content memory

5 comparing device

6 server

7 network

8 database memory

9 contrast device

10 component

11 user interface

14.0 recording content of the component in the reference state

14.1 … 14.3.14.3 recorded content of the Member after Normal use

15 injury record content not currently in use

16 ML method

17 IB method

18 output type of injury

19 expert

20 metadata

I providing the recorded content of the member in the reference state

II obtaining the recorded content of the component after normal use

III comparing the recorded content in the reference state with the recorded content after normal use

To determine the damage

IV compares the recorded content after normal use with the data of the damage database,

to determine the type of injury

Type of V output impairment

VI assessing and correcting the outputted damage type