DE10013137B4 - Process for image-controlled testing and processing of products - Google Patents

Abstract

Process for image-controlled technical testing and / or technical processing of products or product parts
• with first process components (M, feature components) for direct determination of the characteristics of the product to be tested and / or processed, or parts thereof,
With second process components (S, segmental components) for classifying the products to be tested and / or processed or parts thereof,
• with third process components (A, action components) for processing, (after) handling and / or handling of the products or parts thereof and
With fourth process components (R, relation components) for the computer-aided combination of at least two process components (M, S and / or A),
characterized,
a) that each process component (M, S, A, R) is graphically displayed on a display device of a computer and identifiable by a symbolic name,
b) that the process components (M, S, A, R) for creating and / or adjusting the checking and processing method by means of an editor (E) are graphically interactively combinable,
c) that the process components (M, S, A, R) are a ...

Description

1.1 Technical area

The The invention relates to a method for image-controlled technical exam and / or technical processing of products or product parts according to the generic term of claim 1.

The Method is suitable e.g. For a picture-controlled quality assurance.

The Program-related processing processes the image signals of a frame Grabber. It misses e.g. The various components of a complex product, she checks various damages and controls robots to post-process the detected as faulty Components.

1.2 Problems known method

In known methods for image-controlled testing and processing of products become the for a measurement or detection of damage to individual components required process steps in one specific to the product stored program code filed. The process steps are based on information about the structure of the product, the structure of the possible defects, the ones of interest Properties of the product and its components, the resulting product classes and the associated post-processing.

DE-A1-43 38,223, for example, discloses a system and method for Detecting errors or damage in a paint job. The object to be examined - a motor vehicle body - is within a production line on a carrier, for example a chain conveyor, arranged. While the body on the carrier downstream is moved create multiple inspection robots within a coating fault detection station Pictures of previously defined areas of applied paint, For example, the hood, the roof, a side door and the like. Then be the captured images of processing by an image processor subjected. Any errors or damage will occur detected on the basis of the image data and with a grinding or polishing tool, directly on the Test robots attached is repaired. Other testing or machining tasks are not provided, i. the procedure is limited to detecting paint defects or paint damage. The the test procedure underlying program code is self-contained and can not changed become.

at a method known from DE-A1-197 30 885 for automatic Detection of surface defects On body shells is in an optical measuring device by means of a Projection device generates a disk-shaped light tunnel, through the body shell is transported. At the same time, a in a transverse plane of the body shell lying measuring strip the surface irradiated. From this reflected light is under a defined Angle detected by a camera device and as a measurement signal supplied to a computer system. There are surface bumps through detected a comparison with previously filed surface defects. Here too are the for the measurement or detection of damage in the body shell required process steps in one specific to the product stored program code filed. In this can indeed new ones Teach in error types as error patterns. All process steps based on information about the structure of the product, the structure of the possible Errors and the associated post-processing. Related changes are not provided and would, if anything, only through changes in the measuring technique as well as through costly reprogramming possible, what the entire process makes inflexible.

Shorter development cycles to lead however regularly to one always faster change the structure of the product, adding about new components with new possibilities of error come on. Growing quality requirements require additional Surveying and detecting additional errors and their post-processing through robots.

But even easy-to-use script or dialog interfaces to Enable users no really fundamental changes the process steps for the test and processing system. A modification of the product structure or The error structures therefore always require a change of the product tailored program-related process steps. These changes can usually only be made by IT professionals and are with a considerable development and testing effort connected.

An article published in Bild & Ton Bd.45, Issue 11/12 (1992), pp. 320 to 326 describes a specific type of neural networks, namely the so-called hierarchical networks, which are preferred in classifications on the basis of tasks from optical and acoustic quality control - and coding problems are used. Such a network consists of an input layer, an output layer and a number of hidden layers. Their number and the processing lines contained therein depend on the complexity the task. The desired input / output behavior is learned using example pairs. The test methods shown here are always tailored specifically to a product. Changes or variations of the test, measuring and / or comparison parameters or components are not provided.

The Technical Rundschau 50/90 describes on pages 22 to 33 based on Practical examples the advantages and boundary conditions of optoelectronic measurement methods for production-integrated quality assurance. Again, here are the individual components of the illustrated Meßautomaten in a fixed predetermined relationship careful matched to the influence of sources of error within reduce the system and disadvantages of individual elements in the system Balance composite. Transport, sensors and data processing form a self-contained functional unit that has no changes allows. To that System to another measuring task to be able to adapt is a complete reconfiguration not only specialists in optoelectronics, design engineering, Electronics and data processing are necessary, but above all Professionals who work together as part of an overall concept the individual components carefully can coordinate with each other.

The same applies to a in US 5,610,828 disclosed computer-assisted method for the graphically interactive creation of computer programs. This allows computer-aided programming of processes using graphical elements that can be combined on a screen. The graphic elements represent functions, links and data types in the form of icons. They can be graphically combined into a program routine in a block diagram editor and linked together using connection lines. Execution subunits to be included in the program structure execute the user-created program instructions according to the respective compilation.

One major disadvantage of this system is that no static relationships available stand, for example, for a product certain characteristics, errors or to assign components. Therefore it is not possible the tree-like structure of a complex, built up of many parts Model the product. The generated diagrams represent the pure processing structure for the e.g. as part of a Quality inspection accruing Data, but not the structure of the product parts to be tested, errors and features. The user must therefore to create a test system The natural Structure of the problem area into appropriate structures for processing the as part of the exam implement applicable data. This includes the decomposition of the test problem in procedures and the supply of procedures with input values. The creation of a test system However, besides the knowledge of the problem area, this requires considerable Knowledge of procedural data processing, i. relevant programming skills.

task The invention is a method for image-controlled technical exam and / or technical processing of products or product parts to do that quickly without deep programming skills and can be conveniently created. At the same time, any changes, expansion and / or modifications in the product and fault structure as well be quickly and flexibly feasible in the program. Long development times should be avoided as well as expensive programming work.

1.3 Solution Concept

The The object is specified by the characterizing part of claim 1 Characteristics solved. Embodiments are the subject of claims 2 to 13.

The new methods for image-controlled testing and processing of products, results from the use of a novel system graphically interactive combinable components.

The Components encapsulate cameras, robots and process steps. she are specially for such testing and processing tasks suitable, essentially over a classification of the to be tested Products and components solved can be. she represent in this problem area for the User relevant concepts, e.g. Damage, areas, angles, Components, products, product groups, cameras, robots and their relationships.

The Component system contains Segment, feature and action components These are defined by relationship components combined together. There are different types for each component type specific characteristics.

The Segment components represent the Procedural steps for a selection or classification e.g. of pixels, edges, Angles, shapes, damages, Product components, products or product groups.

The feature components represent Sensors for directly determining the features, such as the frame grabber for determining the gray values, or the method steps for calculating features from previously determined features such as the gradient of a pixel, the contrast of an image area, the area of a blob, the number of damage of a product, the shape of a product or its affiliation to a defect class.

The Action components encapsulate the robots, e.g. for polishing or Repainting the damaged ones Products, or process steps for indirect processing, the rely on robot actions.

The Components are over the type and a symbolic name are identified and are graphical interactively combinable.

she support a representation as an icon, possibly a pixel representation in the frame of the examined picture as well as the loading and storing of the relevant Data such as thresholds, examination results, camera or robot parameters.

The for a product-specific testing and processing system required measurement steps, Detection of damage, classification the components and the product, as well as to determine the associated features and to control the machining actions therefore result the product-specific semantic network which goes through Combination of segment, feature and action components using relationship components results.

This specified from the components existing semantic network the structure of the product, the structure of the errors that occur, the product classes, the characteristics used for classification (including the frame grabber) and the actions planned for processing (including the robot).

The Creation or adaptation of a product-specific testing and processing system takes place graphically interactive with an editor essentially by drag and drop.

Of the Editor allows the interactive testing of Sharing the product to be examined, editing thresholds for fault tolerances, that Viewing the test results and the definition of study areas of interest.

Of the Editor allows the display of the detected by the frame grabber and the edited images, the storage of the network and the results in a database as well as the modification of the parameter settings for frame Grabber and robot.

Out The use of the component system yields the following Steps of the method according to the invention.

Procedural steps in Framework for creating or modifying a testing and processing system

• - Take up the error pictures by the frame grabber
• - complete the Component system with product-specific components
• - Modeling the product structure with classification and relationship components
• - Modeling the error structure with classification and relation components
• - Determine the error and adjusting the thresholds
• - Modeling the error characteristics with feature and relationship components
• - To calculate the error characteristics and adjusting the thresholds
• - Modeling the characteristics of the product with feature and relationship components
• - Modeling the error classes of the product with classification and relation components
• - Classify of the product and adjusting the thresholds
• - Modeling the editing actions with action and relationship components
• - Executive Editing actions and adjusting the action parameters

Procedural steps in Frame of the exam and processing of the product

• - Instruct the action components with the processing of the modeled error classes
• - Instruct the classification components with the product classification
• - Instruct the feature components with the calculation
• - Take up of the test to be tested Picture by the frame grabber
• - Determine the error using classification components
• - To calculate the feature components
• - Classify of the product using classification components
• - Executive Action components for processing by robots

extensive Testing and Processing systems require repetitions of certain sequences of Process steps.

1.4 Advantages of the invention

The Use of many small drag and drop combinable components for the Building a product-specific process leads to an easy adaptability to modified product structures or increasing quality requirements. The measurement of a new component or the detection and processing a new mistake usually only require the addition of some segment, Feature and action components and thus few mouse clicks. This allows a fast and thus cost-effective Conversion to a wide variety of company products or quality standards.

Also at completely new products reduce the development time for the testing and processing system considerably.

This even applies in cases in which the existing component system by company or product specific Components must be extended.

There the icons of the components the relevant concepts and relationships the semantic network will reflect the relevant structures the test and processing steps for Customers and employees without studying the underlying code comprehensibly.

The Use of the components leads to a clear clean object-oriented structure of the product-specific Testing and Processing system with appropriate reusability and low error rate.

1.5 Industrial Applicability

The Method is suitable e.g. For a flexible image-controlled quality assurance.

In This case will be based on the image signals of a frame grabber e.g. measuring the various components of a product for damage checked and robots controlled for post-processing.

With the outlined method Components are optically recorded, examined for their shape and shape-specific to be edited.

As well The method can be used to flexibly control fertilizer or weed control systems be based on the optical detection and classification agricultural land.

To an extension of the pictures to 3 dimensional information structures can with the method about tissue changes detected in 3-dimensional medical CT images and treated if necessary become.

The Classification in its general form is not based on input information limited in the form of pictures. In place of pixels are arbitrary Information processable. For example, can make sounds or smells of Sensors are detected and stored in the form of a feature amount. This feature set can then be applied to specific noise or odor segments be inspected, the noise or odor generator can be classified and the corresponding reaction initiated in the form of actions become.

1.6 Exemplary embodiment of claimed invention

1.6.1 The component system

1.6.1.1 Overview

The The core idea of the process is the abstraction of the framework concrete image analysis, component classifications and processing used process steps to a system graphically interactive combinable components.

So are about pixels, edges, angles, damage, product parts, the Product itself and even entire product groups by means of in a certain Modeled according to uniform segment components.

One Segment designates first a certain selected set of pixels of an image.

• – Ein Kantenfinder erzeugt ein Segment von Bildpunkten, die zur Kante gehören. Alle diese Bildpunkte besitzen etwa einen hohen Grauwertgradienten als Merkmal. Aus einer abstrakten Sicht prüft der Kantenfinder Merkmale von Bildpunkten, klassifiziert oder selektiert diejenigen Bildpunkte, deren Merkmal einen gewissen Schwellwert übersteigt, und fasst sie zu einem Kantensegment zusammen.
• – Ein Kantenklassifizierer prüft z.B. die Länge von Ausbrüchen und selektiert diejenigen Kanten, deren Ausbrüche über einem gewissen Schwellwert liegen. Aus einer abstrakten Sicht prüft der Kantenklassifizierer Merkmale der Kantensegmente und erzeugt ein Segment aus Kanten mit großen Ausbrüchen.
• – Ein Winkel kann als ein Segment betrachtet werden, welches aus zwei Kantensegmenten besteht. Die Segmentbildung wird in diesem Falle jedoch in der Regel nicht von dem System sondern von dem Anwender vorgenommen. Der Anwender selektiert zwei Kantensegmente aus der Menge aller Kanten und ordnet sie einem Winkelsegment zu. Das Prüfen eines Winkelsegmentes besteht aus einer abstrakten Sicht zunächst darin, für jede der beiden Kantensegmente die Berechnung der Regressionsgeraden als Merkmale und für das Winkelsegment die Berechnung der Differenz der Steigungen als Merkmal zu initiieren. Anschließend werden wiederum alle Winkelsegmente, die innerhalb eines gewissen Intervalles liegen, zu einem Segment der guten Winkel zusammengefaßt.
• – Ein Spielzeugauto entsteht als Produkt, indem aus einem Lager eine gewisse Menge von Bauteilen selektiert und zu einem neuen Ganzen zusammenfügt werden. Das zu untersuchende Produkt entsteht aus einer abstrakten Sicht also durch Selektion gewisser Bauteile aus der Menge aller Bauteile eines Lagers und durch das Zusammenfassen dieser Bauteile zu einem Produktsegment.
• – Die Menge der Autos, die vor der Auslieferung nachgearbeitet werden müssen, entsteht, indem aus der Menge aller hergestellten Autos bestimmte Autos selektiert werden. Ein Produktklassifikator initiiert z.B. die Berechnung solcher Merkmale wie die Anzahl großer Beschädigungen oder den Mittelwert der Kantenausbrüche. Er selektiert etwa diejenigen Produkte, deren Anzahl oder Mittelwert einen bestimmten Schwellwert übersteigt und fasst sie zu einem Segment der fehlerhaften Produkte zusammen. Schließlich initiiert er eine bestimmte Aktion zur Nachbearbeitung.

The following examples explain how this term and thus the corresponding component of the component system can be generalized.

• - An edge finder generates a segment of pixels belonging to the edge. All of these pixels have about a high gray scale gradient as a feature. From an abstract view, the edge finder checks features of pixels, classifies or selects those pixels whose feature exceeds a certain threshold, and combines them into an edge segment.
• For example, an edge classifier checks the length of breakouts and selects those edges whose breakouts are above a certain threshold. From an abstract view, the edge classifier examines features of the edge segments and creates a segment of edges with large eruptions.
• - An angle can be considered as a segment consisting of two edge segments. However, segmentation in this case is usually not done by the system but by the user. The user selects two edge segments from the set of all edges and assigns them to an angle segment. The examination of an angle segment consists of an abstract view first of all for each of the two edge segments to initiate the calculation of the regression line as features and for the angle segment the calculation of the difference of the slopes as a feature. Subsequently, in turn, all angle segments that are within a certain interval, combined to form a segment of the good angle.
• - A toy car is created as a product by selecting from a warehouse a certain amount of components and assembling them into a new whole. The product to be examined arises from an abstract view, ie by selecting certain components from the set of all components of a bearing and by combining these components into a product segment.
• - The amount of cars that have to be refurbished before delivery is created by selecting specific cars from the quantity of all produced cars. A product classifier, for example, initiates the calculation of such features as the number of major damages or the mean value of the edge breakouts. For example, he selects those products whose number or average exceeds a certain threshold and combines them into a segment of the defective products. Finally, he initiates a specific post-processing action.

In consider all cases the calculation of features is initiated and it will be based Select or classify segments from a set on the characteristics and combined into a new segment.

Therefore can about edges, the amount of broken edges, an angle, that Product car, the amount of defective cars is essentially below Using the same method steps are classified.

These Process steps are summarized in the segment component.

Finally for the segments received initiated actions. For example, must be the amount of as recognized defective toy cars repainted, repolished or be removed.

The Component system therefore defines action components which be coupled to a segment.

A represents such an action component such as the process steps for post-processing the defective cars. It supports usually look at actions taken for the subsegments, in the example the cars to perform are.

The Action component for The cars then use actions again for the individual components of the car Cars to perform are, until finally For example, an action component is executed, which is the painting robot or the polishing robot.

in the Frame of execution This component will then call drivers for the polishing or painting robot and edited the recognized error regions.

The Hang action components in stronger Measures of the specific product and its environment than the segment, Feature and relationship components. Therefore, part of these components usually developed only in consultation with specific companies and customers become. Enable the companies the action components then have the same flexibility as the other components, since the Company and product specific actions then also graphically interactively combinable.

• – Die Segmentkomponenten repräsentieren die Verfahrensschritte für eine Selektion bzw. Klassifikation z.B. von Bildpunkten, Kanten, Winkel, Formen, Beschädigungen, Produktbauteilen, Produkten oder Produktgruppen.
• – Die Merkmalkomponenten repräsentieren Sensoren zur direkten Ermittlung der Merkmale, wie z.B. den Frame Grabber zur Ermittlung der Grauwerte, oder die Verfahrensschritte für eine Berechnung der Merkmale aus zuvor ermittelten Merkmalen wie z.B. den Gradienten eines Bildpunktes, den Kontrast eines Bildbereiches, die Fläche eines Blobs, die Anzahl der Beschädigungen eines Produktes, die Form eines Produktes oder dessen Zugehörigkeit zu einer Fehlerklasse.
• – Die Aktionskomponenten kapseln die Roboter, z.B. zum Nachpolieren oder Nachlackieren der beschädigten Produkte, oder Verfahrensschritte zur indirekten Bearbeitung, die sich auf Roboteraktionen abstützen.
• – Die Relationskomponenten repräsentieren vielfältige Beziehungen zwischen den übrigen Komponenten. Sie speichern z.B. Informationen über die Art der Beziehung oder ihre Darstellung im Rahmen des grafischen Editors.

Overall, the following basic components result

• The segment components represent the method steps for a selection or classification, for example of pixels, edges, angles, shapes, damages, product components, products or product groups.
• The feature components represent sensors for directly determining the features, such as the frame grabber for determining the gray values, or the method steps for calculating the features from previously determined features such as the gradient of a pixel, the contrast of an image area, the area of a blob, the number of damage to a product, the shape of a product or its affiliation to a defect class.
• The action components encapsulate the robots, eg for repolishing or repainting the damaged products, or process steps for indirect processing, which are based on robot actions.
• - The relationship components represent multiple relationships between the remaining components. For example, they store information about the type of relationship or its representation in the context of the graphical editor.

The Components are specially designed for Testing and processing tasks suitable and represent the in this area of responsibility for the user relevant concepts.

The Components are over the type and a symbolic name are identified and are graphical interactively combinable.

she support a representation as an icon, possibly a pixel representation in the frame of the examined image as well as the editing, loading and saving the component-specific data such as threshold values, characteristic values, Camera and robot parameters.

Out these components can then be used for a concrete problem is a problem-specific semantic network for classification and editing graphically interactive by drag and drop are created.

Of the Structure of the semantic network reflects the structure and the properties of the product under investigation and its components, the expected defects and the product classes to be distinguished contrary.

Around impracticable or faulty test and Preventing processing systems to a great extent becomes the semantic one Network before running a consistency check subjected.

For example, allowed to some segments no ordinary Segments are assigned as subsegments. It may not be cyclical Subsegment relationships.

Such Relationship rules can also be extended to company- or product-specific rules.

1.6.1.2 Examples of segments

• - Segment : Represented the basic component. All other segment components inherit the Properties of this component. Such a segment may e.g. one Represent car, the user is assigned the subsegments chassis and engine Has. Chassis and engine are also made by simple segment components modeled. It can represent an angle to which the user has assigned two edge segments as subsegments. It can be one Represent amount of error segments, which the user has assigned to a specific component.
• - ImagePointSetSegment : Represented the set of all pixels. A pixel is in two-dimensional Case defined only by its (x, y) -coordinate. The gray value of the pixel is its characteristic.
• - BinarySegment : Represented one through a binary image given amount of pixels. This can be about those pixels be that belong to an edge or belonging to a soiled area of the component. It receives as input a gray value image and a threshold value.
• - AreaOfInterestSegment : Represented a subset of pixels to be specified by the user. This Range can e.g. by a coordinate rectangle, a vector parallelogram or be given a Binaerbild.
• - ObjectSeparatingSetSegment : Generates one from a segment given by a binary image Set of segments for the individual separated objects.
• - Interval segment : Represented a lot of segments whose characteristic is in a real interval lies. The characteristic and the interval limits as threshold values can be determined by the Users can be specified or at runtime from others Characteristics are calculated.
• - CubeSegment : Represented a set of segments whose feature vector lies in an n-dimensional cube.
• - AndSegment : Receives as input multiple segments, each containing a lot of subsegments have. It selects those subsegments that are in all quantities are included.
• - OrSegment : Receives as input multiple segments, each containing a lot of subsegments have. It selects those subsegments that are in at least one of the quantities are included.
• - NtupelGeneratingSetSegment : Generates all N-tuples of a given set of segments. This is e.g. useful, when found about segment tuples with certain positional relationships should be.
• - FuzzySetSegment : Represented a classification of segments using fuzzy logic. For example, could The FuzzySetSegment has a set of subsegments, each one represent a fuzzy set. In a learning phase could the FuzzySetSegment for each of its fuzzy sets of characteristic feature vectors calculate a center vector and in the classification phase it would for a segment characterized by its feature vector membership to calculate the fuzzy quantities.
• - NeuroNetSegment : Represented a classification of segments using a neural Network.

1.6.1.3 Examples of features

• - RGBSetMark: Represents a frame grabber to determine the gray value or color image together with the determined image. The gray value of a pixel is a feature of the pixel. The RGB feature of a pixel is given, for example, by an RGB triple or a palette index. The RGBSetMark encapsulates the RGB characteristics of all pixels in addition to the frame grabber.
• - BinaryMark : Represented a binary image. Each pixel is thus assigned the characteristic 0 or 1.
• - SobelSetMark : Represented the result of using the Sobel operator e.g. on a RGBMark. Used for directional edge detection.
• - Laplace SetMark : Represented the result of using the Laplace operator e.g. to a RGBMark. Used for direction-independent edge detection.
• - ErosionSetMark : Represented the result of application of erosion e.g. on a RGBMark.
• - DilationSetMark : Represented the result of application of a dilatation e.g. on a RGBMark.
• - OpeningSetMark : Represented the result of applying an opening operation e.g. to a RGBMark.
• - Closing SetMark : Represented the result of applying a closing operation e.g. to a RGBMark.
• - AndSetMark : Receives as input two BinaryMarks. represents the result of a wound up of two Binaerbilder.
• - OrSetMark : Receives as input two BinaryMarks. represents the result of a derivation of two Binaerbilder.
• - HistogramMark : Represented the result of a histogram calculation e.g. for a RGBMark.
• - ContrastMark : Represented the result of a contrast calculation in a by a BinaerSegment given area.
• - GrayValueProfileMark : Represented the result of calculating the gray value profile e.g. for a RGBMark.
• - ObjectSeparatingMark : Receives as input a binary image and creates an image with separated objects.
• - RegressionLineMark : Receives as input the binary image an edge and calculates the associated regression line.
• - AngleMark : Receives as input two regressionLineMarks and calculates the angle between the straight line.
• - DistanceProfileMark : Calculated to an item given by a BinaerSegment using the center of gravity the distance profile (distance between center of gravity and body edge). This is e.g. the shape recognition via a correlation with known Profiles.
• - CorelationMark : Calculates e.g. the correlation between profiles.
• - CentreOfGravityMark : Receives a binary image as input and calculates the focus.
• - SizeMark : Receives a binary image as input and calculates the area.
• - CountMark : Counts the number of subsegments of a segment.
• - FourierMark : Represented the result of using the Fourier operator e.g. on a RGBMark. Serves for the preparation of examinations in the momentum space.
• - CLiftSegmentMark : Transport the value of a feature from a subsegment an overlying one Segment upwards. These features allow access to features that in the semantic network one or more levels below the currently to be calculated feature or segment.
• - MinusMark : Receives each input two real-valued features and calculates the difference. In order to For example, you can calculate the difference between two angles. (For others algebraic operations analogous)
• - DistanceMark : Receives as input two characteristics that have points as a result, and calculated the Euclidean distance of the points. Serves e.g. for distance calculation two focal points.

1.6.1.4 Examples of actions

• - Action : Represented the basic component. All other actions inherit the properties of these Component. This component encounters the execution of all Actions associated with it by means of an ActionActionUsesRelation are.
• - PaintAction : Represented the painting robot for post-processing defective paint spots. The company-specific action component gets from her via a SegmentActionHasRelation associated with the segment that damaged the area descriptive binary image and edit this area by painting. The action component uses the driver calls for the painting robot.

1.6.1.5 Examples of relationships between the components

• - MarkMarkUsesRelation : Need Marks to calculate usually other marks. For example, needs that SizeMark a BinaryMark as input.
• - SegmentMarkHasRelation : A segment has or owns several marks.
• - SegmentMarkUsesRelation : A segment uses or uses one or more markers for selection.
• - SegmentSegmentFilterRelation: A segment selects its subsegments from a through other segment given amount. It filters this set of segments.
• - Segment SegmentHasRelation : A segment has or has subsegments.
• - SegmentActionHasRelation : A segment has or has associated actions.
• - ActionActionUsesRelation : An action uses different actions

1.6.2 Procedural steps as part of the creation of a testing and processing system

The Process steps are exemplified by examination and post-processing a toy car explained.

The Testing and Machining system is designed to be exactly one toy car visually detected and the roof is inspected for errors.

in the In the case of pollution is polished and in case of sufficient huge damage is repainted.

The Presentation of the semantic network is in the present Shape certainly still getting used to.

Of the graphical editor is intended to distinguish between segments, features, Actions and the diverse Support relations through color and shape. This for creation The case tool used in the drawings allows these distinctions However not.

Of the graphic editor is supposed to have different views of a semantic one Network in conjunction with hiding unessential segments or Support relationships.

The Case-Tool used here to create the drawings supports this However not.

1.6.2.1 Recording the Error pictures by the frame grabber

through Frame Grabber will produce images of such toy cars, where the extent of Dirt or paint defects just above or below the tolerance threshold lies.

1.6.2.2 Adding the Component system with product-specific components

In In this process step, the component system is an action component PaintAction added for re-painting. The action component PaintAction gets the damaged area from its segment descriptive binary image and controls the painting robot. The action component encapsulates with it the driver calls for the painting robot.

A appropriate supplement done by an action component for polishing.

1.6.2.3 Modeling the Product structure with classifying and relationship components

(See drawing 1.6.2.3)

In This process step underlies the structure of the toy car Use of segments and SegmentHasRelations modeled. On this way it is specified which components of the car are checked and possibly should be reworked.

in the For example, the tree of the car consists only of roof and Engine Hood.

1.6.2.4 Modeling the Error structure with classifying and relationship components

(See drawing 1.6.2.4)

In This process step models the defect region of the roof.

For the roof be first the areas to be examined are specified by means of two AreaOfInterestSegments "roofRegion1" and "roofRegion2". The Dimensions of the AreaOfInterestSegments can be edited in a dialog or by directly positioning the associated rectangles in the image become.

The entire faulty region "roofFaultRegion" is being used of a binary segment.

this Segment will be a BinaryMark "rooferrorBinaerBild" to record the Result in the form of a binary image assigned. The assignment is made with a SegmentMarkUsesRelation.

Further the "RoofFaultRegion" segment becomes the Selection of the pixels assigned to feature to be used. For example it is the RGBSetMark. This feature represents that Grayscale image generated by the camera.

In a dialog belonging to the BinarySegment will be a preliminary Threshold interval for the Selection of the pixels of the RGBSetMarks fixed.

Finally, the BinarySegment "roofFehlerRegion" is assigned the two AreaOfInterestSegments via a SegmentSegmentFilterRelation ment, in which areas the gray value of the pixels is to be compared with the predetermined threshold interval.

1.6.2.5 Determining the Error and adjusting the thresholds

In This step is the selection of the pixels according to the set Threshold triggered. The BinarySegment "roofFaultRegion" representing the fault region of the roof selects those Pixels that are in the areas of interest and their Gray values are within the set threshold interval. These Pixels are in the binary image marked as "black." The remaining pixels are marked as "white." The result The selection is generated directly by the creator of the test system binary image checked.

The Threshold interval is modified if necessary and the execution of the Selection repeats. At the end of this process, the region marked as "black" corresponds of the roof to the area classified as faulty.

1.6.2.6 Modeling the Error characteristics with feature and relationship components

(See drawing 1.6.2.6)

in the With regard to a quantitative assessment of the fault region suitable features are modeled in this method step. For example will a SizeMark "size" to determine the damaged or polluted area generated. A ContrastMark "contrast" should in the example to distinguish between high-contrast damage and low-contrast soiling. The features are the Segment "roofFaultRegion" by means of a SegmentMarkHasRelation assigned.

1.6.2.7 Calculating the Features and adjusting the thresholds

In This method step initiates the calculation of the characteristics. The Characteristics initiate the selection of the pixels by the segment "roofFaultRegion" Characteristic calculation can be read in a feature-specific dialog become. Possibly. the calculation is repeated with a modified threshold.

1.6.2.8 Modeling the Features of the product with feature and relatons components

(See drawing 1.6.2.8)

In In this process step, the calculated error characteristics are determined via special LiftMarks along the child-father relationships between fault region, Roof and car transported up and so to characteristics of the examined Cars transported. About these Technology can Characteristics of various errors in different components too a feature vector of the car to be uplifted. The characteristics are about MarkMarkUsesRelations linked together.

in the An example will be the size and contrast of the fault region transported. By bumping the calculation of the LiftMarks "autoDachFehlerRegionGroesse" can be the transport to be controlled. The dialog of the LiftMark shows after the calculation the right value.

1.6.2.9 Modeling the Failure of the product with classification and relationship components

(See drawing 1.6.2.9)

In This step is the classification of the examined Cars prepared.

To The amount of all cars is modeled with a segment "autoMenge" and the amount of cars with a sufficiently large Error region on the roof with an interval segment "autoMengeMitDachFehler".

The IntervalSegment comes with the segment representing the amount of all cars represents, over one SegmentSegmentFilterRelation connected. For the IntervalSegment is a preliminary Threshold for the size of the error region on the roof by means of belonging Specify dialog. Furthermore, this interval segment will be notified which real-valued characteristic is to be used for selection.

in the For example, the feature of the car is used, which is the size of the error region on the roof describes.

In addition, will the amounts of cars with a pollution or damage on the Roof is modeled by the interval segments "auto-amount with roof contamination" or "auto-amount with roof damage". Both Segments are over a SegmentSegmentFilterRelation is connected to the interval segment "autoMengeMitDachFehler" in the example These segments use the contrast of the error region for selection. For both segments have to suitable preliminary threshold intervals by dialogue be specified.

1.6.2.10 Classify of the product and adjusting the thresholds

(See drawing 1.6.2.10)

In This step initiates the execution of the selection for the interval segment "autoMengeMitDachBeschaedigung" the specified thresholds in the example shown is the Segment "auto" selected and the IntervalSegment "autoMengeMitDachFehler" and the interval segment "autoMengeMitDachBeschaedigung" assigned as a subsegment.

The Selection may be repeated with modified thresholds, until the assignment in the desired Way done.

in the With regard to a refined classification, the method steps "modeling the error structure with classification and relation components "to" Classify of the car and adjusting the thresholds "again

1.6.2.11 the error structure with classification and relationship components

(See drawing 1.6.2.11.1)

In This process step is the amount of individual damage the roof area with an ObjectSeparatingSegment "roofFaultBlobMenge" modeled Segment is using the entire error region via a SegmentSegmentFilterRelation connected.

(See drawing 1.6.2.11.2)

The ObjectSeparatingSegment creates a set of binary segments, which the separated error blobs and thus the individual damages represent. In the example, for better illustration, 3 subsegments of the ObjectSeparating segment. The graphical editor shows however by default only a subsegment and notes the cardinality the subsegment set in the form of a number.

The Lot of the big ones Blobs are modeled with an interval segment "roof-fault-large-blob-amount".

It will over a SegmentSegmentFilterRelation with the ObjectSeparatingSegment connected. This interval segment "RoofErrorBrosseBlobMenge" is used as part of the associated Communicating which uses real-valued characteristic for selection shall be. In the example, the area of the blobs is used as a feature.

Further is for this segment is a tentative threshold for the area to specify the blobs.

1.6.2.12 Modeling the Error characteristics with feature and relationship components

(See drawing 1.6.2.12)

in the With regard to a quantitative assessment of the damage suitable features are modeled in this process step.

in the For example, SizeMarks are used to determine the area of the blobs. These are assigned to the error blobs by means of a SegmentMarkHasRelation.

to Recording the number of big ones Blobs are assigned a CountMark "numberBigBlobs" to the "roofFaultGrosseBlobMenge" interval segment. This counts the Number of subsegments of the interval segment and therefore determines the number of big ones Blobs.

1.6.2.13 Calculate the Error characteristics and adjustment of the threshold values

(See drawing 1.6.2.13)

In This step initiates the calculation of the CountMark.

The CountMark assigns the interval segment "roof faultBrosseBlobMenge" with the execution of the Selection. This interval segment initiates the execution of the ObjectSeparatingSegments "roofErrorBlobMenge" and thus the generation of the Blobs.

The IntervalSegment "roofFehlerGrosseBlobMenge" then commissions for each Blob the calculation of the area and selects those blobs whose area is above the specified threshold lies.

Of the Process step is repeated with modified thresholds as long as until the number of big ones Blobs matches the expected number.

in the For example, 2 of the 3 blobs were classified as sufficiently large and the associated ones BinarySegments to the segment "RoofErrorBrosseBlobMenge" as subsegments assigned. The dialog belonging to CountMark "numberBigBlobs" shows therefore the value 2.

1.6.2.14 Modeling the Features of the product with feature and relationship components

(See drawing 1.6.2.14)

In In this process step, the CountMark "numberBigBlobs" will be along special LiftMarks the child-father relationships between crowd of big blobs, Roof and car transported up and so to a feature of the transported examined cars.

By Nudge the calculation of the LiftMarks "autoDachAnzahlBrosseBlobs" can the transport to be controlled. The dialog of this LiftMarks shows after the Calculate the correct number, in the example 2.

1.6.2.15 Modeling the Error classes of the product with classification and relation components

(See drawing 1.6.2.15)

In This step is the classification of the examined Cars further refined.

To For this purpose, the quantities of cars with small or large damage to the Roofs are modeled using the interval segments "autoMengeMitSmall RoofDamage" and "autoMengeMitGrosserDachBeschaedigung". Both Segments use the characteristic "autoDachNumberGrosseBlobs" for the selection, as part of the associated dialogs become provisional Threshold intervals specified.

1.6.2.16 Classify of the product and adjusting the thresholds

(See drawing 1.6.2.16)

In This process step initiates the execution of the selection.

The Car whose picture is for the preparation of the test and machining system is used in the example, the interval segment "autoMengeMitGit RoofDownPresentation" as a subsegment. Execution If necessary, the selection is repeated with modified threshold values.

1.6.2.17 Modeling the Editing actions with action and relationship components

(See drawing 1.6.2.17)

the IntervalSegment "autoMengeMitGrosserDachBeschaedigung" is the action "paintAutoDaecher" using SegmentActionHasRelation assigned. She only instructs her about an ActionActionUsesRelation Assigned action "paintAutoDach" of the segment "auto" with the execution. In in the same way, this action commissions the PaintAction "PaintedDownFaultRegion" associated with the "roof" segment. This action was before the preparation of the test and machining system added to the existing component system. The Action represents the lacquering of a given area by a binary segment. In the example, this area is determined by the binary segment "roofFaultRegion".

The Action allows to set certain parameter settings for the postprocessing, such as such as the type of paint or the spray distance adjust.

1.6.2.18 Running the Editing actions and adjusting the action parameters

In This process step triggers the execution of the action "paintAutoDaecher" Procedural step will be done with modified action parameters repeated until the post-processing shows the desired result.

1.6.2.19 The complete testing and processing system

in the Below is an overview of the discussed in the above steps part of the testing and processing system for the Toy car given.

(See drawing 1.6.2.19)

As already mentioned above supports the graphical editor the creation of individual views, hiding non-relevant relationships and the optical separation of the different ones Components and relations through color and shape.

A "complete" presentation of the semantic network is in more extensive testing and processing systems no longer useful.

Also in this overview is e.g. waived a representation of the investigations for the hood.

The Overview presented here but still leaves once the object-oriented oriented to the relevant concepts Structure of the test and processing system from a few common components recognize.

in the middle area of the picture becomes the mostly tree-like structure of the product visible, in the example, the car consists only of Roof and hood.

among them From left to right, the structure of the errors to be tested builds on.

The AreaOfInterestSegment "roof region" represents the region to be examined representing the binary segment "roofFaultRegion" represents the entire detected error range and the interval segment "roof-fault-large-blob-amount" after all single selected blobs. Here are a variety of other error structures conceivable.

The Error segments each have a link to the corresponding ones Binary images. These binary images contain gray values of pixels and thus represent features of Pixels. Therefore, they are characteristics of the ImagePointSetSegment by means of SegmentMarkHasRelation assigned.

The Built-up network of error segments forms the basis for the calculation of characteristics for quantification. In the example, these are the size of the error region, the contrast, the sizes of the individual Blobs and the total number of big blobs.

These Features are over Special LiftMarks along the child-father relationships within the Semantic network transported up and so to characteristics of the examined car. The transported upward characteristics of individual components of the examined Cars then enable its classification.

The Different error classes can be found in the upper part of the testing and processing system. The segments for classifying the product form with theirs Filter relationships a decision tree that provides expert knowledge about the sequence the one to be carried out Representations.

Go through this Product this decision tree, so depending on the path different Features for the individual components calculated.

Especially are thus subsequent investigations, i. Calculations of features, depending on the result of previous investigations. The investigation chain is thus controlled dynamically. For example, the calculation of the feature "autoDachAnzahlBrosseBlobs" is triggered only if the car is assigned to the interval segment "autoMengeMitDachBeschädigung". The calculation of the feature "autoDachNumberBrosseBlobs" is not triggered if the car is assigned to the interval segment "autoMengeMitDachVerschmutzung".

The Finally, the classifications made form the basis for the various actions to correct the detected errors.

1.6.3 Process steps as part of the exam and processing of the product

in the The method steps which are described during the execution of the test and Processing system performed become. These process steps represent those within the semantic network running dynamics.

1.6.3.1 Commissioning the Action components with the processing of the modeled error classes

The System comes across this Procedural step for all modeled error classes the associated post-processing actions at.

in the Thus, for example, the action "paintAutoDaecher" for the interval segment "autoMengeMitGrosserDachBeschaedigung" is triggered.

1.6.3.2 Commissioning the Classification components with the product classification

The Action "paint car roofer" commissioned in this Process step their segment "autoMengeMitGroßDachBeschaedigung", the associated cars to investigate.

This Segment therefore prepares the selection of the subsegments. To this Purpose assigns it again the preceding segment "autoMengeMitDachBeschädigung", the associated cars to investigate.

So the order is planted to determine the associated cars from error class to error class along the decision tree continued. In our example, the chain of care ends at the segment "autoMenge" has in the example considered only a subsegment, which already was assigned in the modeling.

1.6.3.3 Commissioning the Feature components with the calculation

The Error class "autoMengeMitDachFehler" starts first with the selection of the associated Cars. For this purpose, she commissioned the calculation of the LiftMarks "autoDachFehlerRegionGroesse" for the segment "auto". This LiftMark then commissioned the calculation of the LiftMarks "roofFehlerRegionGroesse" of the segment "roof".

thereupon This LiftMark commissioned the calculation of the SizeMarks "size" of the binary segment "roofFehlerRegion".

The SizeMark "groesse" commissions the binary segment "roofFehlerRegion" with the determination of the Error region.

For the AreaOfInterestSegmente was the selection of the subsegments in the form of the associated pixels already undertaken as part of the modeling by the user.

The BinarySegment "roofFehlerRegion" instructs the RGBSetMark "autoGrauWertBild" with the determination of the gray value image.

1.6.3.4 Recording the to be tested Picture by the frame grabber

Of the Frame Grabber generated with the for the product specific camera setting the gray scale image of the currently to be examined and possibly nachzubearbeitenden cars.

1.6.3.5 Determining the Errors using classification components

The BinarySegment "roofFaultRegion" created using of the RGBSetMarks "autoGrauWertBild" and by the AreaOfInterestSegment given areas the BinaryMark "roof errorBinaerBild".

1.6.3.6 Calculating the feature components

In This step is followed by the calculation of the SizeMarks "size", the LiftMarks "roofFehlerRegionGroesse" and the LiftMarks "autoDachFehlerRegionGroesse".

The sought feature is thus transported up to the car.

1.6.3.7 Classify of the product using classification components

The IntervalSegment "autoMengeMitDachFehler" checks in this Step the LiftMark "autoDachFehlerRegionGroesse" and add in the considered Example the segment "auto" from the subsegments of the segment "autoMenge" its own subsegments added.

1.6.3.8 Order the Feature components with the calculation

The Error class "autoMengeMitDachBeschädigung" instructs the calculation of the LiftMarks "autoDachFehlerRegionKontrast" for the segment "auto". This LiftMark then instructs the calculation of the LiftMarks "RoofFaultRegionContrast" of the "roof" segment.

thereupon This LiftMark commissioned the calculation of the ContrastMark "contrast" of the binary segment "roofFehlerRegion".

1.6.3.9 Calculating the feature components

The BinaryMark "roofFehlerBinaerBild" already exists.

Therefore In this process step, the calculation of the ContrastMark "contrast", of the LiftMark "roof errorRegionContrast" and of the LiftMarks "autoDachFehlerRegionKontrast".

The sought feature is thus transported up to the car.

1.6.3.10 Classify of the product using classification components

The IntervalSegment "autoMengeMitDachBeschädigung" checks the LiftMark "autoDachFehlerRegionKontrast" and adds in the considered Example the segment "auto" from the subsegments of the segment "autoMengeMitDachFehler" with its own subsegments added.

1.6.3.11 Commissioning the Feature components with the calculation

The Error class "autoMengeMitGrosserDachBeschädigung" instructs the calculation the characteristic "autoDachNumberBigBlobs" for the segment "auto". This LiftMark then instructs the calculation of the LiftMarks "roof number large Blob" of the segment "roof". Then commissioned this LiftMark the calculation of the CountMark "numberBigBlob" of the interval segment "roofFaultBigBlobMenge".

1.6.3.12 Determining the Errors using classification components

The CountMark "Count BigBlobs" commissioned in this Step of the process is its interval segment "roofFaultBigBlobMenge" with the selection the big Blobs.

The IntervalSegment therefore encounters execution of the over An objectSeparatingSegment "roofFaultBlobMenge" associated with a filter relationship. This segment generates binary segments "blobi".

The IntervalSetSegment "roofErrorBrosseBlobMenge" initiates the calculation the SizeMarks for the blobs and selects those whose size is above the specified threshold lies.

1.6.3.13 Calculate the feature components

in the next Step is the calculation of the CountMarks belonging to the IntervalSegment "numberBigBlobs", of the LiftMarks "roof number largeBlobs" and the LiftMarks "auto roof number large Blobs" So it is transported up to the car.

1.6.3.14 Classify of the product using classification components

The IntervalSegment "autoMengeMitGrosserDachBeschädigung" checks the characteristic "autoDachAnzahlBrosseBlobs" and adds in the considered Example the segment "auto" from the subsegments of the "AutoMengeMitDachBeschädigung" segment with its own subsegments added.

1.6.3.15 Running the Action components for editing

In This process step is finally the error class belonging Action triggered. in the For example, the action "paintAutoDaecher" is executed.

she assigns the assigned action "paint car roof" of the segment "auto" with the execution. In In the same way, this action commissions the "paint roof defect error" action assigned to the "roof" segment. This action controls the painting robot e.g. using the error region determined Binary image.

1.6.3.16 The dynamics at a glance

After the presentation of the individual process steps, an overview of the dynamic processes takes place:
The assignment of the segment "auto" propagates along the decision tree modeled by the error classes, whereby each error class initiates the calculation of a specific feature of the product to be examined. This feature calculation leads to a cascade of feature calculations for the individual components running downwards in the semantic network Finally, in the context of binary image generation, the feature computations lead to the determination of the gray value image by the frame grabber and image processing operations for determining the errors.

To Termination of these image processing operations are the determined Characteristics then transported up the network and for classification of the product.

To Successful classification causes the action belonging to the error class one of the first comparable second cascade within the semantic Network from top to bottom. However, this now has the goal of robot actions for the individual components to correct the detected errors.

If the product goes through the decision tree, depending on the path different characteristics for the calculated individual components. In particular, these are the following Dependent on examinations from the result of previous investigations. The investigation chain is thus controlled dynamically.

The Assignment of a car to a specific error class will be for the user graphically visible directly. The values of the characteristics can be displayed become. Thus, the correct function of the testing and processing system can be followed step by step and possibly influenced by correcting the thresholds.

1.6.4 Further examples for product specific processing systems

Image AutoClassificationAngles

(See drawing 1.6.4.1)

The AreaOfInterestSegmente specify the area in which after a Edge to be searched. The Binaer segments "kantei" represent the set of edge points with their characteristics. They are each assigned to a binary image, in which for each pixel is described by the feature value 0 or 1, whether he belongs to the edge or not. The angle segment is created by the user from the Set of all edges two selected and the angle segment as subsegment assigns. The feature "anglevalue" initiates the calculation the two regression lines. These are features of the associated edges. The product finally becomes again depending on the quality classifies the measured angle and if necessary post-processed.

A analogous procedure results in the determination of distances between edge pairs.

Therefore can Optically measure products in a flexible way.

The Surveying two additional Edges only draws the graphically interactive complement of the semantic Network around two edge segments after itself.

The features of the ImagePointSetSegment and their user relationships as the lowest Ebe ne of the semantic network reflect the known data flow representations of concatenated image processing operations. In the example, for example, the Laplace operation is applied to the gray value image CRGBMark. This is expressed by a usage relationship between the RGBSetMark and the LaplaceSetMark.

Therefore finds a conventional Datenflußdarstellung of image editing operations by accustomed users at this point a starting point to the usual point of view.

Image AutoClassificationForm

(See drawing 1.6.4.2)

In This example explains how different actions are performed dependent on executed by the recognized form can be.

Furthermore the example shows the learning ability of the system.

Of the First of all, the user puts a picture of a car in the illustrated test system Form A and initiates the partial execution of the test system, here in particular the generation of the binary image for the car with the known Form A, the calculation of the center of gravity and the determination of the distance profile "autoReferenceProfileA" Serve pictures as a pattern. Therefore, its value is "frozen" by the user, i.e. this feature will not be redetermined when submitting additional images.

Analogous moves the user with a picture of the form B.

Now will the full semantic network applied to a car of unknown form. From The three distance profiles are now recalculated only the feature "autoProfile" because the other two were frozen. Building on this calculates the features "correlationFormA" and "correlationFormB". Corresponding The result is the car is now selected from the segment "autoMengeFormA" or neither. Based on this classification will turn triggered different actions.

Picture AutoClassificationAndOr

(See drawing 1.6.4.3)

The Example shows a bit like using the And and OrSegments Product classes logically linked can be.

Image AutoClassification2Tuple

(See drawing 1.6.4.4)

The Example shows how groups of components that in a certain Relationship to each other, can be selected.

1.6.5 The graphical editor

Image user interface

(See drawing 1.6.5)

Of the Editor allows interactive creation of product-specific Testing and Processing systems.

The semantic network of segment, feature and action components specifies the structure of the product, the structure, as described above the occurring error, the product classes used for classification characteristics used and the actions foreseen for processing.

The Components are identified by shape and color type-specific icons shown. The relations are directed by form and color type specific Arrows shown.

During the Creation allows the user to calculate characteristics, selection of subsegments by a segment or the execution of a Action by marking the component and activating the corresponding one To push the menu item.

He can subsequently directly check the value of a feature by activating the associated dialog. On the generated relations he can recognize which of the subsegments selected from a segment become. The effect of a threshold change is therefore immediately visible. This gives him the opportunity the test and processing system interactively and incrementally from the components build.

Of the graphical editor allows hiding components and supports the Creation of multiple views of a product-specific inspection and processing system.

In order to the user can better focus on the really essential structures of the test system focus.

The plan for the properly functioning test system can first be stored in the database of the system and later, if necessary, for the examination or processing of a product charged and carried out depending on the type of product.

The Components of the framework allow a pure for many problems graphically interactive solution. The system supports Furthermore the creation of product or company specific components a Wizzard, the code framework for generated new components. The new components enable along with the base components for subsequent problems then again a purely graphical interactive creation or modification the product-specific test systems.

1.6.6 The overall system

Picture complete system

(See drawing 1.6.6)

On Desire will be the results of testing in the database of the system stored. In this way, the graphical representation of the error trends allows.

In order to is about a quick reaction to a change in production conditions possible.

The System supported the parallel check or Editing multiple products by managing a pool of several remote inspection systems, multiple cameras and several Robot.

The System is fail-safe, as the inspection PCs with two control PCs work together. falls a control PC off, takes over the other the controller.

About suitable Actions and features can the movements and settings of the cameras depending on controlled by inspection results.

For example, can turn on the camera after detecting a specific error a close-up can be arranged.

to Detection of errors can e.g. several images with different depth of field used become.

Claims (13)

Process for image-controlled technical testing and / or technical processing of products or product parts • with first process components (M, feature components) for direct determination of the characteristics of the product to be tested and / or parts thereof, • with second process components (S, segment components) for classifying the products or parts thereof to be checked and / or processed, • with third process components (A, action components) for processing, (after) treating and / or handling the products or parts thereof and • with fourth process components (R, relationship components) for the computer-aided combination of at least two process components (M, S and / or A), characterized in that a) each process component (M, S, A, R) is graphically displayed on a display device of a computer and identifiable by a symbolic name, b) that the process components (M, S, A, R) for Ers c) that the process components (M, S, A, R) include a structure of the product, the structure of the errors that occur, the product classes, form the semantic network used for the classification and / or the actions intended for processing, which represent the concepts and their relationships that are relevant in the problem area of the user, d) that the segment components (S) comprise method steps for a classification of the to be checked and / or machining products or parts thereof, e) the classification being based on a classifying segment component (S) from a definable set of segment components (S1, S2, S3, ... Sn) to be selected and a subset (T) selected from them selected segment components (T1, T2, ... Tn) by means of relation component (R) a direct relationship the quantity of segment components (S1, S2, S3,... Sn) to be classified by the classifying segment component (S) is represented by a further segment component (SM) which is associated both with the segment components (S1, S2 , S3, ... Sn) and directly to the classifying segment component (S) by means of relation components (R), and f) wherein the selection by the classifying segment component (S) is based on an evaluation of feature components (M) for the g) that the relational components (R) specify those selected segment components (T1, T2, ... Tn) which have been selected by a classifying segment component (S) , h) that the relationship components (R) are the feature components (M) of a classifying segment specify component (S), i) that the relational components (R) specify feature components (M1, M2, ... Mn) that require a feature component (M) for its computation, j) that the relational components (R) feature components (M1, M2, ... Mn) using a classifying segment component (S) for selecting segment components to be classified (S1, S2, S3, ... Sn), k) the relation components (R) specifying further segment components (SM) representing the set of the segment components (S1, S2, S3, ... Sn) to be classified by the classifying segment component (l), l) the relationship components (R) representing the action component (A) to be executed for the classifying segment component (S) specify and m) that the relational components (R) specify the action components (A1, A2, ... An) to be triggered by an action component (A), n) during the creation or modification of the checking and processing the following method steps are carried out: complete the component system (M, S, A, R) with product-specific components, model the product structure by means of segment and relation components (S, R), model the error structure by means of segment and relation components (S, R), - picking up the features of the product or parts thereof by at least one feature component (M), - determining the defects using the segment components (S) and adjusting the thresholds (P), - modeling the defect features and the associated thresholds (P) by means of feature and relation components (M, R), - calculating the feature values by means of feature components (M) and adjusting the thresholds (P), - modeling the features of the product with feature and relation components (M, R), - modeling the error classes of the Product with segment and relation components (S, M), - Classification of the product by means of segment components (S) and adaptation n the thresholds (P), - modeling the actions with action and relation components (A, R), and - executing the edit actions and adjusting the action parameters. o) and wherein during the examination and processing of the product the following method steps are carried out: - Assignment of the action components (A) with the processing of the modeled error classes (S), - Assignment of the segment components (S) with the product classification, - Assignment of the feature components ( M) with the calculation, - recording the image to be examined by the frame grabber using feature component (M), - determining the defects by means of segment components (S), - calculating the feature values for the defects and the product using feature components (M), - classifying of the product by means of segment components (S) and - execution of the action components (A) for processing by robots. Method according to claim 1, characterized in that that the Feature components (M) process steps for the control of facilities, For example, sensors to determine the characteristics of the to be checked or To be processed product or parts thereof, such as Control of a frame grabber to determine a gray value image, or method steps for calculating features from previously determined Features include, for example, the calculation of the gradient a pixel from the previously determined gray value image, the calculation the contrast of an image area, the area of a defect area, the Number of damages a product the size of one noise or the shape of a product. Method according to claim 1 or 2, characterized that the Evaluation of the feature components (M) by means of one-dimensional or multidimensional Threshold method, using a neural network or a fuzzy classification he follows. Method according to one of claims 1 to 3, characterized that the Action components (A) Machining, processing, handling machinery etc.. or robots are. Method according to claim 4, characterized in that that the Action components (A) Procedural steps for controlling the machining, Processing, handling machines and the like. or robots, For example, required process steps for polishing or Painting damaged Products, or process steps to control other action components (A1, A2, ... to). Method according to one of claims 1 to 5, characterized that this display a screen is and that the Process components (M, S, A, R) are displayed as icons. Method according to Claim 6, characterized that the Process components (M, S, A, R) can be generated by drag and drop and can be combined with each other. Method according to one of claims 1 to 7, characterized in that different views of the semantic network can be displayed. Method according to claim 8, characterized in that that negligible Views of the semantic network can be hidden. Method according to one of claims 1 to 9, characterized that this subjected to a semantic network before each execution of a consistency check becomes. Method according to one of claims 1 to 10, characterized that state and parameter values of each process component (M, S, A, R) in the frame a dialog on the screen, for example Characteristic values, threshold values, camera or robot parameters. Method according to claim 11, characterized in that that the Parameter values by means of an input device (E), for example an editor, are specifiable. Method according to one of claims 1 to 12, characterized that the Process components (M, S, A, R) and their specific states and Parameters are storable in a memory.