Method and device for identifying a workpiece surface structure

Abstract

The invention relates to a method and a device for identifying a workpiece surface of a workpiece which preferably consists at least partly of wood, wood material, plastic or the like, said method having the steps of receiving workpiece surface data relating to a workpiece, identifying a defined workpiece surface, preferably on the basis of the received workpiece surface data and by means of a trained neural network, and determining, on the basis of the received workpiece surface data and using the trained neural network, whether there is a deviation from the defined workpiece surface, the neural network being trained using training workpiece surface data relating to a plurality of workpieces.

Description

[0001] Method and device for identifying a workpiece surface structure

[0002] Technical field

[0003] The invention relates to a computer-implemented method according to the preamble of claim 1, a computer program product according to the preamble of claim 6, a method according to the preamble of claim 7 and a device according to the preamble of claim 8 for identifying a workpiece surface structure of a workpiece which preferably consists at least partially of wood, wood-based material, plastic or the like.

[0004] State of the art

[0005] In the furniture and building component industry, workpieces, mostly made of wood, wood-based materials, plastic, or similar materials, are processed in a variety of ways, such as sawing, drilling, milling, painting, or sealing. The processed workpieces are characterized, for example, by different shapes, dimensions, colors, or surface textures.

[0006] The surface finish of machined workpieces is of particular importance in the final product, as it significantly determines the overall visual impression. For example, workpiece surfaces should ideally have a smooth structure. However, due to faulty machining or improper handling during the machining process, undesirable quality defects such as scratches, waves, dents, pores, notches, or cracks can sporadically occur on workpiece surfaces.

[0007] If such workpiece defects are clearly visible to the human eye, they are quickly detected by trained machine operators or even end users. In contrast, so-called serial quality defects are more difficult to perceive visually, meaning that these usually reach the end customer in larger quantities. Different finishes or colors of the workpieces also influence the visibility of these defects.

[0008] To reliably identify large, flat workpieces with the workpiece defects described above during the manufacturing process, the state of the art employs complex inspection units that perform an inclination-measuring method for 3D shape determination (so-called "shape-from-shading" method). This method utilizes physical relationships between the angle of inclination, surface properties, light incidence, and viewing direction to determine a curvature pattern from at least three images.

[0009] However, the acquisition costs, space requirements, and technical complexity of these systems are high, as they necessitate, for example, sophisticated lighting and camera technology, as well as conveyor units synchronized with it. A further disadvantage is the complex integration of the inspection units into existing production processes and the need for a trained and experienced operator to achieve satisfactory inspection results. In particular, individual adaptation of the system and measuring technology is required for inspecting surface structures on different workpiece variants.

[0010] Furthermore, the results of the shape-from-shading process are significantly distorted by various interfering factors such as dust particles on the workpiece surfaces, necessitating the use of additional antistatic units. Quality control of a narrow side and an edge surface, or the surface of the edge material of a workpiece, is not satisfactorily possible with the shape-from-shading method, so a separate inspection unit must be used for this purpose.

[0011] Description of the invention

[0012] The invention is based on the objective of providing a method and a device which effectively and cost-effectively identify surface defects of differently machined workpieces that are detrimental to human perception of quality during a continuous process.

[0013] This problem is solved according to the invention by a computer-implemented method according to claim 1, a method according to claim 7, and a device according to claim 9 for identifying a workpiece surface of a workpiece which preferably consists at least partially of wood, wood-based material, plastic, or the like. Particularly preferred embodiments of the invention are specified in the dependent claims.

[0014] The invention is based on the idea that optically perceptible deviations from a predefined workpiece surface exhibit a specific pattern. In other words, workpiece surface defects exhibit an optical pattern if they disturb the end user's perception of quality. Consequently, they can be detected with sufficient certainty, provided that the corresponding patterns are known to a technical system or can be detected by it.

[0015] The use of trained neural networks leads to optimal results in the identification of surface defects, since certain patterns can be learned from a large number of different workpieces with different surface structures or workpiece surface defects and subsequently used for identification, which play a role in human perception of quality.

[0016] For this purpose, a computer-implemented method according to the invention for identifying a workpiece surface of a workpiece, which preferably consists at least partially of wood, wood-based material, plastic or the like, comprises the following steps:

[0017] Receiving workpiece surface data of a workpiece;

[0018] Identifying a predefined workpiece surface, preferably using received workpiece surface data and a trained neural network; and

[0019] Determine based on the received

[0020] workpiece surface data and, using the trained neural network, determine whether there is a deviation from the predefined workpiece surface, where the neural network is trained with training workpiece surface data from a large number of workpieces.

[0021] A (predefined) workpiece surface can refer to two things. First, it refers to a surface structure that exhibits a (predefined) surface quality, such as a specific surface texture, which can be defined using corresponding surface parameters. A surface with a predefined surface texture is considered essentially "smooth." Second, a (predefined) workpiece surface can also include a specific surface decoration, pattern, color, or shape / dimensions. For example, the average roughness (Ra) of a (predefined) machined workpiece surface ranges from 0.1 pm to 10,000 pm, depending on the material, but preferably from 1 pm to 1,000 pm.A predefined workpiece surface can, for example, have a wood grain finish, particularly one resembling the natural wood structure of sawn timber, and be rectangular. A specific number of drill holes can also (additionally) define a workpiece surface.

[0022] A deviation from the predefined workpiece surface is understood to mean, for example, the presence of workpiece surface structures that deviate from a substantially (predefined) smooth workpiece surface, in particular those that interrupt a substantially smooth workpiece surface and thereby form protrusions and / or indentations. Such deviations include, for example, scratches, waves, dents, pores, notches, cracks, grooves, drill holes, joints, or notches (also referred to as surface defects). The substantially (predefined) smooth workpiece surface structure can also be interrupted by particles or particle aggregations adhering to the smooth workpiece surface structure, such as dust, chalk marks, or chips. Deviations from a substantially smooth workpiece surface are preferably deviations that are visually perceptible to the human eye.However, these are not limited to this, but can also refer to deviations that can only be detected with an optical recording device, such as a camera.

[0023] The use of trained neural networks in the method according to the invention has the advantage that workpiece surface data, in particular image data of a workpiece surface, can be evaluated in which certain data patterns relating to workpiece surfaces are found. In other words, not only can (identically) previously known workpiece surfaces be identified, but also unknown workpiece surfaces whose associated data exhibit a data pattern on which a neural network can be trained to recognize. For example, several workpieces may have different wood decor structures, which, however, exhibit a higher-order, identifiable general (wood) pattern.

[0024] Furthermore, a (predefined) workpiece surface can be identified even if surface defects are already present. In other words, the pattern of a (predefined) workpiece surface can be recognized or identified with sufficient certainty, provided that a (basic (data)) pattern is (still) present. Similarly, the surface defects described above also exhibit an identifiable (data) pattern that a neural network can recognize. This allows not only the identification of (identical) previously known workpiece surface defects, but also the identification of unknown workpiece surface defects whose associated data exhibits a data pattern on which a neural network can be trained to recognize them.

[0025] In the method according to the invention, the detection sensitivity of the neural network is adjustable via a so-called defect size, which defines a surface defect degree. Based on the defect size, surface defects can be determined with a specific sensitivity. In other words, the neural network can recognize, based on the data patterns, the degree or extent of a surface defect, or how pronounced it is. The degree or extent of a surface defect correlates, for example, with its three-dimensional dimensions or color deviation. This results in a flexible yet easy-to-use identification method whose detection sensitivity with respect to surface defects is adjustable.

[0026] Training the neural network with training workpiece surface data from a large number of workpieces offers the advantage that the respective neural networks can be individually adapted to specific workpiece surfaces (defects) that occur with a particular machining technique or in a specific machining system. This means that the neural networks can be specifically trained only for workpieces in a particular machining system. However, workpiece properties or surface defects that occur in a variety of (different) machining systems can also be taken into account during training.

[0027] The described method is particularly advantageous for machined workpieces, especially plate-shaped ones, which consist at least partially of wood, wood-based materials, plastic, or the like. The workpiece properties or surface structures of these workpieces exhibit corresponding data patterns that can be advantageously captured using a neural network. Furthermore, these materials are present in sufficient number and variation in the relevant production facilities, allowing for particularly efficient training of the neural networks.

[0028] In a preferred implementation, the trained neural network comprises a trained neural detector network, specifically for identifying a predefined workpiece surface, and a trained neural inspector network, specifically for determining deviations from the predefined workpiece surface, which are trained independently of each other. This allows for particularly efficient training of each neural network, as only data necessary for identifying the predefined workpiece surface or deviations from it are considered. Separate training requires less computing power than combined training. Consequently, a CPU with lower processing power can be used for training.Furthermore, two separately trained neural networks offer the possibility of performing, for example, the identification process of the predefined workpiece surface and the determination process for deviations from the predefined workpiece surface separately. This also reduces the computing power and time required for each respective process.

[0029] Preferably, the training of the neural detector network can be carried out using training identification workpiece surface data that is compressed, preferably 2x to 32x, and more preferably 8x. This data relates in particular to surface decorations, surface patterns, surface colors, surface details, or surface shapes / dimensions. Data compression allows for particularly efficient training of the detector network. That is, the training is limited to data that is only necessary for determining the workpiece surface. Consequently, less computing power is required than with uncompressed data.

[0030] Even more effectively, the training of the neural inspector network can be performed using training-identification workpiece surface data. This data relates specifically to deviations from a predefined workpiece surface as described above and includes, for example, scratches, waves, dents, pores, notches, cracks, grooves, drill holes, joints, notches, particles, particle aggregations, dust, chalk marks, or chips. Since the training-identification workpiece surface data is not compressed, the neural inspector network can be trained particularly efficiently. That is, the training considers a large dataset, which is advantageous for the particularly precise identification of workpiece surface structures. In this way, the neural inspector network is trained with a particularly comprehensive dataset.

[0031] In a preferred case, the workpiece surface data can be assigned to a narrow face, an edge surface, and / or a front and / or back surface of a plate-shaped workpiece. This allows the narrow face, edge surfaces, and front and / or back surfaces of the workpiece to be identified and potential deviations from a predefined surface to be determined. This enables all surfaces of a machined workpiece to be checked for defects in a uniform procedure. It is understood that "workpiece surface data" also includes corresponding training workpiece surface data for a narrow face, an edge surface, and / or a front and / or back surface of a large number of workpieces.An expert understands an edge surface to be the surface of an edge material that is applied to a narrow face of a workpiece using an edge banding machine. The workpieces can be, but do not have to be, flat; they can also have other three-dimensional shapes. A computer program product according to the invention comprises commands which, when the program is executed by a computer, cause it to carry out the steps of the computer-implemented method described above. The computer program product can be flexibly installed on any process computer of a production plant and executed there.

[0032] A method according to the invention for identifying a workpiece surface of a workpiece, which preferably consists at least partially of wood, wood-based material, plastic or the like, comprises the following steps:

[0033] Simultaneous exposure and capture of a first workpiece section with an exposure unit and a camera;

[0034] Generating workpiece surface data for the first workpiece section of the workpiece;

[0035] Conveying the workpiece with a conveying unit to expose and capture a second section of the workpiece or to generate workpiece surface data for the second workpiece section, wherein the workpiece can be conveyed at a variable conveying speed, which in particular depends on a conveying speed of a through-feed machine, especially during a through-feed process;

[0036] From carrying out the computer-implemented procedure described above with the generated workpiece surface data.

[0037] The described process steps provide a method that can be flexibly integrated into an existing machining process. In other words, the method is flexibly adaptable to the cycle times of upstream and downstream machining / production machines / systems, as the detection and exposure timing of the individual workpiece sections can be varied. The conveying unit can, in particular, be the conveying unit of a continuous feed machine, especially a continuous feed machining center.

[0038] An inventive device for identifying a workpiece surface of a workpiece, which preferably consists at least partially of wood, wood-based material, plastic or the like, comprises: a camera, preferably only one, which is configured to capture a section of a workpiece and generate workpiece surface data; an exposure unit, preferably only one, in particular a bar light for exposing the section of the workpiece; and a process unit, which is configured to carry out one of the methods described above.

[0039] The device according to the invention has the advantage that only a camera and an exposure unit are required. This gives the device a smart setup that can be installed quickly and cost-effectively into an existing production plant. It is particularly advantageous to install the camera and exposure unit in the vicinity of a conveyor unit, which can be part of an existing production plant. An area around the conveyor unit means, for example, that the camera or the exposure unit is located at a distance of 1 cm to 10 m, preferably 1 cm to 1 m, from the conveyed workpiece. The distance between the camera and the exposure unit and the conveyed workpiece can be the same or different.Furthermore, it is conceivable that the device according to the invention itself may have a conveying unit that preferably conveys plate-shaped workpieces, which preferably consist at least partially of wood, wood-based material, plastic, or the like. In addition, it is possible that the device may, for example, have exactly two, three, four, five, or six cameras and a corresponding number of exposure units in order to capture and expose a corresponding number of surface faces and / or edge faces of a preferably plate-shaped workpiece.

[0040] Preferably, the camera and / or the exposure unit can be configured to capture or expose a narrow side, an edge surface, and / or a front and / or back surface of the workpiece. This allows all surfaces of a machined workpiece to be captured and checked for surface defects using a single, compact, and easily installed device.

[0041] In a preferred embodiment, the exposure unit can expose the surface of the workpiece with a radiation angle of less than 90 degrees, preferably less than 45 degrees, and more preferably 10 to 20 degrees. This allows adverse radiation effects to be reduced (to a minimum).

[0042] Brief description of the drawings

[0043] Further features and advantages of the method and the device will become apparent from the following description of embodiments with reference to the accompanying drawings. Of these drawings, the following is shown:

[0044] Fig. 1 a flowchart of an implementation form of a publicly disclosed computer-implemented method; Fig. 2 a schematic training example of a neural network of a publicly disclosed method;

[0045] Fig. 3 shows a flow diagram of an implementation form of a disclosed process; and

[0046] Fig. 4 shows a schematic view of an embodiment of a device according to the disclosed information for identifying a workpiece surface of a workpiece;

[0047] Description of execution forms

[0048] Identical reference symbols shown in different figures denote identical, corresponding, or functionally similar elements. It is apparent to a person skilled in the art that individual features described in different embodiments can also be implemented in a single embodiment, provided they are not structurally incompatible. Likewise, different features described within a single embodiment can also be provided individually or in each subcombination in several embodiments.

[0049] Figure 1 shows a flow diagram of an embodiment of a computer-implemented method according to the invention. This method can, for example, be carried out with a process unit 8, which can also be used to control and regulate machining processes of workpieces 3 in corresponding through-feed machines. Thus, the method according to the invention can be integrated into existing processes without additional hardware. It is understood that the method can also be carried out on a stand-alone process unit 8, which can be networked with other process units of workpiece machining centers.

[0050] In a first step S1, workpiece surface data of a workpiece 3 are received. This data can be in binary format or a data format known to those skilled in the art and is transmitted to the process unit 8 via an interface. Based on this data, a predefined workpiece surface 3 is identified in a second step S2. A predefined workpiece surface 3 is, for example, a surface of a workpiece 3 with a wood decor featuring a natural wood structure of sawn timber, wherein the workpiece 3 is rectangular and has a mean roughness Ra of, for example, 0.1 pm to 10,000 pm, preferably 1 pm to 1,000 pm, and particularly preferably 5 pm. In the present example, a predefined workpiece surface 3 can also be square, green in color, and have a mean roughness Ra of, for example, 0.1 pm to 10,000 pm.000 pm, preferably 1 pm to 1000 pm, particularly preferably 5 pm, and comprise five boreholes 11 .

[0051] Identifying a predefined workpiece surface using received workpiece surface data means assigning this data to a specific workpiece surface (data) category. This can also include the received workpiece surface data consisting solely of information about a predefined workpiece surface (data) category. In other words, the received workpiece surface data is data relating to a (predefined) workpiece surface (data) category, which is specified, for example, by a user.

[0052] Preferably, the identification is carried out using a neural network. This has the advantage that the received workpiece surface data can be evaluated by the trained neural network to recognize a data pattern. In this way, it is possible to identify a large number of different workpieces using the method according to the invention or to assign them to a workpiece surface category. For example, the same predefined workpiece surface "wood decor" can be identified for several workpieces, even if the several workpieces have a wood decor with a natural wood structure of sawn timber, whose grain and / or number of knots, knot size, and / or knot design differ. This is possible because the respective wood structure of the wood decor has a common data pattern, which is recognizable by the neural network.It is also conceivable that a predefined number of drill holes 11 can be identified in a workpiece surface. The neural network can particularly preferably identify data structures that define the drill holes 11.

[0053] Knotholes in the wood decor distinguish it.

[0054] In a third step, S3, it is determined whether a deviation from the predefined workpiece surface exists. In other words, the trained neural network identifies surface defects that extend in three dimensions along, from, and into the workpiece surface, such as scratches, waves, dents, pores, notches, cracks, grooves, drill holes, joints, notches, dust particles, chips, etc. Each of these surface defects exhibits a characteristic data pattern that can be determined by the neural network. This is advantageous because it allows both the identification of a predefined workpiece surface and the determination of surface defects, either additionally or separately.

[0055] Furthermore, a defect size can be set, particularly for determining deviations from a predefined workpiece surface using the neural network. This size defines the degree of the surface defect, its extent, and its severity. In other words, the defect size defines different data patterns for a surface defect that can be recognized by the neural network. A specific rating can then be assigned to each surface defect, correlating with the optical perception capabilities of the human eye.

[0056] Figure 2 shows a schematic training example of a neural network of a disclosed method. This includes separate and independent training of a neural detector network and a neural inspector network. Both the neural detector network and the neural inspector network are pre-trained neural networks which can be specifically trained for the application according to the invention using the training data described later. For example, the software "HALCON" (MVTec Software GmbH) can be used to train the two neural networks.

[0057] The independent and separate training method ("deep learning") of two specially selected neural networks proves particularly advantageous for identifying the surface of a workpiece consisting at least partially of wood, wood-based material, or plastic, compared to classical "machine learning" or a combination of "deep learning" and "machine learning," since the trained neural networks achieve a better identification and determination rate. At the same time, the training can be specifically adapted for identifying a predefined workpiece surface and determining deviations from a predefined workpiece surface, enabling particularly efficient training; that is, the required computing power can be reduced for the application according to the invention.For training purposes, a large number of workpieces are used, consisting at least partially of wood, wood-based materials, or plastic. Training with at least one hundred workpieces is particularly advantageous, especially those exhibiting different characteristic surface defects that occur during a specific machining process. For this purpose, workpieces can be selected from a faulty machining process using a machine tool or deliberately inserted manually or mechanically into a tool surface.

[0058] The training process is illustrated by way of example in Figure 2, which shows a workpiece 3 having a wood decor with a natural wood structure of sawn timber, wherein the workpiece is rectangular and has a mean roughness Ra of, for example, 0.1 pm to 10,000 pm, preferably 1 pm to 1,000 pm, particularly preferably 5 pm. In addition, the workpiece comprises a borehole 11 and a surface defect 12 in the form of an optically visible indentation 12.

[0059] For training the neural detector network, the workpiece surface of the workpiece described above is captured, and its data is provided to the training software described above in the form of image data. The image data represents, for example, an image 13 with a resolution of 4096 pixels (width) by 1024 pixels (length), which depicts a workpiece surface 3 captured in a (production) environment 14. The resolution of image 13 is subsequently reduced, particularly by means of the training software, for example, to a resolution of 512 pixels (width) by 128 pixels (length). In other words, the image data is compressed. The compressed image data is then used for training the neural detector network, i.e., provided to the training software described above.In the present example, the neural detector network will identify a data pattern in the compressed data that, according to the operator, is assigned to a workpiece with a wood finish, where the workpiece is rectangular and has the average roughness described above. In other words, the neural detector network is trained to identify this predefined workpiece surface. This also includes the neural network being trained to identify the corresponding outlines of the workpiece 3 and to separate them from the (production) environment, i.e., to extract the workpiece surface image 3 from the (production) environment 14. It is understood that a large number of workpieces with different wood finishes, or other surface finishes, shapes, etc., are provided to the neural detector network in the same way to ensure sufficient data.

[0060] To achieve a training effect.

[0061] It is important to note that neural networks require a constant input size with respect to the captured image data, meaning the captured image must be as large as possible, i.e., for example, 4096 pixels (width) by 10000 pixels (length). The width of the captured image is limited to 4096 lines by the sensor chip of camera 1, but not to a specific length. Since the images in this case are always captured with a maximum length of 10000 pixels, images with a length of less than 10000 pixels must be padded to this length, i.e., the image data must be supplemented with zeros to provide the neural network with a constant input size.

[0062] The original (uncompressed) image data is then used again to train the neural inspector network with image data at maximum resolution. However, the extracted image of workpiece surface 3 has a dynamic width and length. To provide the neural network with a constant input size for training, it is trained with image data whose constant input size is, for example, 256 pixels by 256 pixels. The 256-pixel by 256-pixel image data forms image squares, which are overlaid as a partially overlapping image-square grid onto the image of workpiece surface 3, thus enabling the entire image of workpiece surface 3 to be represented (despite the dynamic width and length).

[0063] In the corresponding individual images (with a constant input value), surface defects can then be identified and manually marked by an operator using the training software, allowing the neural network to learn the corresponding data pattern of surface defect 12. In the present example, the borehole 11 can also be marked and its data pattern used for training. The image data of the overlapping grids with the constant input value are then reassembled using the training software in such a way that the workpiece surface is represented with the corresponding location of surface defect 12 and borehole 11.

[0064] Figure 3 shows a flow diagram of an embodiment of a disclosed process with a first step VI in which a first workpiece section of workpiece 3 is simultaneously exposed and captured by an exposure unit 2 and a camera 1. In a second step V2, workpiece surface data for the first workpiece section of workpiece 3 are generated. In a subsequent third step V3, the workpiece 3 is conveyed by a conveyor unit 4 to expose and capture a second section of workpiece 3. In other words, the workpiece is conveyed to repeat the first step VI, but to capture and expose a second section of the workpiece.Depending on the conveying speed of the conveying unit 4, which is preferably in a range of 1 m / min to 500 m / min, preferably 20 m / min to 100 m / min, more preferably 60 m / min, and the length of the workpiece to be detected, steps VI and V2 are repeated until the workpiece has been completely detected and corresponding workpiece data has been generated.

[0065] The conveying speed of the conveying unit, within a range of 1 m / min to 500 m / min, preferably 20 m / min to 100 m / min, can be variably controlled. This means that the workpiece is conveyed at a variable speed while steps VI and V2 are performed (repeatedly). This is particularly advantageous if the conveying speed can be varied depending on the conveying speed of another conveying unit of a continuous-feed machine during a continuous process. This allows the cycle time of upstream and downstream processing equipment in the production process to operate independently of the inventive method. In other words, the cycle time of the exposure or detection (and identification) of the workpiece (or the determination of possible deviations) is determined by the cycle time of a processing machine or an overall processing process.The beginning and end of (the repeating) steps VI and V2 can be determined by means of a part sensor 10, which is set up to indicate the presence of the workpiece in the exposure and detection range of the process unit 8.

[0066] In a fourth step, V4, the computer-implemented procedure described above is then executed; that is, a predefined workpiece surface is identified, and deviations from the predefined workpiece surface are determined. If a deviation from the predefined workpiece surface is detected, a message can be sent to an operator and / or the defective workpiece can be directly excluded from the process.

[0067] Figure 4 shows a schematic view of an embodiment of a device according to the disclosed disclosure for identifying a workpiece surface of a workpiece 3. This device comprises only a single camera 1, in particular an area scan camera with a block-scan mode or a commercial line scan camera. This camera is arranged above a workpiece 3, preferably at a distance of 1 cm to 10 m, more preferably 1 cm to 1 m from the workpiece, in order to capture at least a section of the surface of the workpiece 3, which in this case is a section of a first front or back surface of a plate-shaped workpiece 3. Additionally, only one exposure unit 2 is provided above the workpiece, preferably at a distance of 1 cm to 10 m, more preferably 1 cm to 1 m, from the workpiece, in order to expose the first front or back surface of the plate-shaped workpiece 3. For this purpose, the exposure unit 2 can be a bar light or a similar type of light source.This configuration is particularly cost-effective. In Figure 4, the surface of the workpiece and the directed light source rays of the exposure unit 2 form a radiation angle of 10 to 20 degrees. In this way, the emission effects can be kept low. The exposure unit has, for example, an illuminance of approximately 1000 W / m². 2 .

[0068] Although not shown in Figure 4, an additional second camera and / or exposure unit can be provided to capture or expose a second front or back side of the workpiece 3. However, a configuration is provided in which two cameras are arranged for capturing a first and / or second front or back side of the workpiece 3. This can increase the precision of the workpiece 3 capture. By arranging an (additional) camera, a narrow side, an edge surface, or a surface of an edge material can also be captured. The camera can be arranged such that any narrow side or edge of a plate-shaped workpiece can be captured. It is also possible to arrange several cameras for capturing several different narrow sides or edge surfaces.Similarly, a corresponding exposure unit for exposing this narrow side (n) or edge (n) can be provided, but is not required. With this configuration, comprehensive acquisition and inspection of all workpiece surfaces of a plate-shaped workpiece is possible.

[0069] Figure 4 also shows a conveying device 7, preferably in the form of a continuous conveyor belt, as part of the device according to the invention for conveying the workpiece 3 at a conveying speed. The conveying device 7 can also be a controlled component of a processing or through-feed machine. In the latter case, the device according to the invention, in particular the camera 1 and the exposure unit 2, is configured to be installable on the controlled conveying device 7 of the processing or through-feed machine and to capture workpieces in such a way that the corresponding capture data can be processed in the method according to the invention.

[0070] As shown in Figure 4, the device according to the invention comprises a process unit 8, which carries out the methods according to the invention. For this purpose, the workpiece data captured by the camera 1 are transmitted to the process unit 8 via an interface. The process unit is also configured to control or regulate the conveyor 7. That is, the process unit controls, for example, the conveying speed of the conveyor 7 as a function of a conveying speed / cycle time of a processing or through-feed machine that is upstream and / or downstream of the conveyor 7.

[0071] In the event that the conveyor 7 is part of a processing or throughput machine in a production plant and is controlled or regulated by the process units therein, the process unit 8 of the device according to the invention is configured to communicate with the process units of the processing machines in order to control or regulate the detection and exposure timing of the surface workpiece section by means of the camera 1 or exposure unit 2. The detection and exposure timing can also be carried out, for example, by means of an encoder 9. In addition, a part sensor 10 can be part of the device according to the invention, which is configured to determine a start and an end of the workpiece detection and exposure. For example, the part sensor 10 can be a light barrier or a push-button switch.

[0072] Preferably, the device according to the invention is configured to expose and capture workpieces per minute at a conveying speed of at least 10, preferably 10 to 20, more preferably 15 to 20 workpieces per minute. Preferably, the device according to the invention is configured to expose and capture workpieces per minute at a conveying speed of at least 20 m / s, preferably 20 to 50 m / s, more preferably 40 to 50 m / s. Preferably, the device according to the invention is configured to capture and expose workpieces up to a maximum length of 3100 mm and / or width of 1600 mm, and to execute the inventive method with the captured data.

Claims

REQUIREMENTS 1. Computer-implemented method for identifying a workpiece surface of a workpiece which preferably consists at least partially of wood, wood-based material, plastic or the like, comprising the following steps: Receiving workpiece surface data of a workpiece (Sl) ; Identifying a predefined workpiece surface, preferably using the received workpiece surface data and a trained neural network (S2); and Determine, using the received workpiece surface data and the trained neural network, whether there is a deviation from the predefined workpiece surface (S3) , wherein the neural network is trained with training workpiece surface data of a large number of workpieces.

2. Computer-implemented method according to claim 1, wherein the trained neural network comprises a trained neural detector network, in particular for identifying a predefined workpiece surface, and a trained neural inspector network, in particular for determining a deviation from the predefined workpiece surface, which are trained independently of each other.

3. Computer-implemented method according to claim 2, wherein the training of the neural detector network is carried out on the basis of, preferably 2-fold to 32-fold, more preferably 8-fold, compressed training identification workpiece surface data.

4. Computer-implemented method according to claim 3, wherein the training of the neural inspector network is performed on the basis of training-determination workpiece surface data.

5. Computer-implemented method according to one of the preceding claims, wherein the workpiece surface data are assigned to a narrow surface, an edge surface and / or a front and / or back surface of a plate-shaped workpiece.

6. Computer program product comprising instructions which, when the program is executed by a computer, cause it to perform the steps of the method according to any of the preceding claims.

7. Method for identifying a workpiece surface of a workpiece (3) , which preferably consists at least partially of wood, wood-based material, plastic or the like, comprising the following steps: Simultaneous exposure and capture (VI) of a first workpiece section of a workpiece (3) with an exposure unit (2) and a camera (1) ; Generating workpiece surface data (V2) for the first workpiece section of the workpiece (3) ; Conveying the workpiece (V3) with a conveying unit (7) to expose and capture a second section of the workpiece (3) or to generate workpiece surface data for the second workpiece section of the workpiece (3), wherein the workpiece can be conveyed at a variable conveying speed, which is in particular dependent on the conveying speed of a continuous feed machine, especially during a throughput process depends; Performing the method according to one of claims 1 to 5 with the generated workpiece surface data (V4) .

8. Device for identifying a workpiece surface of a workpiece (3), which preferably consists at least partially of wood, wood-based material, plastic or the like, comprising: a, preferably only one, camera (1) which is configured to capture a section of a workpiece and generate workpiece surface data; an, preferably only one, exposure unit (2), in particular a bar light for exposing the section of the workpiece (3); and a process unit (6) which is configured to carry out the method according to one of claims 1 to 5 or 7.

9. Device according to claim 8, wherein the camera (1) and / or the exposure unit (2) is configured to capture or expose a narrow side, an edge surface and / or a front and / or back surface of the workpiece (3).

10. Device according to claims 8 or 9, wherein the exposure unit (2) exposes the surface of the workpiece (3) with a radiation angle of less than 90 degrees, preferably less than 45 degrees, more preferably 10 to 20 degrees.

Images