A system and method for controlling the quality of panels within a production line

Abstract

The invention relates to a system and method for controlling the quality of panels within a production line, by making use of a sensor configured to capture and / or determine at least one characteristic of at least one panel displaced within the production line and a processing unit configured to receive sensor data related to at least one characteristic of the panel from said at least one sensor which is processed such that the system can provide an operational instruction as output.

Description

[0001] A system and method for controlling the quality of panels within a production line

[0002] The invention relates to a system for controlling the quality of panels within a production line. The invention also relates to a method for controlling the quality of panels within a production line.

[0003] Production of decorative panels, such as for example floor panels, wall panels, ceiling panels and / or building panels are produced in automated production lines which have a high production capacity. Due to the high speed at which these production lines operate, it is a challenge to provide inline quality control at all aspects of the production. Defects are therefore often only identified after a batch of panels has been produced. Current scanning systems further face the challenge in the quality control of a decorative panel of the existence of “hidden” or “unreachable” features, such as is the case with interlocking mechanisms that feature intricate designs that are usually only measurable by means of destructive analysis, specifically lock systems with a tongue-and-groove design with cavities extending into the panel’s edge or body. Such hidden features cannot be mapped by a single viewpoint and are mapped according to the invention by utilizing at least one further scan from at least one different angle, preferably at least perpendicular to the board in at least one plane, most preferably at least in the plane and facing the machine direction of the double-end-tenoner by which the interlocking mechanism is applied, thereby capturing the shape of the hidden feature. It is conceivable that supplementary equipment can be used in support of 3D scanning equipment, such as automated feeler gauges, and / or sensors. It is conceivable that where parts of the 3D model are missing information, simulation software may be utilized to predict at least partially the dimensions of hidden areas based on known properties of the panel and / or feature being scanned. Currently, the panel profiles are manually checked by cutting a portion of a panel and visually inspecting its dimensions. This process takes a considerable amount of time which delays discovering a defective panel while administering invasive processes.

[0004] As production lines in flooring can reach speeds of up to 100m / min, the high volume of boards produced on a production line make it difficult to scan each and every board coming from the production line, as depending on the specific scanning technology used, the scan time for a single board may be several minutes or longer depending on the scanning size or area and / or quantity of parameters to scan for.

[0005] Since effective quality control is a critical component of any production process, as defects can lead to increased waste, decreased efficiency and customer satisfaction there is always a need to further optimize this process. Traditional quality control methods often rely on manual inspection, which can be timeconsuming, expensive, and prone to errors. Furthermore, traditional quality control methods may detect defects only after they have occurred, leading to waste and rework. A further goal of the invention is to provide a system and method for controlling the quality of panels within a production line for decorative panels which overcomes at least part of the abovementioned drawbacks.

[0006] The invention provides thereto a system for controlling the quality of (decorative) panels within a production line, said production line comprising at least one conveyor for displacing at least one decorative panel within the production line, said system comprising:

[0007] - at least one sensor configured to capture and / or determine at least one characteristic of at least one panel displaced within the production line; and / or

[0008] - at least one processing unit configured to: o receive sensor data related to at least one characteristic, in particular at least one determined characteristic of the panel from said at least one sensor; and / or o identify if the panel comprises one or more visual identifiers, in particular one or more defects optionally based on the received sensor data and / or on at least one determined characteristic of the panel, in particular by using at least one machine learning model; and / or o determine a qualitative acceptability value for at least one and preferably each identified visual identifier, in particular at least one and preferably each identified defect, optionally based on the received sensor data, in particular by using at least one machine learning model; and / or o provide at least one operational instruction in particular to the production line in case the limit value for qualitative acceptability of an identified visual identifier, in particular an identified defect is exceeded. The system according to the present invention is designed to improve production quality by automatically identifying defects, correcting defects as they occur, as well as providing real-time feedback to production operators or automated production systems to make immediate corrective actions. The combination of at least one sensor which is configured to determine at least one characteristic of at least one panel displaced within the production line and at least one processing unit according to the invention enables the characteristics of a panel or board can be linked to visual identifiers, such as defects, during the production process. The sensor according to the invention is in particular configured to capture, generate, and / or provide sensor data related to at least one determined characteristic of the panel. The sensor data is then processed and analysed by the at least one processing unit, which is configured to identify defects and generate real-time alerts, or operational instructions, to for example production operators and other stakeholders. However, it is also conceivable that the operational instructions are sent directly to at least one operating system of the production line. The at least one processing unit preferably applies at least one machine learning model. It is also conceivable that at least one processing unit is an Al and / or self-learning processing unit. The processing unit is preferably capable of learning from previous defect occurrences, making recommendations for process improvements to prevent similar defects from occurring in the future, and / or adjusting the production parameters directly.

[0009] This invention then provides a system and method for checking and controlling the quality of panels within a production line for decorative panels by using one or more image sensors deployed in the said production line. This invention further uses at least one machine learning model to identify unexpected or unknown defects by using the image or visual data acquired by the said image sensors. This is contrary to the use of loT-based sensor analysis of defects which is costly and complex due to the plurality of sensors involved. The at least one machine learning model is configured to analyze sensor data and / or determine a qualitative acceptability value for at least one and preferably each identified visual identifier, in particular at least one and preferably each identified defect. It is also conceivable that at least one machine learning model is configured to determine at least one quality classification. At least one operational instruction can be provided in case the at least one quality classification of an identified visual identifier falls outside the scope of a threshold quality classification. A quality classification can be used alternatively or in addition to at least one qualitative acceptability value.

[0010] The system according to the present invention utilized scanning and / or imaging technology of at least one sensor to capture detailed and accurate data such as images of at least part of the visible surfaces of the panels, enabling the identification of even the smallest defects. At least one characteristic of the panel can be determined by at least one sensor by capturing at least one image. Within the context of the present invention, at least one visual identifier can be a defect, or vice versa. Non-limiting example of visual identifiers or defects are for example colour deviations, gloss deviations, uneven gloss or matte level, relief deviations, bulges, scratches, warping, missing paint, shiny spots, incorrect print design, unsynchronized embossing with the printed design, incorrect dimensions, cracks, missing or broken parts, undesired cavities, and / or combination thereof. It is also possible that the irregularities to the side edges of the panel with or without coupling parts are classified as visual identifier or defect.

[0011] At least one processing unit is configured for co-action with at least one sensor and preferably with multiple sensors, if applied. At least one processing unit is preferably directly or indirectly connected to the at least one processing unit. It is for example possible that the processing unit and sensor(s) are wirelessly connected. When it is referred to a processing unit, also a processor can be meant or referred to. Within the context of this invention, the terms are interchangeable.

[0012] The operational instructions as provided by the at least one processing unit according to the present invention can be real-time operational instructions. It is also possible that at least one operational instruction comprises an alert, a warning and / or an advice. At least one and preferably multiple operational instructions can be sent to for example production operators and other stakeholders. However, it is also conceivable that the operational instruction(s) are send directly to the operating system of the production line.

[0013] The system according to the present invention enables improving the production quality by automatically identifying defects, optionally correcting defects as they occur, as well as providing real-time feedback to production operators or automated production systems to make immediate corrective actions. This practice adheres to quality management best practices. By continuously monitoring and analysing production data in real-time, the system can identify trends, correlations and possible root causes for the deviations that may not be immediately apparent to production operators, allowing for continuous improvement of the production process of, for example, decorative panels. It is conceivable that the system can implement direct or indirect corrective actions in real-time to correct defects and / or to prevent similar defects from occurring in the future. Overall, the present invention provides a highly effective quality checking system or mechanism for the production of flooring, wall, ceiling and / or building panels, that is both efficient and cost- effective.

[0014] In a possible embodiment, the at least one processing unit is configured to identify at least one feature in the panel and / or of the panel and the at least one processing unit is preferably configured to provide an at least one operational instruction based on the determined feature. The operational instructions can for example be directed to the production line and / or to an operator thereof. At least one operational instruction can for example be directed to the operating system of the production line. It is possible that at least one setting of the operating system of the production line is adjusted based on at least one operational instruction. It is for example possible that at least one feature of the panel is a colour, a colour change, an irregularity of the surface of the panel, a scratch, misalignment of the decorative print, a warped and / or buckled surface, a stain, delamination, an uneven thickness and / or combinations thereof. It is also possible that at least one feature is chosen from the group of interlocking mechanism accuracy, cut smoothness, embossing or texture design, embossing accuracy, embossing depth, board thickness, length or width, board size, bevel size, squareness, straightness, laminated layer thickness, thickness distribution, laminated layer material type, backing or pre-attached underlayment thickness and / or accuracy, or damages such as but not limited to cracks, breakages, holes, and the like.

[0015] The system according to the invention is, in particular, configured to use in combination with a production line for (decorative) panels. The production line in particular comprises at least one conveyor for displacing at least one decorative panel within the production line. The conveyor can, for example, comprise a belt and / or at least one pair of rollers. The production line preferably comprises at least one module, preferably a plurality of modules configured to at least partially process a decorative floor, wall, or ceiling panel. At least one module is preferably chosen from the group of a ripsaw, a punch, a digital printer, a digital embossing printer, a press, an extruder, a laminator, a heater, a cooler, a conveyor, a coating applicator, a UV lamp, a LED lamp, an excimer lamp, a resin applicator, an embossing applicator, a foil texture applicator, a double end tenoner, a programmable logic controller (PLC), or any combination thereof. In a possible embodiment, the system according to the invention may comprise at least one module. The system according to the invention may be integrated within a production line. It is also possible that the production line forms part of the system, or vice versa.

[0016] At least one sensor may be configured to capture at least one image of at least part of at least one panel. Hence, in a possible embodiment of the system according to the invention at least one sensor is configured to capture at least one image of at least part of at least one panel displaced within the production line and that the at least one processing unit is configured to receive sensor data related to at least one image captured by the at least one sensor. The at least one processing unit may further be configured to determine and / or identify if the panel comprises one or more visual identifiers, in particular one or more defects, based on the sensor data. Subsequently, the at least one processing unit may be configured to determine a qualitative acceptability value and / or quality classification for at least one and preferably each identified visual identifier, in particular at least one and preferably each identified defect preferably by using at least one machine learning model and / or to provide at least one operational instruction in case the limit value for qualitative acceptability of an identified visual identifier, in particular an identified defect is exceeded and / or is the determined quality classification is falls outside the scope of a threshold quality classification. A limit value for qualitative acceptability of an identified visual identifier, in particular an identified defect, can for example be maximum a colour deviation, a maximum gloss deviations, a maximum bulge height, a maximum scratch depth, a maximum degree of warping and / or combination thereof.

[0017] It is beneficial if at least one sensor is configured to capture at least one image of at least part of at least one panel. Such image can provide relevant sensor data for the at least one processing unit for determining the presence of visual indicators or defects. Images consist of pixel data, with each pixel encoding colour information typically represented as RGB values. Hence, a non-limiting example of sensor data is pixel data. It is also possible that the images captured by at least one sensor are contour images. Further, images often contain metadata detailing information such as camera settings, location, and capture time, facilitating organization and analysis. The at least one processor can be configured to extract visual identifiers like edges, shapes, textures, and colours from images. The at least one image, and preferably a plurality of images, as provided by at least one sensor can form a data source for the at least one processing. Based upon at least one captured image at least one characteristic of at least one panel can be determined. It is also conceivable that at least one sensor is configured to capture image and / or video data. It is also possible that 3D imaging data is applied.

[0018] At least one sensor may be an optical sensor. Preferably, the at least one sensor comprises at least one camera, in particular at least one optical camera. The use of at least one optical sensor and / or optical camera enables capturing high-resolution images which can be applied for inspection purposes. The system according to the system may also comprise multiple sensors. It is possible that the system comprises multiple optical sensors. The use of multiple sensors can increase the coverage, reliability and / or accuracy of the determined and / or captured characteristics of the panel. Further, combining sensor data from multiple sensors could contribute to enhanced image processing, such as an enhanced resolution and / or the ability to provide a 3D image of at least part of a panel. It is possible that the at least one processing unit analyzes data from multiple sensors possibly in combination with data obtained from the production line such as system input parameters. The at least one processing unit and / or the at least one machine learning model may be configured to determine correlations and / or causations between specific production parameters and the occurrence of defects. Based on this analysis, it is then conceivable that recommendations are determined for changes to the production parameters to prevent similar defects from occurring in the future. For example, if the system identifies that a certain temperature, pressure setting or formulation is causing a specific type of defect, it could adjust those settings automatically or provide warnings or recommendations to the human operator to prevent the defect from reoccurring. Over time, the processing unit and / or machine learning model could use this feedback loop to continuously learn from past data and improve the accuracy of its recommendations, or operational instructions, allowing the production process to become more efficient and less prone to defects.

[0019] It is further possible that at least one sensor comprises or is formed by an optical sensor, optical camera, infrared sensor, ultrasonic sensor, temperature sensor, humidity sensor, position sensor, tilt sensor, color sensor, proximity sensor, capacitive sensor, load cell, or any combination thereof. It is possible that a combination of sensors is applied. The sensors may form a sensor (sub)system. The at least one sensor, and preferably all sensors are configured to check the panels as they move along the production line. It is conceivable that based upon sensor data, at least one panel is isolated from the production line and maneuvered to a dedicated conveyor belt for quality control and checking.

[0020] Preferably, at least one image is provided to the at least one processing system in order to determine the presence of visual identifiers or defects. It is also possible that at least one sensor is displaced in at least three positions relative to the sample panel to take multiple images and / or that multiple sensors are applied to capture the panel from different angles.

[0021] At least one sensor could also comprise at least one structured light scanner. Such structured light scanner can be configured to project a pattern of light onto at least part of the outer surface of the panel and capture the distortion of the pattern by means of at least one camera. It is conceivable that the sensor and / or the processing unit determines the characteristics, size, and / or physical location of the pattern data (unit) or area’s shape and size through the analysis of the pattern distortions. At least one pattern data unit may be selected from pattern dimensions such as size, colour and / or depth. The use of at least one structured light scanner can be beneficial due to being fast, accurate and cost-effective. It is conceivable that at least one structured light scanner is complemented by at least one laser scanner and / or photogrammetry scanner in particular to create at least partially a digital visual representation or a digital model. The at least one digital visual representation can be a histogram, heat map, topographic map, 3D point cloud and / or a triangulation model. The at least one digital model or at least one visual representation of the scanned panel can be represented through simulation software such as AutoCAD, SolidWorks, Rhinoceros + Grasshopper. It is also possible that the system according to the invention, or at least one sensor thereof comprises at least one spectral imager, wherein multiple bands across the electromagnetic spectrum are used for imaging to produce high-quality images. For example, (a) wavelength-scan methods that measure the images one wavelength at a time; (b) spatial-scan methods that measure the whole spectrum of a portion of the image at a time and scan the image or (c) time-scan methods that measure a set of images where each one of them is a superposition of spectral or spatial image information.

[0022] The system in particular enables that the quality of all panels within the production line can be checked. However, the system may also make use of a sampling strategy, wherein for example a subset of panels is selected at random for scanning. The number of panels to be scanned may depend on the specific requirements of the quality control process, as well as the statistical variability of the parameters across the production line, such as the accuracy of the interlocking mechanism, flatness of surfaces, texture design, embossing depth, panel thickness and size, straightness / squareness, etc. Generally, a larger sample size will provide a more accurate representation of the quality of the measured production parameter across the production line but will also require more time and resources. The system can also be designed to operate at the production line speed and / or can be programmed to capture data from specific regions of interest, such as the interlocking mechanism or panel thickness. On the other hand, in other production processes where the speed of production is lower and the features to be measured are simpler, it may be possible to measure certain features in-line and non-stop, providing real-time feedback to the production process.

[0023] In a beneficial embodiment, the system comprises an ejection mechanism. The determination step of at least one qualitative acceptability value may comprise classifying one or more panels as rejected or accepted, and at least one operational instruction may comprise the ejection of one or more panels classified as rejected by means of the ejection mechanism. It is also conceivable that the ejection mechanism forms part of the production line and that the at least one processing unit is configured to instruct the ejection mechanism. The use of an ejection mechanism can further enhance the efficiency of the process, since the rejected panels can be separated at an early stage. It is conceivable that the at least one processing unit is configured to analyze if multiple panels are rejected within a defined time interval. This may be an indication that the defects are not isolated defects but that there is a substantial error in the production line. In case this is determined, it is possible that the operational instructions include the advice to pause the production process.

[0024] At least one processing unit may comprise a system on a chip (SoC) comprising one or more CPUs, graphics processing units (GPUs), memory modules, and / or hardware accelerators. Preferably, the at least one processing unit comprises at least 10GB, preferably at least 24GB of working memory, preferably a video random-access memory (VRAM). This will enable scanning of the panels which are processed at high operational speeds within the production line. At least one processing unit comprises preferably a processor selected from artificial intelligence processing unit (AIPU), tensor processing unit (TPU), field programmable gate arrays (FPGA), Al-specific application-specific integrated circuit (ASIC), neuromorphic processor, quantum processor, GPU, Al accelerator, or any combination thereof. The at least processing unit comprises a computational power of at least 285 Al tera operations per second (TOPS), preferably at least 1321 Al TOPS. In a possible embodiment, the at least one processing unit comprises a plurality of parallel computing processors, stream processors, shader cores, tensor cores, Al accelerators, deep learning accelerators, neural network processors and / or compute unified device architecture (CUDA) cores. It is also conceivable that the at least one processing unit is integrated in at least one edge device and / or comprises at least one machine learning model comprising a general physical quality assessment machine learning model, a decor- and / or SKU-specific finetuned visual quality assessment machine learning model, and / or a production defect cause assessment machine learning model.

[0025] At least one qualitative acceptability value for at least one visual identifier, in particular at least one defect, is preferably determined by using at least one machine learning model. At least one machine learning model preferably comprises a plurality of parameters, weights and / or at least one self-attention layer, preferably a plurality of self-attention layers. It is possible that at least one machine learning model comprises at least one transformer, a diffusion model, a generative adversarial network (GAN), a variational auto-encoder, neural network (NN), convolutional neural network (CNN), a combination of a CNN and a deformable part model (DPM), a deep neural network (DNN), a latent diffusion model, an LLM- based diffusion model, a U-Net architecture, or any combination thereof.

[0026] Preferably, at least one machine learning model has been trained using at least one and preferably at least two of the following options:

[0027] ■ sensor data depicting at least one characteristic of at least one panel, in particular a decorative floor, wall or ceiling panel; and / or

[0028] ■ data representative of at least one physical and / or visual identifier of said decorative floor, wall or ceiling panel, and / or

[0029] ■ data representative of at least one production parameter; and / or

[0030] ■ data representative of a quality defect, a qualitative acceptability value, and / or a accept / reject classification.

[0031] By applying any of more of these data characteristics, the system can be further optimized and the reliability and / or accuracy thereof can be improved. It is for example possible that at least one machine learning model is trained using a labeled dataset comprising sensor data and / or preferably at least 3,000, most preferably at least 5,000 images of panel, which may be classified as defective and / or non-defective panels.

[0032] The system may further comprise at least one visual identifier database, in particular a defect database storing historical visual identifier / defect information for continuous learning and improvement of the at least one machine learning model and / or wherein the at least one machine learning model is periodically retrained using updated data to adapt to variations in panel designs and visual identifier types in particular defect types. This will positively contribute to the effectiveness and accuracy of the system.

[0033] The at least one processing unit is in particular configured to provide an operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded. Thus, in case a panel is considered to be of an unacceptable quality, the processing unit can provide an operation instruction. The operational instruction can for example be a an alert, a warning and / or an advice to an operator but also an instruction or adjustment to the part of the control system of the production line. The operational instruction could for example be a corrective action. It is conceivable that the one or more operational instruction comprises:

[0034] - determining, preferably by using one or more machine learning models, one or more causes associated with the determined qualitative acceptability value;

[0035] - determining, preferably by using one or more machine learning models, one or more corrective actions associated with rectifying the one or more causes; and / or

[0036] - automatically adjusting one or more production parameters according to said corrective actions and / or outputting one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard.

[0037] Preferably, the corrective actions comprise modifying parameters related to speed, temperature, pressure, including but not limited to adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a roller speed, adjusting a coating weight, adjusting a nip width, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof. The system may further comprise at least one controller. It is conceivable that said controller is configured to perform one or more production line parameter adjustments according to said corrective actions via at least one module in the production line such as, but is not limited to, a ripsaw, a punch, a digital printer, a digital embossing printer, a press, an extruder, a laminator, a heater, a cooler, a conveyor, a coating applicator, a UV lamp, a LED lamp, an excimer lamp, a resin applicator, an embossing applicator, a foil texture applicator, a double end tenoner, a programmable logic controller (PLC), or any combination thereof. In case a graphical user interface and / or dashboard is applied, said graphical user interface and / or dashboard preferably aggregates defect statistics, trend analysis, and / or overall production quality metrics. At least one interface, for example a user interface, allows interaction between the system and an operator. The at least one interface may comprise a touchscreen which the operator can use to adjust system parameters, monitor the processes, and send a corrective action to a device connected to the panel production line.

[0038] At least one corrective action, if applied, may include adjusting at least one parameter of the production line. When the at least one processing unit identifies a defect, for example, in the printed decor layer such as unwanted lines in the print, then at least one processing unit outputs a set of corrective actions that the operator can manually select or in some cases automatically performed by the system through the corrective action module. A control signal can be sent to part of the production line responsible for the printed decor. For example, when unwanted lines are identified in the print then a command could be sent to the digital printing line to adjust its print discharge or clean the print heads by printing a test page. The system according to the invention or an additional quality checking system can then validate if the issue has been resolved. If not, a temporary halt in the printing process or the production process and alert the operator of the issue can be suggested or executed.

[0039] In a beneficial embodiment, at least one qualitative acceptability value is determined based on defect size and / or location upon the panel. The defect size and location thereof have a considerable impact on the panel aesthetics. Further, in case the defect is for example present at a coupling part or profile of the panel, the defect may also have an impact on the installation and / or mechanical quality of the panel. The qualitative acceptability value can be a dynamic value. It is imaginable that the qualitative acceptability value is updated based upon information gained from at least one machine learning model. The qualitative acceptability value may be potentially changing as the processing unit receives additional data, updates its evaluation criteria, or incorporates feedback from downstream processes or enduser performance metrics.

[0040] At least one qualitative acceptability value according to the invention is distinct from conventional classification labels in both methodology and output characteristics. While classification labels such as "plaque," "stain," "scratch," "chip," "abrasion," or "no defect" represent discrete categorical identifications of specific defect types, qualitative acceptability values may provide a comprehensive assessment of overall panel suitability that transcends simple defect categorization. Classification labels may function as binary or multi-class identifiers that assign detected features to predefined categories based on visual or physical characteristics. In some conventional systems, a machine learning model may analyze sensor data and output a label indicating the presence or absence of a specific defect type. These labels may serve primarily as diagnostic tools that identify what type of defect exists without necessarily indicating the impact of that defect on panel functionality or acceptability. In contrast, qualitative acceptability values may represent a evaluation that considers not only the presence and type of defects but also their severity, location, frequency, and cumulative impact on panel performance. The processing unit may determine these values by analyzing the relationship between detected characteristics of the panel and requirements of the intended purpose. The qualitative acceptability value determination may involve analysis that weighs multiple factors simultaneously, including defect severity gradients, spatial distribution patterns, and contextual considerations related to panel application requirements.

[0041] It is conceivable that the qualitative acceptability value is expressed as a numerical range, for example a continuous numerical range. The numerical range may comprise a score between 0 and 1 or a percentage. This may result in a nuanced quality decisions, where panels with minor defects may receive acceptability values that reflect their reduced but still adequate quality level, rather than being categorically rejected based on defect presence alone.

[0042] The invention also relates to a method for controlling the quality of panels within a production line. The method according to the invention can be applied in combination with a system according to the present invention. The method may also be performed by the system according to the invention. Any of the embodiments described for the system could also apply to the method, and vice versa.

[0043] The invention relates to a method for controlling the quality of panels within a production line, comprising the steps of:

[0044] - capturing and / or determining at least one characteristic of at least one panel displaced within the production line; and / or

[0045] - receiving sensor data related to at least one characteristic of at least one panel from at least one sensor applied in at least one production line; and / or

[0046] - identify if the panel comprises one or more visual identifiers, in particular one or more defects based on the at least one characteristic of the panel; and / or

[0047] - determining a qualitative acceptability value for at least one and preferably each identified visual identifier, in particular at least one and preferably each identified defect preferably by using at least one machine learning model; and / or providing at least one operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded; and / or optionally executing at least one operational instruction.

[0048] Optionally, at least one panel is cleaned and / or illuminated prior to at least one sensor provides sensor data about the panel. It is possible that at least one panel is cleaned via a cleaning and / or preparation module, and / or illuminated via at least one illumination module preferably prior to and / or during the step of capturing and / or determining at least one characteristic of at least one panel displaced within the production line. At least one cleaning and / or preparation module and / or at least one illumination module could form part of the system according to the invention or to the production line as such. Cleaning of the panels to prepare for scanning can prevent disrupting influence of dirt, dust, and / or debris. It is conceivable that an airblowing system is utilized, which comprises high-pressure air nozzles to blow away dust and debris from at least part of the surface of the panels. For more stubborn contaminants, it is conceivable that a brushing system is used, conceivably together with at least one air-blowing system. Such a brushing system would then scrub the surface of the panels and remove dirt, dust, and other debris through means of rotating brushes or rollers. Such a system is then particularly effective for removing larger surface debris. It is further conceivable that a vacuum system is utilized to complement at least one air-blowing and / or at least one brushing system, which utilizes a negative pressure or suction to remove dust and debris from the surface of the boards. It is of course conceivable to combine multiple cleaning methods, such as air-blowing, brushing, and vacuuming, to achieve the best results. It can be tailored to the specific requirements of the scanning system and the production process and can be automated to operate in conjunction with the conveyor and / or ejection system.

[0049] The method may further or alternatively comprise the step of identifying at least one feature in the at least one panel based on at least part of the received sensor data prior to determining, using one or more machine learning models and based at least on the sensor data, a qualitative acceptability value. At least one operational instruction may comprise determining one or more causes associated with the determined qualitative acceptability value by using one or more machine learning models, determining one or more corrective actions associated with rectifying the one or more causes by using one or more machine learning models, and / or automatically adjusting one or more production parameters according to said corrective actions and / or output one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard. One or more of the corrective actions may comprise adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof. The received sensor data may comprise image data. This may be image data preprocessed via one or more image processing steps such as quality enhancement, noise reduction, resizing, colour correction, cropping, normalization, contrast or brightness adjustment, sharpening, filtering, image segmentation, image compression, image watermarking, image synthesis, image interpretation, or any combination thereof.

[0050] At least one operational instruction may comprise determining a qualitative acceptability value and / or classifying one or more panels as rejected or accepted, and optionally rejecting one or more panels classified as rejected by means of an ejection mechanism.

[0051] The invention will be further elucidated based on the following non-limitative clauses.

[0052] 1 . A system for controlling the quality of panels within a production line, said production line comprising at least one conveyor for displacing at least one panel within said production line, said system comprising:

[0053] - at least one sensor configured to capture and / or determine at least one characteristic of at least one panel displaced within the production line; and

[0054] - at least one processing unit configured to: o receive sensor data related to at least one characteristic of the panel from said at least one sensor; and / or o identify if the panel comprises one or more defects; and / or o determine a qualitative acceptability value based on the received sensor data by using at least one machine learning model; and / or o provide at least one operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded. 2. The system according to clause 1 , wherein the at least one processing unit is configured to identify at least one feature of the panel and provide at least one operational instruction based on the determined feature.

[0055] 3. The system according to claim 1 or clause 2, wherein at least one sensor comprises an optical camera which is configured to capture at least one image of at least a part of at least one panel.

[0056] 4. The system according to any of the previous clauses, wherein at least one operational instruction is directed to at least one operating system of the production line.

[0057] 5. The system according to any of the previous clauses, wherein at least one sensor is chosen from the group of optical camera, image sensor, infrared sensor, ultrasonic sensor, temperature sensor, humidity sensor, position sensor, tilt sensor, color sensor, proximity sensor, capacitive sensor, load cell, or any combination thereof.

[0058] 6. The system according to any of the previous clauses, wherein the system comprises at least one ejection mechanism, and wherein the determination of at least one qualitative acceptability value comprises classifying one or more panels as rejected or accepted, and wherein at least one operational instruction comprises the ejection of one or more panels classified as rejected by means of the at least one ejection mechanism.

[0059] 7. The system according to any of the previous clauses, wherein the at least one processing unit comprises a system on a chip (SoC) comprising one or more CPUs, graphics processing units (GPUs), memory modules, and / or hardware accelerators.

[0060] 8. The system according to any of the previous clauses, wherein the at least one processing unit comprises at least 10GB, preferably at least 24GB of working memory, preferably a video random-access memory (VRAM). 9. The system according to any of the previous clauses, wherein the at least one processing unit comprises a processor selected from artificial intelligence processing unit (AIPU), tensor processing unit (TPU), field programmable gate arrays (FPGA), Al-specific application-specific integrated circuit (ASIC), neuromorphic processor, quantum processor, GPU, Al accelerator, or any combination thereof.

[0061] 10. The system according to any of the previous clauses, wherein the at least one processing unit comprises a plurality of parallel computing processors, stream processors, shader cores, tensor cores, Al accelerators, deep learning accelerators, neural network processors and / or compute unified device architecture (CUDA) cores.

[0062] 11 . The system according to any of the previous clauses, wherein the at least one processing unit comprises a computational power of at least 285 Al tera operations per second (TOPS), preferably at least 1321 Al TOPS.

[0063] 12. The system according to any of the previous clauses, wherein the at least one processing unit is integrated in at least one edge device and / or comprises at least one machine learning model comprising a general physical quality assessment machine learning model, a decor- and / or SKU-specific finetuned visual quality assessment machine learning model, and / or a production defect cause assessment machine learning model.

[0064] 13. The system according to any of the previous clauses, wherein at least one machine learning model comprises a plurality of weights and / or at least one selfattention layer, preferably a plurality of self-attention layers.

[0065] 14. The system according to any of the previous clauses, wherein at least one machine learning model comprises at least one transformer, a vision transformer (ViT), a diffusion model, a generative adversarial network (GAN), a variational autoencoder, neural network (NN), convolutional neural network (CNN), a combination of a CNN and a deformable part model (DPM), a deep neural network (DNN), a residual network (ResNet), a latent diffusion model, an LLM-based diffusion model, a U-Net architecture, or any combination thereof. 15. The system according to any of the previous clauses, wherein at least one machine learning model has been trained using at least two of the following options:

[0066] ■ sensor data depicting at least one characteristic of at least one panel;

[0067] ■ data representative of at least one physical and / or visual identifier of said decorative floor, wall or ceiling panel;

[0068] ■ data representative of at least one production parameter; and / or

[0069] ■ data representative of a quality defect, a qualitative acceptability value, and / or an accept / reject classification.

[0070] 16. The system according to any of the previous clauses, wherein at least one machine learning model is trained using a labeled dataset comprising sensor data and / or preferably at least 3,000, most preferably at least 5,000 images of defective and / or non-defective panels.

[0071] 17. The system according to any of the previous clauses, further comprising a defect database storing historical defect information for continuous learning and improvement of the at least one machine learning model and / or wherein the at least one machine learning model is periodically retrained using updated data to adapt to variations in panel designs and defect types.

[0072] 18. The system according to any of the previous clauses, wherein the one or more operational instruction comprises:

[0073] - determining, by using one or more machine learning models, one or more causes associated with the determined qualitative acceptability value;

[0074] - determining, by using one or more machine learning models, one or more corrective actions associated with rectifying the one or more causes; and / or

[0075] - automatically adjusting one or more production parameters according to said corrective actions and / or outputting one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard; wherein the corrective actions comprise modifying parameters related to speed, temperature, pressure, including but not limited to adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a roller speed, adjusting a coating weight, adjusting a nip width, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof.

[0076] 19. The system according to any of the previous clauses, wherein at least one qualitative acceptability value is determined based on defect size and / or location upon the panel.

[0077] 20. A method for controlling the quality of panels within a production line, comprising the steps of:

[0078] - capturing and / or determining at least one characteristic of at least one panel displaced within the production line;

[0079] - receiving sensor data related to at least one characteristic of at least one panel from at least one sensor applied in at least one production line;

[0080] - identify if the panel comprises one or more defects based on the at least one determined characteristic of the panel;

[0081] - determining a qualitative acceptability value for each identified defect by using at least one machine learning model; and

[0082] - providing an operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded; and

[0083] - optionally executing said operational instruction.

[0084] 21 . The method according to clause 20, wherein the panel is cleaned via a cleaning module and / or illuminated via an illumination module prior to and / or during the step of capturing and / or determining at least one characteristic of at least one panel displaced within the production line.

[0085] 22. The method according to any of clauses 20-21 , further comprises the step of identifying at least one feature in the at least one panel based on at least part of the received sensor data prior to determining, using one or more machine learning models and based at least on the sensor data, a qualitative acceptability value.

[0086] 23. The method according to any of clauses 20-22, wherein the one or more operational instruction comprises:

[0087] - determining one or more causes associated with the determined qualitative acceptability value by using one or more machine learning models; - determining one or more corrective actions associated with rectifying the one or more causes by using one or more machine learning models; and

[0088] - automatically adjusting one or more production parameters according to said corrective actions and / or output one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard; wherein the corrective actions comprise adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof.

[0089] 24. The method according to any of clauses 20-23, wherein the received sensor data are image data preprocessed via one or more image processing steps such as quality enhancement, noise reduction, resizing, colour correction, cropping, normalization, contrast or brightness adjustment, sharpening, filtering, image segmentation, image compression, image watermarking, image synthesis, image interpretation, or any combination thereof.

[0090] 25. The method according to any of clauses 20-24, wherein the one or more operational instruction comprises:

[0091] - determining a qualitative acceptability value and / or classifying one or more panels as rejected or accepted; and

[0092] - ejecting one or more panels classified as rejected by means of an ejection mechanism.

[0093] 26. The method according to any of clauses 20-25, wherein the method is performed by the system according to any one of clauses 1 - 19.

[0094] It will be clear that the invention is not limited to the exemplary embodiments which are described here, but that countless variants are possible within the framework of the attached claims, which will be obvious to the person skilled in the art. In this case, it is conceivable for different inventive concepts and / or technical measures of the above-described variant embodiments to be completely or partly combined without departing from the inventive idea described in the attached claims. The verb 'comprise' and its conjugations as used in this patent document are understood to mean not only 'comprise', but to also include the expressions 'contain', 'substantially contain', 'formed by' and conjugations thereof.

Claims

Claims1 . A system for controlling the quality of panels within a production line, said production line comprising at least one conveyor for displacing at least one panel within said production line, said system comprising:- at least one sensor configured to capture and / or determine at least one characteristic of at least one panel displaced within the production line; and- at least one processing unit configured to: o receive sensor data related to at least one characteristic of the panel from said at least one sensor; o identify if the panel comprises one or more defects; and o determine a qualitative acceptability value based on the received sensor data by using at least one machine learning model; and o provide at least one operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded.

2. The system according to claim 1 , wherein the at least one processing unit is configured to identify at least one feature of the panel and provide at least one operational instruction based on the determined feature.

3. The system according to claim 1 or claim 2, wherein at least one sensor comprises an optical camera which is configured to capture at least one image of at least a part of at least one panel.

4. The system according to any of the previous claims, wherein at least one operational instruction is directed to at least one operating system of the production line.

5. The system according to any of the previous claims, wherein at least one sensor is chosen from the group of optical camera, image sensor, infrared sensor, ultrasonic sensor, temperature sensor, humidity sensor, position sensor, tilt sensor, color sensor, proximity sensor, capacitive sensor, load cell, or any combination thereof.

6. The system according to any of the previous claims, wherein the system comprises at least one ejection mechanism, and wherein the determination of at least one qualitative acceptability value comprises classifying one or more panels as rejected or accepted, and wherein at least one operational instruction comprises the ejection of one or more panels classified as rejected by means of the at least one ejection mechanism.

7. The system according to any of the previous claims, wherein the at least one processing unit comprises a system on a chip (SoC) comprising one or more CPUs, graphics processing units (GPUs), memory modules, and / or hardware accelerators.

8. The system according to any of the previous claims, wherein the at least one processing unit comprises at least 10GB, preferably at least 24GB of working memory, preferably a video random-access memory (VRAM).

9. The system according to any of the previous claims, wherein the at least one processing unit comprises a processor selected from artificial intelligence processing unit (AIPU), tensor processing unit (TPU), field programmable gate arrays (FPGA), Al-specific application-specific integrated circuit (ASIC), neuromorphic processor, quantum processor, GPU, Al accelerator, or any combination thereof.

10. The system according to any of the previous claims, wherein the at least one processing unit comprises a plurality of parallel computing processors, stream processors, shader cores, tensor cores, Al accelerators, deep learning accelerators, neural network processors and / or compute unified device architecture (CUDA) cores.11 . The system according to any of the previous claims, wherein the at least one processing unit comprises a computational power of at least 285 Al tera operations per second (TOPS), preferably at least 1321 Al TOPS.

12. The system according to any of the previous claims, wherein the at least one processing unit is integrated in at least one edge device and / or comprises atleast one machine learning model comprising a general physical quality assessment machine learning model, a decor- and / or SKU-specific finetuned visual quality assessment machine learning model, and / or a production defect cause assessment machine learning model.

13. The system according to any of the previous claims, wherein at least one machine learning model comprises a plurality of weights and / or at least one selfattention layer, preferably a plurality of self-attention layers.

14. The system according to any of the previous claims, wherein at least one machine learning model comprises at least one transformer, a vision transformer (ViT), a diffusion model, a generative adversarial network (GAN), a variational autoencoder, neural network (NN), convolutional neural network (CNN), a combination of a CNN and a deformable part model (DPM), a deep neural network (DNN), a residual network (ResNet), a latent diffusion model, an LLM-based diffusion model, a U-Net architecture, or any combination thereof.

15. The system according to any of the previous claims, wherein at least one machine learning model has been trained using at least two of the following options:■ sensor data depicting at least one characteristic of at least one panel;■ data representative of at least one physical and / or visual identifier of said decorative floor, wall or ceiling panel;■ data representative of at least one production parameter; and / or■ data representative of a quality defect, a qualitative acceptability value, and / or an accept / reject classification.

16. The system according to any of the previous claims, wherein at least one machine learning model is trained using a labeled dataset comprising sensor data and / or preferably at least 3,000, most preferably at least 5,000 images of defective and / or non-defective panels.

17. The system according to any of the previous claims, further comprising a defect database storing historical defect information for continuous learning and improvement of the at least one machine learning model and / or wherein the at leastone machine learning model is periodically retrained using updated data to adapt to variations in panel designs and defect types.

18. The system according to any of the previous claims, wherein the one or more operational instruction comprises:- determining, by using one or more machine learning models, one or more causes associated with the determined qualitative acceptability value;- determining, by using one or more machine learning models, one or more corrective actions associated with rectifying the one or more causes; and / or- automatically adjusting one or more production parameters according to said corrective actions and / or outputting one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard; wherein the corrective actions comprise modifying parameters related to speed, temperature, pressure, including but not limited to adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a roller speed, adjusting a coating weight, adjusting a nip width, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof.

19. The system according to any of the previous claims, wherein at least one qualitative acceptability value is determined based on defect size and / or location upon the panel.

20. A method for controlling the quality of panels within a production line, comprising the steps of:- capturing and / or determining at least one characteristic of at least one panel displaced within the production line;- receiving sensor data related to at least one characteristic of at least one panel from at least one sensor applied in at least one production line;- identify if the panel comprises one or more defects based on the at least one determined characteristic of the panel;- determining a qualitative acceptability value for each identified defect by using at least one machine learning model; and- providing an operational instruction in case the limit value for qualitative acceptability of an identified defect is exceeded; andoptionally executing said operational instruction.21 . The method according to claim 20, wherein the panel is cleaned via a cleaning module and / or illuminated via an illumination module prior to and / or during the step of capturing and / or determining at least one characteristic of at least one panel displaced within the production line.

22. The method according to any of claims 20-21 , further comprises the step of identifying at least one feature in the at least one panel based on at least part of the received sensor data prior to determining, using one or more machine learning models and based at least on the sensor data, a qualitative acceptability value.

23. The method according to any of claims 20-22, wherein the one or more operational instruction comprises:- determining one or more causes associated with the determined qualitative acceptability value by using one or more machine learning models;- determining one or more corrective actions associated with rectifying the one or more causes by using one or more machine learning models; and- automatically adjusting one or more production parameters according to said corrective actions and / or output one or more recommended corrective actions to a graphical user interface (GUI) and / or dashboard; wherein the corrective actions comprise adjusting a printer speed, cleaning printer heads, adjusting a conveyor speed, adjusting a sawing or pressing position, speed or force, adjusting an extrusion rate, melt temperature or melt pressure, or any combination thereof.

24. The method according to any of claims 20-23, wherein the received sensor data are image data preprocessed via one or more image processing steps such as quality enhancement, noise reduction, resizing, colour correction, cropping, normalization, contrast or brightness adjustment, sharpening, filtering, image segmentation, image compression, image watermarking, image synthesis, image interpretation, or any combination thereof.

25. The method according to any of claims 20-24, wherein the one or more operational instruction comprises:- determining a qualitative acceptability value and / or classifying one or more panels as rejected or accepted; and- ejecting one or more panels classified as rejected by means of an ejection mechanism.

26. The method according to any of claims 20-25, wherein the method is performed by the system according to any one of claims 1 - 19.