Production analysis modeling for product quality testing
By analyzing image data of the adhesion-promoting layer using machine learning models, the problem of inadequate monitoring during the application of the adhesion-promoting layer was solved, enabling real-time monitoring and prediction of product quality and improving quality control in the production process.
Patent Information
- Application Number
- CN202480068006.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-10-31
- Filing Date
- 2024-10-01
- Publication Date
- 2026-06-09
Smart Images

Figure CN122180974A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to production analysis modeling for product quality inspection. Such techniques can be particularly useful for predicting quality or quality defects associated with products from a particular manufacturer, i.e., by modeling production data from multiple different manufacturers of the product. Background Technology
[0002] Artificial neural networks (ANNs) are networks that process information by modeling networks of neurons, such as those in the human brain, to process information sensed in a specific environment (e.g., stimuli). Similar to the human brain, neural networks typically consist of multi-neuron topologies (e.g., which may be referred to as artificial neurons). An ANN operation refers to the operation of using artificial neurons to process input to perform a given task. ANN operations can involve executing various machine learning algorithms to process the input. Example tasks that can be processed by performing ANN operations include machine vision, speech recognition, machine translation, social network filtering, and / or medical diagnosis.
[0003] Multiple manufacturing facilities offer foam formulations and expertise for dual-belt lamination customers in manufacturing sandwich metal sheets with rigid polyurethane or polyisocyanurate cores. For high-index foam formulations, a primer or adhesion promoter is applied across the entire metal finish before polyurethane casting. This is done using a rotating disc that spreads droplets of the reacting primer onto the application surface. Application of the adhesion promoter is considered mandatory when the foam used is a high-index polyisocyanurate type, but is recommended for other polyurethane foams as it ensures a durable bond between the foam and the metal sheet.
[0004] Proper primer distribution is crucial for producing panels free of potential defects that can develop into blistering over time after panel installation and subsequently lead to claims from downstream customers. The problem is that current monitoring of the adhesion promoter layer is inadequate, or even nonexistent. Currently, there is no quantitative inspection of primer distribution during production. Summary of the Invention
[0005] This disclosure relates to improvements in using machine learning techniques to predict product properties (e.g., quality, performance, defects, failures, etc.). Image-based visual analytics utilizing machine learning models can be used to monitor and analyze production processes to predict product properties based on production analysis data. Image-based visual analytics can utilize machine learning models to interpret and extract data from captured images by identifying relevant patterns and features within the image. These features may include, but are not limited to, edges, textures, shapes, colors, and / or other visual attributes. Predictions of product properties can be received from the machine learning model, and these predictions can be used to adjust the production process or determine whether to reject the product.
[0006] As a concrete example, data can be collected from multiple different manufacturing sites to generate machine learning models for predicting the quality of products produced by different manufacturing sites. Machine learning models can be used to identify production parameters that may be the cause of specific properties. As used herein, production parameters refer to data collected during the production of a particular product. For example, production parameters used for depositing an adhesion-promoting layer may include, but are not limited to: droplet distribution data, layer uniformity data, composition data, mixing ratio data, time data between the deposition of the adhesion-promoting layer and subsequent layers, humidity data and / or temperature data, and other data that may potentially affect the quality of products produced using the adhesion-promoting layer. Furthermore, the machine learning model can generate ranges of production parameters for multiple production parameters, which can be provided to multiple different manufacturing sites to ensure product quality levels.
[0007] The foregoing description of this invention is not intended to describe every disclosed embodiment or every implementation thereof. The following description illustrates exemplary embodiments in more detail. Throughout this application, guidance is provided by a list of examples, which may be used in various combinations. In each case, the enumerated list serves only as a representative group and should not be construed as an exclusive list. Attached Figure Description
[0008] Figure 1 An example of a system for production analysis modeling for product quality inspection is shown.
[0009] Figure 2 This is an example illustration of a thermal image of the applied surface.
[0010] Figure 3 This is an example illustration showing the curing rate and mixing ratio of the adhesive-promoting layer.
[0011] Figure 4 An example of a method for production analysis modeling for product quality inspection is shown.
[0012] Figure 5An example of a machine-readable medium is shown for production analysis modeling for product quality inspection.
[0013] Figure 6 An example of equipment used for production analysis modeling for product quality inspection is shown. Detailed Implementation
[0014] This disclosure relates to methods and apparatus for production analysis modeling for product quality inspection, which utilize machine learning models to predict product properties of one or more products generated using different production analysis data. An example of a machine learning model is an ANN (Application Not Node). An ANN provides learning by forming probabilistic weight associations between inputs and outputs. These probabilistic weight associations can be provided by multiple nodes comprising the ANN. Nodes, along with weights, biases, and / or activation functions, can be used to generate the ANN's output based on its inputs. Multiple nodes of the ANN can be grouped to form layers within the ANN.
[0015] A machine learning model can be a function or equation used to identify patterns in data. A machine learning module can be multiple machine learning models used together to identify patterns in data. In a specific example, a machine learning module can be organized as a neural network. A neural network can include a set of instructions that can be executed to identify patterns in data. Some neural networks can be used to identify underlying relationships in a set of data in a way that mimics the operation of the human brain. Neural networks can adapt to changing or altered inputs, allowing them to produce acceptable results without redesigning the output criteria.
[0016] The production of a specific product can utilize different conditions during the production process. These production conditions can have varying impacts on product quality. Different types of production conditions will have different degrees of influence on the final product produced. Production conditions can be monitored using different types of equipment. For example, imaging equipment, sequencing equipment, sensor equipment, and / or other types of equipment can be used to monitor different types of conditions to generate production data specific to the product.
[0017] Production data can be generated from multiple different manufacturing sites and implemented in a uniform format (e.g., production analytics data) to allow different manufacturing sites to provide federated data for generating or training machine learning models. Production analytics data may include image data collected during the production process, production setup data collected during the production process, and / or environmental data collected during the production process. Machine learning models can utilize production data or production analytics data and corresponding quality data associated with the produced products to generate production ranges, which can be used by each of the multiple manufacturing sites to improve product quality. In this way, machine learning models can generate product reviews for a specific product to identify one or more production data points from the production data to be reviewed by the product reviewer, attributing to a specific defect or quality level of that product.
[0018] As used herein, unless otherwise expressly stated, the singular forms “a / an” and “the” include both singular and plural indicators. Furthermore, throughout this application, the word “may” is used in a permissible sense (e.g., possibly, able to) rather than a mandatory sense (e.g., must). The term “comprising” and its derivatives mean “including but not limited to”.
[0019] As will be understood, elements shown in the various embodiments herein may be added, interchanged, and / or eliminated to provide multiple additional embodiments of this disclosure. Furthermore, as will be understood, the proportions and relative scales of the elements provided in the figures are intended to illustrate certain embodiments of the invention and should not be considered limiting.
[0020] Figure 1 An example of a system 100 for production analysis modeling for product quality inspection is illustrated. System 100 illustrates a system for applying an adhesion-promoting layer to an application surface. Although system 100 illustrates a specific system for applying an adhesion-promoting layer to a particular surface, this disclosure is not limited thereto. For example, other types of systems may utilize the functions and / or features described herein in a similar manner for production analysis modeling for product quality inspection of other types of products.
[0021] System 100 includes a primer distribution device 102 that can dispense a chemical compound used as an adhesion-promoting layer (e.g., primer, etc.) onto a surface. The primer distribution device 102 may include a distribution disk that can rotate to distribute the adhesion-promoting layer onto the surface. The distribution disk may rotate at different speeds to spread or dispense the chemicals constituting the adhesion-promoting layer. In some examples, the adhesion-promoting layer may include multiple chemicals distributed onto the distribution disk. In some examples, the multiple chemicals may be distributed onto the distribution disk at different flow rates to create a specific chemical ratio among the multiple chemicals. As used herein, flow rate describes the rate at which a given amount of chemical is dispensed onto the distribution disk. When the primer distribution device 102 dispenses the chemical onto the surface, the chemical may be dispersed with a specific droplet distribution and / or a specific layer uniformity. As used herein, droplet distribution may be the number of droplets deposited in multiple sub-regions on a specific area. For example, droplet distribution may represent a specific number of droplets in each of the multiple sub-regions to identify when a particular sub-region has a relatively more or fewer droplet count compared to other sub-regions within the specific area.
[0022] Similarly, a specific layer uniformity can represent the layer thickness of the adhesion-promoting layer in multiple sub-regions of a specific area. In this way, a specific layer uniformity can identify when a first sub-region has a layer thickness with a different chemical substance than a second sub-region. Fluctuations in the amount of chemical substances dispersed on the surface of the material can be captured and used as production data and / or production analysis data, as droplet distribution and / or layer uniformity affect the quality of the final product. In some embodiments, production setup data may be included with image data, such that the production setup used to generate a specific layer uniformity can be used as production analysis data.
[0023] An adhesion-promoting layer on the surface can be exemplified by surface portion 104. Surface portion 104 can be a metallic surface including an adhesion-promoting layer deposited by primer distribution device 102. System 100 may include a light source 110 that can be directed to surface portion 104. Light source 110 can be a visible light source and / or other types of light sources based on the type of image to be captured. The position or angle of light source 110 can be changed over a period of time to capture multiple different types of images of surface portion 104 during that period.
[0024] Surface portion 104 can be positioned on a transport system to transport it to different areas, thereby allowing different substances or chemicals to be deposited on surface portion 104. In these embodiments, light source 110 can be positioned at a first location, and different light sources can be positioned at second locations. In this way, as surface portion 104 moves from a location receiving deposits from primer distribution device 102 to a different distribution device, different light sources can be used to capture the surface portion.
[0025] The transport of a surface portion from a first location to a second location can be monitored by a sequencing device to determine the amount of time taken for the surface portion to move from the first location to the second location. System 100 may include additional sensors to monitor the amount of time taken for the surface portion 104 to move from the location where it receives chemical deposits from primer distribution device 102 to the different distribution device. The chemical substance deposited by primer distribution device 102 may undergo a specific chemical reaction. In this way, the amount of time taken when the surface portion 104 arrives at the different distribution device can be used to determine the state of the chemical reaction.
[0026] System 100 may include imaging device 112. Imaging device 112 may be a device for capturing images of surface portion 104 to identify the properties of the adhesion-promoting layer deposited on surface portion 104 by primer distribution device 102. Imaging device 112 may include one or more of the following: a camera, a still image camera, a thermal imaging camera, a hyperspectral camera, and / or other types of devices capable of capturing images and / or video of surface portion 104. Imaging device 112 may be a standardized imaging device that can be utilized by each of multiple manufacturing sites to create federated data from each of the multiple manufacturing sites. In a similar manner, the position and / or angle of imaging device 112 may be a standardized position and / or angle to capture images of surface portion 104 at the same or similar location as primer distribution device 102. In this way, data captured by imaging device 112 can be correlated with data captured at different manufacturing sites.
[0027] System 100 may include a control panel 106 that can be utilized by user 108. Control panel 106 may be used to display notifications generated by system 100 to notify user 108 when the current conditions or measurements of system 100 exceed a specific range. As further described herein, data collected during the production of surface portion 104 and / or the product utilizing surface portion 104 may be used to generate production ranges that can be utilized by system 100.
[0028] Images captured by imaging device 112 can be provided to machine learning model 114 operating on edge computing device 116. Machine learning model 114 can be used to analyze the images received from imaging device 112. In other embodiments, machine learning model 114 can be used to organize data collected by system 100 into a data file that can be correlated with the final product utilizing surface portion 104. For example, humidity data, temperature data, droplet distribution data of chemical layer, layer uniformity data of chemical layer, and / or time data between the deposition of chemical layer and the deposition of different layers on the chemical layer can be correlated and associated with the product utilizing a specific surface portion 104. In this way, the quality of surface portion 104 can be determined over a period of time and can be correlated with data associated with surface portion 104.
[0029] System 100 may include an edge computing device 116. In some examples, the edge computing device 116 is a computing device that includes processor resources and machine-readable medium for storing instructions executed by the processor resources to perform a specific function. The edge computing device 116 may be used to communicate with a remote device 124. The remote device 124 may be a cloud device that receives data from multiple manufacturing sites to augment the dataset used to train a machine learning model 114 for a specific product and / or a portion of a product, such as surface portion 104. The edge computing device 116 may be used to remove data associated with a specific manufacturing site that the site does not wish to share with other manufacturing sites.
[0030] In other embodiments, edge computing device 116 may utilize production ranges (e.g., production condition ranges, condition thresholds, etc.) to monitor data provided by machine learning model 114. When data received by machine learning model 114 exceeds the threshold range, edge computing device 116 may send a notification to control panel 106 to inform user 108. In this way, changes to production data and / or production settings can be made in real time. When edge computing device 116 determines that production data exceeds the data threshold range, surface portion 104 may be discarded or not used in a particular product. As described herein, the threshold range may be provided by remote device 124 when remote device 124 is utilizing the machine learning model.
[0031] As used herein, production settings can be adjustable, defining production parameters that define how system 100 deposits a chemical layer onto surface portion 104. For example, production settings may include, but are not limited to: the positioning of imaging device 112 and / or light source 110 for altering or influencing the image; the positioning of primer distribution device 102; the rotational speed of primer distribution device 102; the flow rate of the chemical substance; the temperature of the chemical substance; the temperature of primer distribution device 102; the linear velocity at which surface portion 104 moves from a first position to a second position to alter or influence the distribution; the curing time of the chemical substance or primer; and / or the chemical ratio of the chemical substance or primer. In this way, system 100 can change or adjust production settings to alter or adjust the manner in which primer distribution device 102 applies the chemical layer to surface portion 104. In this way, production settings can be changed to ensure that production data remains within threshold ranges provided by machine learning models.
[0032] Remote device 124 can receive data from edge computing device 116 and data organized in the same manner from multiple other manufacturing sites 120. In this way, remote device 124 can utilize machine learning models to perform advanced data analysis 118 on the data received from edge computing device 116. Remote device 124 can use this data from edge computing device 116 and other manufacturing sites 120 to generate supporting knowledge for control panel 106. As further described herein, supporting knowledge may include product review analysis for specific products manufactured at a particular time and / or a range of production data that can be used by edge computing device 116 to generate real-time notifications to control panel 106.
[0033] Figure 2 This is an example illustration 230 illustrating a thermal image of the applied surface. (See illustration 230.) Figure 1 The illustrated imaging device 112 may be a thermal imaging device (e.g., an infrared camera, a thermal imaging camera, etc.). As used herein, a thermal imaging device can be a device that captures and / or displays thermal energy emitted by an object. For example, a thermal imaging device can detect and generate a visual representation or image of thermal markings emitted by an object or substance. Figure 230 may be an example of how a thermal imaging device can be used to determine the droplet distribution of an adhesion-promoting layer on a metal surface. In other embodiments, a hyperspectral camera may be used to determine the droplet distribution of the adhesion-promoting layer. As further described herein, a hyperspectral camera is an imaging device that captures and analyzes the electromagnetic spectrum across a range of wavelengths. That is, hyperspectral image data may be used to determine the droplet distribution of the adhesion-promoting layer.
[0034] Figure 230 illustrates cold regions 232 and warm regions 234-1, 234-2, where the cold regions correspond to the metal surface or the surface to which the material is applied, and the warm regions correspond to the chemical substances in the adhesion-promoting layer. In this way, the number and location of the warm regions 234-1, 234-2 can be used to identify the droplet distribution of the adhesion-promoting layer. The droplet distribution of a specific surface region can be used to generate production analysis data for that specific surface region. As described herein, production setup data (e.g., setups used during production) can be used in conjunction with the droplet distribution data to generate production analysis data. As illustrated in Figure 230, the upper left quadrant of a specific surface region may contain less adhesion-promoting layer compared to the lower left quadrant. Images captured by the thermal imaging device illustrated in Figure 230 can be used to monitor the droplet distribution in real time and generate production analysis data to be provided to a machine learning model.
[0035] As described in this paper, the machine learning model can utilize generated production analytics data and corresponding quality data to generate a range for multiple production data points. For example, the production analytics data may include the droplet distribution of the adhesive promoting layer, and the machine learning model can generate a droplet distribution range for a specific manufacturing location. In this way, a specific manufacturing location can use thermal imaging equipment to determine the droplet distribution in real time and generate a notification when the droplet distribution exceeds the range. Furthermore, the machine learning model can update the droplet distribution range and / or other production data based on additional production analytics data and / or additional quality data received from multiple different manufacturing locations.
[0036] Images acquired from thermal imaging cameras can be streamed to enable machine learning models (e.g., ...) Figure 1 The analysis is performed using machine learning models (e.g., 114, cited in the text). These models utilize imaging algorithms to validate imaging parameters, such as the distribution of the adhesion-promoting layer. The machine learning model measures percentage coverage, which is measured as the number of pixels with an adhesion-promoting layer response compared to the number of pixels in the metallic background. This image is compared to a database of images with "normal" distributions to detect anomalous distributions, such as uneven spreading or adhesion-promoting layer aggregation at edges. When the percentage distribution of the adhesion-promoting layer exceeds a threshold range for droplet distribution, these images can be saved to the database or stored for further analysis and / or further training of the machine learning model. Further analysis can be used to generate notifications for the control panel.
[0037] Production analysis data for a specific surface area can be correlated with product quality data for that specific surface area and / or products generated using that surface area. In this way, production analysis data for multiple surface areas generated at multiple different manufacturing sites and the resulting product quality data can be used to generate and / or train machine learning models. Furthermore, the machine learning models can generate production setup changes that can be implemented at specific sites to produce production analysis data or to generate products that achieve production analysis data associated with relatively high-quality products.
[0038] Figure 3 This is an example illustration 340 illustrating the curing rate and mixing ratio of an adhesion-promoting layer. Illustration 340 may include graphical representations, which may be visual representations of the deposited chemicals for the adhesion-promoting layer generated by a hyperspectral camera. That is, when the imaging device includes a hyperspectral camera (e.g., an imaging spectrometer, etc.), graphical representations similar to those illustrated in Illustration 340 may be used to identify the curing rate and / or mixing ratio of the deposited chemicals.
[0039] As used in this article, a hyperspectral camera is an imaging device that captures and analyzes the electromagnetic spectrum across a range of wavelengths. Hyperspectral cameras operate based on spectroscopic principles that involve analyzing the interaction of light with materials. These cameras use specialized sensors capable of detecting and quantifying the intensity of light in hundreds or even thousands of spectral bands spanning from ultraviolet (UV) wavelengths to near-infrared (NIR) wavelengths.
[0040] Graph 341 may be a representation of a spectrometer graph (e.g., a spectrum, a spectral curve, etc.). Graph 341 may be a visual representation of how light intensity or electromagnetic radiation varies with wavelength or frequency. For example, graph 341 may include wavelength on the horizontal axis and intensity on the vertical axis. Graph 341 may illustrate a visual representation of how a machine learning model can identify the curing rate of chemicals associated with the adhesion-promoting layer. A portion 342 of graph 341 may be further analyzed by the machine learning model to determine the curing rate of the chemicals over a period of time. For example, each curve in graph 341 may correspond to a different curing time of the chemicals associated with the adhesion-promoting layer. The curing rate may be monitored by the machine learning model by monitoring the growth of polyurethane markers within each of multiple time periods. In this way, a larger amount of polyurethane may correspond to the chemicals being further or closer to complete curing during the curing process.
[0041] Section 342 is further magnified in graph 344. Each of the multiple curves may correspond to the chemical substance at the following times: time 0 (t-0), time 5 minutes (t-5), time 10 minutes (t-10), and / or time 2 hours (t-2hr). In this example, time 0 may be the start time of curing, and time 2 hours may be 2 hours after the start time of curing. Section 342 may illustrate a visual representation of how a machine learning model can determine the curing time of the adhesion accelerator layer when a second or additional layer is applied to it. This curing time may be added to and / or utilized as production analysis data. Production analysis data may also include the production settings used when applying the adhesion accelerator layer. As described herein, section 342, or similar data used to generate section 342, may be used by a machine learning model to monitor the curing time of the adhesion accelerator layer in real time to generate a notification when the curing time exceeds a threshold curing time. Furthermore, the curing time may be used to generate production analysis data provided to the machine learning model. In these examples, curing time, production setting data, and / or environmental data may be included as production analysis data.
[0042] Graph 345 illustrates a visual representation of how a machine learning model can determine the mixing ratios of chemicals associated with an adhesive-promoting layer. For example, the mixing ratio could be for a two-component polyurethane adhesive. Graph 345 could also be a spectrophotometer plot comparing the mixture to pure samples of its individual components. Graph 345 illustrates isocyanate components (such as vornate) ™ ) and polyol components (such as Voramer) ™ Specific examples of mixtures are provided. However, the examples in this article are not limited to these. Additional two-component or multi-component mixtures can also be represented using this type of visualization, illustrated by Figure 345. For example, two-component or multi-component mixtures, such as epoxy resins, can be utilized in a similar manner.
[0043] Part 346 of graph 345 is illustrated as a magnified portion in graph 348. Figure 348 illustrates a visual representation of how a machine learning model can determine the differences between pure and mixed samples to determine the ratio of the mixture. The ratio of the mixture can alter the performance of the adhesion-promoting layer. For example, the ratio of the mixture can change the curing time of the adhesion-promoting layer. In this way, the ratio of the mixture can be used to generate production analysis data to be provided to the machine learning model and / or for real-time monitoring of the ratio of the mixture.
[0044] Figure 4An example of a method 460 for production analysis modeling for product quality inspection is illustrated. In some examples, method 460 may be performed by a computing device as described herein. Method 460 may be used to identify specific production data that may be the cause of a particular defect and / or to identify threshold ranges for specific production data that can be utilized by a particular manufacturing site.
[0045] At step 461, method 460 may include receiving a review request for a product that has been identified as defective. As used herein, the review request may be a request to analyze a specific product. The review request may include an identifier that can be used to identify production analysis data for that product. As used herein, the identifier may be an indicator or unique identifier that allows the product to be associated with the production analysis data. In this way, specific production analysis data used during the production of the product (e.g., image data, calculations based on image data, environmental data, production setup data, etc.) can be compared with other production analysis data for other products. A defective product may be an indication that its quality level is below a threshold quality level. For example, a defective product may be a portion of the product that is visually or functionally below a threshold quality level for the product.
[0046] The review request can be provided by one of several different users. For example, the review request could be provided by one of the product's manufacturer, customer, seller, distributor, or other users associated with the product. A physical product may not be required to identify the production analytics data for that product. For example, an identifier can be provided within the review request and used to identify the production analytics data for that product. Identified failures can be used as known outputs from the production analytics data for that product, which are then taken as input, to (further) train a machine learning model. In this way, the production analytics data for that product can be associated with the identified failure data.
[0047] At step 462, method 460 may include identifying production analysis data associated with the production process for the product. Production analysis data includes multiple data sets captured during the production of the product. Identifying production analysis data for a specific product may include using a product identifier to extract production analysis data associated with the production process during the production of the product. By identifying production analysis data associated with the production process of the product, data collected during the production of the specific product can be compared with data collected during the production of similar or identical products.
[0048] Production analysis data for this product may include, but is not limited to, droplet distribution data, layer uniformity data, and / or environmental data at specific points in time when the product is manufactured at a particular production site. As described herein, identifying production analysis data for this product may include using identifiers to determine a specific date and time corresponding to the production of the product or a portion thereof by a particular manufacturer. In this way, data from a particular manufacturer can be used to identify production analysis data during a specific date and time of production of the product or a portion thereof.
[0049] As described herein, production analysis data may include production and analysis data during the production of the product and / or multiple parts of the product. In this way, identified failure points (e.g., the part of the product that failed) can be identified by a machine learning model, and the failed part of the product can be further analyzed. Production analysis data may include production analysis data for the adhesion accelerator layer deposition process. Therefore, production analysis data for the adhesion accelerator layer deposition process may include, but is not limited to: mixing ratio, curing time, time between the deposition of the adhesion accelerator layer and the deposition of subsequent layers, humidity, room temperature, and other conditions monitored during the adhesion accelerator layer deposition process.
[0050] At step 463, method 460 may include: providing production analysis data for the product to a machine learning model used to produce the product, to compare multiple data points from the production analysis data with a specified range for the product. Providing the production analysis data to the machine learning model may include using the machine learning model to compare the production analysis data for the product with production analysis data for multiple other products. These other products may be produced in a similar manner to the product and may be manufactured at the same or different manufacturing locations. For example, the same type of product may be produced or manufactured at multiple different manufacturing locations. In this way, the machine learning model may identify other products that include the same or similar production analysis data to determine when production conditions exceed threshold production conditions used to manufacture the product. Furthermore, the machine learning model may use the production analysis data to identify the causes of identified failures. For example, the machine learning model may identify one or more production data values within the production analysis data that exceed a threshold range that can be determined or calculated by the machine learning model.
[0051] Training the machine learning model may also include identifying a first plurality of products whose production analysis data falls within a specific range that includes production analysis data for products identified as faulty or defective. Identifying the first plurality of products may include comparing production analysis data for multiple similar products or for a specific type of product. The first plurality of products may be identified as similar manufactured products produced under conditions similar to the product being analyzed (e.g., the product being analyzed). The first plurality of products may be products manufactured under the same or similar production conditions. In this way, the first plurality of products can be analyzed or used to train a machine learning model to identify products that are identified as faulty, including those that are the same as or similar to the product being analyzed. The first plurality of products may have different degrees of identified faults, and by comparing the degree of identified faults with the production analysis data for the product and the first plurality of products, the cause of the identified faults can be determined.
[0052] In these examples, training the machine learning model also includes identifying a second plurality of products that include the identified failures. Identifying the second plurality of products may include comparing the identified failures of a particular product (e.g., the product to be analyzed, etc.) with defects or failures associated with a plurality of other products. The second plurality of products may be identified as similar failed products that include the same or similar types of failures as the particular product to be analyzed. The machine learning model is able to identify the second plurality of products based on the characteristics of the identified failures. For example, “bubbling” or “bumps” on the surface of a product may have specific characteristics or properties. In this example, the machine learning model may identify a second plurality of products that include similar “bubbling” or “bumps”. In these implementations, the machine learning model may identify production analysis data associated with the second plurality of products to identify potential product failures. For example, production analysis data of products similar to the second plurality of products may be identified as causes of failure. Identified failures and / or identified causes of failures may be used to update and / or train the machine learning model.
[0053] Method 460 may include comparing the identified failure of the product with the failure status of a first plurality of products. In this manner, the identified failure may be provided to a machine learning model, which may compare the identified failure with the failure status of the first plurality of products. In some embodiments, a portion of the first plurality of products may be identified as having the same or similar production analysis data and the same or similar defects as a product being analyzed that has an identified failure. As described herein, the quality data or failure status of the first plurality of products may be compared with the identified failure to determine whether the same or similar type of failure exists between the product and the first plurality of products. As used herein, the failure status may include an indication of whether the product has an identified failure or not. In other embodiments, the failure status may include an indication of the type of failure if an identified failure exists. In these embodiments, the production analysis data of the product is compared with production analysis data corresponding to a second plurality of products. As described herein, the production analysis data or production conditions of the second plurality of products may be compared with the production analysis data or production conditions of the product to identify similarities. The identified similarities may be further analyzed to determine whether these production conditions are a cause or potential cause of the identified failure.
[0054] Machine learning models can be trained using production analytics data from multiple data sources associated with various production sites (e.g., manufacturing sites) that produce components of the product or similar components related to the product. In some examples, different production sites may manufacture or produce a portion of the product. In these examples, the production analytics data can be generated in a uniform manner to allow for federated use of this data. That is, data from multiple different production sites can be used to generate and / or train machine learning models.
[0055] As described herein, when a product includes an adhesion-promoting layer, production analysis data from multiple data sources includes droplet distribution data and layer uniformity data. Droplet distribution data and / or layer uniformity data can be used to determine the consistency of a particular adhesion-promoting layer at specific locations on the application surface. In this way, the product may have been manufactured using a specific portion or segment of the application surface, and the consistency of the adhesion-promoting layer within that specific portion or segment can be identified and associated with the product, while other unused portions or segments of the product can be associated with other products. Production analysis data may include environmental data and process sequence data associated with the time of adhesion-promoting layer deposition. Environmental data may include, but is not limited to: humidity at the manufacturing site, temperature at the manufacturing site, concentration of specific substances at the manufacturing site, and / or other characteristics of the site related to the environment or conditions of the manufacturing site.
[0056] As described herein, process sequence data can refer to the amount of time between the deposition of the adhesion-promoting layer and the deposition of the subsequent layer. In some examples, the application of the subsequent layer may be timed to occur during a specific stage of the curing process. As described herein, the curing time of the adhesion-promoting layer can be monitored and may be affected by mixing ratios and / or environmental conditions. Process sequence data, together with curing time data, can be used to determine what curing stage the adhesion-promoting layer is in when the subsequent layer is deposited. Therefore, process sequence data can be used to determine the curing stage of the adhesion-promoting layer.
[0057] At step 464, method 460 may include identifying data from production analytics data that exceeds a specified range identified by the machine learning model. In some examples, the production analytics data may be specific data that compares the product with multiple other products, including the same or similar products that have been identified as faulty. In other examples, the specified range identified by the machine learning model may be generated based on previously collected data related to other products similar to the product.
[0058] At step 465, method 460 may include identifying the cause of identified failures during the production of the product based on the determined data. The determined data may be outside the scope defined by the machine learning model. For example, the machine learning model may generate a scope for production analysis data based on historical production analysis data and / or identified failures. In this example, the machine learning model may determine that one or more data points are outside these generated scopes, and determine that the cause of the failure is due to one or more data points being outside the generated scope.
[0059] Figure 5 An example of a machine-readable medium 570 for production analysis modeling for product quality inspection is illustrated. The machine-readable medium 570 is communicatively connected to a processor resource 571 via a communication path 572. In some examples, the communication path 572 may include a wired or wireless connection that allows communication between devices and / or components within a single device. As used herein, the processor resource 571 may include, but is not limited to, a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a metal programmable cell array (MPCA), a semiconductor-based microprocessor, or other combinations of circuitry and / or logic for coordinating the execution of instructions 573, 574, 575, and 576. In a specific example, the processor resource 571 utilizes a non-transitory computer-readable medium storing instructions 573, 574, 575, and 576, which, when executed, cause the processor resource 571 to perform corresponding functions.
[0060] Machine-readable storage medium 570 can be an electronic, magnetic, optical, or other physical storage device that stores executable instructions. Therefore, a non-transitory machine-readable medium (MRM) (e.g., machine-readable medium 570) can be, for example, a non-transitory MRM including random access memory (RAM), read-only memory (ROM), electrically erasable programmable ROM (EEPROM), a storage drive, an optical disk, etc. Machine-readable medium 570 can be disposed within a controller and / or computing device. In this example, executable instructions 573, 574, 575, and 576 can be "installed" on the device. Additionally and / or alternatively, machine-readable medium 570 can be a portable, external, or remote storage medium, for example, that allows a computing system to download instructions 573, 574, 575, and 576 from a portable / external / remote storage medium. In this case, the executable instructions can be part of an "installation package".
[0061] Machine-readable medium 570 includes instructions 573 for training a machine learning model for the production process of a product type using production analytics data captured during the production process of a product type at multiple different locations. In these examples, the machine learning model correlates the production analytics data with the quality level of the product type. As described herein, the product type may include an application layer or application surface having an adhesion-promoting layer applied during the production process.
[0062] Machine-readable medium 570 may include instructions for generating a machine learning model of the product's production process using production analytics data captured during the production process at multiple different locations. The machine learning model can correlate the production analytics data with the product's quality level. As described herein, the machine learning model can be generated or trained using production data provided by multiple different manufacturing sites. Different manufacturing sites may each be instructed to collect data in the same or similar manner. For example, different manufacturing sites may be instructed to collect the same type of data (e.g., infrared images, hyperspectral images, etc.). In some examples, different manufacturing sites may be instructed to collect data at specific times (e.g., the amount of time after depositing a specific layer, etc.). Furthermore, different manufacturing sites may be instructed to collect different environmental data during the product's production.
[0063] Data collected from different manufacturing sites can be organized into production analytics data, which can be provided as federated data to train or generate machine learning models. Different manufacturing sites can be instructed to analyze raw data and provide specific analyses to the machine learning model. For example, different manufacturing sites can be instructed to calculate the droplet dispersion percentage when a specific portion of the substrate is transferred from a first dispersion device to a second dispersion device. In this way, the droplet dispersion percentage can be provided to the machine learning model without providing the raw data, which may contain sensitive information. In this way, different manufacturing sites can each provide data to the machine learning model without risking the provision of sensitive information. In some implementations, raw data refers to unprocessed, unfiltered, and unmodified data collected directly from the source. For example, raw data could be unaltered or unupdated image data collected directly from an imaging device. This type of raw data may include sensitive information that may not be wished to be disclosed to external sources.
[0064] Machine-readable medium 570 includes instructions 574 for providing site-specific production analysis data to a machine learning model, wherein the production analysis data includes image data and environmental data over a period of time during the production of this type of product at the site. In some examples, production analysis data from different manufacturing sites may be used to generate or train the machine learning model. The trained machine learning model can be used to analyze the production of this type of product at the site. The site may be one of these different manufacturing sites; however, it may also be another site for which production analysis data was not provided to train the machine learning model. The site-specific production analysis data may be compared with production analysis data from different sites to determine whether the site is producing the product within the range of data calculated by the machine learning model.
[0065] Machine-readable medium 570 includes instructions 575 for identifying data in a production process, including values exceeding the range determined by a machine learning model, based on production analysis data for a specific location. Multiple data measures captured during the production process at multiple different locations may include thermal imaging data to determine droplet distribution and layer uniformity. As described herein, data measures captured during the production process at multiple different locations may include production settings utilized during the production process. For example, production settings may be adjustable settings that define production parameters for distributing chemical layers onto surfaces of mechanical components. Furthermore, production settings may be settings used to collect image data or other data to be used as production analysis data.
[0066] Multiple data metrics captured during production processes at multiple different sites may include hyperspectral imaging data used to determine chemical formulation data for chemical substance layers. Multiple data metrics captured during production processes at multiple different sites may also be federated data received from multiple different sites. As used herein, federated data refers to an approach in which data from multiple sources or entities remain separate and distributed, while still allowing for collaborative analysis of the combined data and the insights gained. Instead of centralizing data in a single location or organization, a federated data system enables data to remain decentralized and distributed across various independent entities or locations. In some implementations, each entity or organization retains control and ownership of its own data, thereby protecting data privacy and security. The federated approach does not involve directly sharing or aggregating data, but rather involves implementing protocols that enable data analysis and processing across distributed data sources without requiring data movement or direct access.
[0067] Machine-readable medium 570 includes instructions 576 for generating notifications to change production settings for a site-specific production process, thereby changing values to within a range determined by a machine learning model. The notification may be an instruction to change specific production settings of the production process. In other embodiments, the instructions may be executed without human interaction. Changing the production process settings may include sending a notification to a control panel at the site to instruct the site to change specific settings of the production process (e.g., distribution disk rotation speed, chemical temperature, and flow rate), which may be adjusted by the control panel or by a human user. In some examples, the adhesion-promoting layer may include multiple chemicals that may be distributed by a rotating disk of a distribution device. The rotation speed of the rotating distribution disk may change the droplet distribution of the adhesion-promoting layer. In other examples, changing the settings may include updating the production process range utilized by the site. For example, the site may utilize a first droplet dispersion percentage range, and the machine learning model may indicate that the site should instead utilize a second droplet dispersion percentage range.
[0068] In other examples, changing production process settings may include modifying the production process at a specific location. For instance, a control panel may be accessible, and production process settings may be changed. In this example, droplet dispersion device settings may be changed to alter the droplet dispersion percentage to a value within the range of droplet dispersion percentages defined by the machine learning model. Other types of settings or processes may be changed based on identified causes of failure.
[0069] Machine-readable medium 570 may include instructions for updating a machine learning model using production analytics data captured at a specific location. The production analytics data from that specific location can be used to update the machine learning model. In these embodiments, the updated machine learning model can modify production processes at multiple different locations. For example, the data range for specific data in the production analytics data may change based on the production analytics data provided by the specific location. In this example, the updated range may be provided to multiple different locations.
[0070] Figure 6 An example of an apparatus 601 for production analysis modeling for product quality inspection is illustrated. In some examples, apparatus 601 is a computing device that includes processor resources 671 and machine-readable medium 670 for storing instructions 603, 605, 607, 609, 611 that are executed by processor resources 671 to perform specific functions. Figure 6 This illustrates how a computing device can execute instructions to perform the functions described herein. Device 601 can be a machine learning model operating on a computing device.
[0071] Device 601 can be communicatively coupled to imaging device 612 via communication path 684. As described herein, imaging device 612 can capture an image of the application surface 604 when a chemical layer (e.g., an adhesion-promoting layer, etc.) is applied to the application surface 604. The captured image from imaging device 612 can be transmitted to device 601 via communication path 684, where the captured image can be analyzed.
[0072] Device 601 includes instructions 603 stored by machine-readable medium 670, which are executed by processor resource 671 to monitor received image data of the application surface during the application of the adhesion-promoting layer on the application surface at the production site. The image data can be received from imaging device 612 via communication path 684. The image data can be infrared image data, hyperspectral image data, visual image data, and other types of image data. The image data can be generated by machine learning models (e.g., such as...). Figure 1 The machine learning model cited in the text (114, etc.) and / or edge computing devices (e.g., such as...) Figure 1 The analysis was performed on edge computing devices (such as 116) referenced in the text.
[0073] The apparatus 601 includes instructions 605 stored in a machine-readable medium 670, which are executed by a processor resource 671 to calculate production analysis data associated with applying an adhesion-promoting layer to an application surface based on image data. The production data may include, but is not limited to: production setup, environmental characteristics of different locations over a period of time, droplet distribution of the chemical layer, layer uniformity of the chemical layer, and the amount of time between depositing the chemical layer and depositing different layers on the chemical layer, as well as other data that may be collected during the application of the adhesion-promoting layer.
[0074] Apparatus 601 includes instructions 607 stored by machine-readable medium 670, which are executed by processor resource 671 to provide production analysis data to a machine learning model, thereby comparing the production analysis data with threshold ranges of the machine learning model. In these embodiments, the machine learning model utilizes production data and quality data from multiple different production sites that produce products including an applied surface and an applied adhesion-promoting layer. In some embodiments, the machine learning model may correlate production analysis data with multiple quality data associated with a product including an applied surface. As described herein, multiple production data and / or production analysis data may be correlated with multiple quality data. In this way, production data can be combined with quality data such that specific quality data for a particular product can be used with production data for that particular product. Production data and quality data may be correlated under a specific product identifier to associate a specific product with the production data and quality data.
[0075] Device 601 includes instructions 609 stored by machine-readable medium 670, which are executed by processor resource 671 to identify when data values of production analysis data exceed a threshold range of production data identified by a machine learning model. As described herein, production data can be monitored in real time and for periods of time that might not be possible without imaging equipment (such as imaging device 612). For example, a surface area with an applied adhesion-promoting layer can be transported to deposit or attach an additional layer, such that the additional layer is applied within a threshold time period. In this way, using a machine learning model to monitor production data can prevent products from being delivered to customers.
[0076] As described herein, multiple production data points and multiple quality data points can be provided to a machine learning model as production analytics data. The production analytics data can be used as federated data for multiple different production sites. The multiple production data points and multiple quality data points provided to the machine learning model may not include raw data collected by imaging device 612 (e.g., excluding initial data, etc.). As used herein, raw data refers to data collected during the production process. Production data, quality data, and / or production analytics data may not include the actual data collected, but rather include computational results obtained using the actual data collected. In this way, a higher level of security is provided for multiple production sites. The multiple production data points and multiple quality data points can be provided to the machine learning model to train the machine learning model.
[0077] Device 601 includes instructions 611 stored by machine-readable medium 670, which are executed by processor resource 671 to generate notifications based on identified production data to change the settings for applying the adhesive promoting layer. As described herein, device 601 may generate notifications that the settings for applying the adhesive promoting layer exceed specific production data and may be defective. In some embodiments, device 601 may identify a manufactured product or a portion of a manufactured product as defective and send notifications that the manufactured product or a portion of a manufactured product is defective. In some cases, the identified product may be prevented from shipment or removed from the production of another product.
[0078] Device 601 includes instructions stored by machine-readable medium 670, which are executed by processor resource 671 to receive a product review from a machine learning model. This product review includes changes to the application of an adhesive-promoting layer to the application surface to alter production data and improve the quality level of the product, including quality levels. The product review may include analysis of quality and production data for a specific product or product range. The product review may include analysis of the expected quality of the product based on production data for a specific product. In other embodiments, the product review may include analysis of the causes of identified product failures based on production data. In this way, the product review can compare the product's production data with production and quality data from different products.
[0079] Device 601 may include instructions stored by machine-readable medium 670, which are executed by processor resource 671 to provide multiple production data and multiple quality data to update a machine learning model associated with a product including the applied surface. In these embodiments, the machine learning model may utilize production data and quality data from multiple different production sites that produce the product including the applied surface. In other embodiments, device 601 may provide a product review request to the machine learning model for a product including a quality level. As described herein, a product review request may be a request to analyze the cause of an identified failure. However, a product review request may be a request to analyze a specific product or product range to determine whether the product or product range is manufactured within the current set of production ranges determined by the machine learning model.
[0080] Device 601 may include instructions stored by machine-readable medium 670, which are executed by processor resource 671 to receive multiple threshold ranges based on multiple production data and multiple quality data provided by multiple different production sites. As described herein, the multiple threshold data may include threshold ranges that can be applied to multiple production data during the production of the product. The multiple threshold ranges may be updated or modified at one or more of the multiple different production sites to improve the quality data of the produced product. In specific examples, the multiple threshold ranges may include, but are not limited to: droplet dispersion threshold, mixing ratio threshold, curing time threshold, sequence threshold between different applications or operations, and other thresholds.
[0081] Device 601 may include instructions stored by machine-readable medium 670, which are executed by processor resource 671 to monitor received applied image data during the application of the adhesion-promoting layer on an application surface at a production site. As described herein, imaging device 612 may be used to monitor production data in real time during the production of a product. In these examples, device 601 may monitor production data of the adhesion-promoting layer based on the received image data. Monitoring production data may include comparing the production data with production thresholds identified by a machine learning model.
[0082] Device 601 can monitor environmental data at the production site during the application of the adhesion-promoting layer and generate a notification when either the production data or the environmental data exceeds a threshold range provided by a machine learning model. As described herein, an edge computing device for the production site can generate a notification when either the environmental data or the production data exceeds a threshold range identified by a machine learning model. Device 601 can also generate a notification when a combination of environmental data and production data exceeds a combined threshold range. For example, environmental data such as external temperature may exceed a specific threshold, which alters the threshold for the curing time of the adhesion-promoting layer. In this example, the combination of environmental data and a specific curing time may exceed a threshold for the combination of environmental data and production data. Other combinations of environmental data and production data can be used to generate notifications.
[0083] Device 601 may include instructions stored by machine-readable medium 670, which are executed by processor resource 671 to receive federated data instructions from a machine learning model. This set of federated data instructions indicates parameters for calculating multiple production data and collecting image data. The set of federated instructions may be instructions from the machine learning model regarding how to generate federated data that can be provided to the machine learning model. As described herein, product analytics data may be received from multiple production sites, each generating production analytics data according to specifications from the machine learning model. In this way, the machine learning model can be used to collect and compare data. Furthermore, the collected data can protect the privacy of the production sites providing the data.
[0084] In this manner, federated data instructions may include instructions on how to configure imaging device 612 such that similar portions of application surface 604 are captured by multiple different production sites. Similarly, lighting or other settings for capturing data can be standardized across multiple different production sites. Calculations used to determine production data and / or quality data can be standardized across multiple different production sites via federated data instructions. In these embodiments, this set of federated data instructions includes instructions for positioning imaging device 612 relative to application surface 604 and for positioning light sources relative to application surface 604. Other types of federated instructions may be provided to different production sites to ensure that data collected and provided to machine learning models is standardized.
[0085] Although specific embodiments have been described above, these embodiments are not intended to limit the scope of this disclosure, even where only a single embodiment is described with respect to a particular feature. Unless otherwise stated, the examples of features provided in this disclosure are intended to be illustrative and not restrictive. The above description is intended to cover such alternatives, modifications, and equivalents that will be obvious to those skilled in the art who benefit from this disclosure.
[0086] The scope of this disclosure includes any feature or combination of features disclosed herein (express or implicit), or any generalization thereof, whether or not it alleviates any or all of the problems addressed herein. Various advantages of this disclosure have been described herein; however, embodiments may provide some, all, or none of these advantages, or may provide other advantages.
[0087] In the foregoing specific embodiments, for the purpose of simplifying this disclosure, some features are combined in a single embodiment. This approach of the disclosure should not be construed as reflecting an intention that the disclosed embodiments of the disclosure must use more features than are expressly recited in each claim. Rather, as reflected in the following claims, the subject matter of the invention does not consist of all features of a single disclosed embodiment. Therefore, the following claims are hereby incorporated into the specific embodiments, wherein each claim exists independently as a separate embodiment.
Claims
1. A method, the method comprising: Receive review requests, including for products that have been marked as expired; Identify production analysis data related to the production process for the product, wherein the production analysis data includes data captured during the production of the product; The production analysis data for the product is provided to a machine learning model used to produce the product to compare the production analysis data with a specified data range for the product. Identify data from the production analysis data that exceeds the specified range identified by the machine learning model; as well as The identified data is used to pinpoint the cause of the identified failure during the production of the product.
2. The method according to claim 1, further comprising: The machine learning model is trained using production analysis data from multiple data sources associated with multiple different production sites that produce the components of the product or similar components associated with the product.
3. The method of claim 2, wherein when the product includes an adhesion-promoting layer, the production analysis data from the plurality of data sources includes droplet distribution data and layer uniformity data.
4. The method of claim 3, wherein the production analysis data includes environmental data and process sequence data associated with the time of deposition of the adhesion-promoting layer.
5. The method of claim 1, wherein the production analysis data for the product having the adhesion-promoting layer includes: The droplet distribution data, layer uniformity data, and environmental data at specific points in time during the production of the product at a particular production site.
6. The method according to claim 1, further comprising: The machine learning model is trained by identifying a first plurality of products, including corresponding production analysis data falling within the scope of the production analysis data of the product, as products with similar manufacturing conditions.
7. The method according to claim 6, further comprising: The machine learning model is trained by identifying a second or more products, including the already identified failed products, as similar failed products.
8. The method according to claim 7, further comprising: The machine learning model is trained using the following steps: The identified failure of the product is compared with the failure status of the first plurality of products to identify products from the first plurality of products that include the same failure type as the identified failure. as well as The production analysis data of the product is compared with the production analysis data of the second plurality of products to identify products from the second plurality of products that include production analysis data falling within the range of the production analysis data of the product.
9. A machine-readable medium storing machine-readable instructions that, when executed by a processor of a device, cause the processor to: A machine learning model for the production process of a type of product is trained using production analysis data captured during the production process of a product at multiple different locations, wherein the machine learning model correlates the production analysis data with the quality level of the product type. The production analysis data for a specific location is provided to the machine learning model, wherein the production analysis data includes image data and environmental data over a period of time during the production of the type of product at the specific location; Identify data in the production analysis data for the specific location that includes values outside the defined range of the machine learning model; as well as A notification is generated to change the production settings of the production process for the specific location, thereby changing the value to the range determined by the machine learning model.
10. The machine-readable medium of claim 9, wherein the production setup of the production process comprises at least one of the following: Distributed disk positioning; Distribute disk rotation speed; Chemical substances at temperature; and Chemical flow rate.
11. The machine-readable medium of claim 9, wherein the production analysis data captured during the production process at the plurality of different locations includes thermal imaging data or hyperspectral image data for determining the droplet distribution and layer uniformity of the chemical layer of the product of the type.
12. The machine-readable medium of claim 11, wherein the production analysis data captured during the production process at the plurality of different locations includes hyperspectral imaging data for determining chemical formulation data for the chemical substance layer.
13. The machine-readable medium of claim 9, wherein the production analysis data captured during the production process at the plurality of different locations is federated data received from the plurality of different locations.
14. The machine-readable medium of claim 9, wherein the machine-readable medium includes instructions for updating the machine learning model using production analysis data captured at the particular location.
15. A system comprising: An imaging device for capturing image data of the application surface during the application of the adhesion-promoting layer on the application surface at a production site; The device is configured to: During the application of the adhesion-promoting layer to the application surface at the production site, image data of the application surface is monitored and received. Based on the image data, production analysis data associated with applying the adhesion-promoting layer to the application surface is calculated; The production analysis data is provided to a machine learning model to compare the production analysis data with threshold data of the machine learning model, wherein the machine learning model utilizes production data and quality data from multiple different production sites that produce products including the applied surface and the applied adhesion promoting layer. The system identifies when the production data in the production analysis data exceeds a threshold range identified by the machine learning model. as well as A notification is generated based on identified production data to change the production settings for applying the adhesion-promoting layer.
16. The system of claim 15, wherein the device is configured to receive a set of instructions indicating parameters for calculating the production analysis data.
17. The system of claim 16, wherein the set of instructions includes instructions for positioning the imaging device relative to the applying surface and for positioning the light source relative to the applying surface.
18. The system of claim 15, wherein the production data and the quality data provided to the machine learning model do not contain raw data collected from the plurality of different production sites.
19. The system of claim 15, wherein the device is configured to receive a plurality of updated threshold data based on the production data and the quality data provided by the plurality of different production sites.
20. The system of claim 15, wherein the device is configured to: Environmental data of the production site are monitored during the application of the adhesion-promoting layer; and The notification is generated when the environmental data exceeds the threshold range provided by the machine learning model.
