Inspection and production of printed circuit board assemblies

Abstract

A method of inspecting a printed circuit board, PCB, assembly (C) comprising the steps of: acquiring an image (IM) of the PCB assembly (C) and analyzing the image (IM), wherein the analyzing comprises object-based analysis of the image (IM) to recognize at least one component (A) placed on the PCB (B), wherein the object-based analysis is performed on the basis of an object-based analysis program, wherein the object-based analysis program comprises a trained machine learning model (ML); and determining, by the object-based analysis program, whether the at least one component (A) is placed on the PCB (B) on the basis of a comparison between findings of the object-based analysis and stored assembly information of the PCB (B); and outputting, by the object-based analysis program, an error, preferably displaying an image of the PCB assembly and error information related to a missing or wrongly placed component or a wrongly placed component on the PCB, if one or more components are missing or wrongly placed or one or more wrong components are placed on the PCB as determined by the object detection analysis; inputting, by a worker, a result of a visual inspection of the PCB assembly, the result of the visual inspection indicating a false error of the object detection analysis; writing one or more settings for soldering the PCB assembly (C) by a soldering device.

Description

Technical Field

[0001] This disclosure relates to printed circuit board (PCB) assemblies and their production via soldering. More specifically, this disclosure relates to the inspection of PCB assemblies during the production process. Furthermore, this disclosure relates to the fields of artificial intelligence and machine learning and their industrial applications. Background Technology

[0002] As electronic devices become smaller and packaged with higher density, automated inspection of printed circuit board (PCB) assemblies is becoming increasingly important. In the long run, automated inspection outperforms manual inspection in terms of consistency, speed, and cost.

[0003] A printed circuit board (PCB) uses conductive traces, pads, and other features etched from one or more laminated copper layers onto and / or between layers of a non-conductive substrate to mechanically support and electrically connect electrical or electronic components. Components are typically soldered onto the PCB to electrically connect and mechanically secure them to the PCB.

[0004] Common defects on PCB components include missing components, misalignment, component tilting, component standing up / open circuit, component error, value error, bridging / short circuit, bent leads, incorrect polarity, extra components, raised leads, insufficient solder, and excessive solder.

[0005] US20150246404A1 discloses a power supply unit, control unit, management device, and power supply control device for a welding system.

[0006] EP0871027A2 discloses the inspection of printed circuit board assemblies, and KR20090049009A discloses an optical inspection apparatus and method for printed circuit boards. Summary of the Invention

[0007] Today, with the wide variety of PCB components to be produced, workers assembling PCBs with electrical components face a large number of different parts to be mounted on the same or similar PCB types. This confusion between different layouts can lead to malfunctions when components are placed on specific PCBs. Typically, PCB assemblies are only inspected after components have been soldered onto the printed circuit board. This results in many PCB assemblies being discarded, leading to material loss and waste.

[0008] Therefore, the object of this invention is to improve the use of materials, simplify the production process, and thereby reduce the number of defective PCB components produced.

[0009] This objective is achieved through the following aspects.

[0010] According to the first aspect, the objective is achieved by a method for inspecting a printed circuit board (PCB) assembly. The method includes the steps of acquiring an image of the PCB assembly (e.g., using a camera) and analyzing the image, wherein the analysis includes object-based analysis of the image to identify at least one component placed on the PCB. The method further includes the step of determining whether the at least one component is placed on the PCB based on a comparison between the findings of the object-based analysis and stored component information of the PCB.

[0011] According to the second aspect, this objective is achieved through a method for training a machine learning algorithm for an object-based analysis program. The method includes acquiring multiple images of a PCB assembly, preferably different types of PCB assemblies, most preferably during the production process of the PCB assembly. The method further includes selecting images from the multiple images suitable for training the machine learning algorithm. The method further includes automatically labeling the multiple images based on a template used to label the PCB assembly. The method further includes training the machine learning algorithm based on the labeled images.

[0012] According to a third aspect, this objective is achieved by an inspection system for inspecting printed circuit board (PCB) assemblies. The system includes a camera for acquiring images of the PCB assembly. The system further includes a control unit for analyzing the images, wherein the analysis includes object-based analysis of the images to identify at least one component placed on the PCB. The control unit is further configured to determine whether to place the at least one component on the PCB based on a comparison between the findings of the object-based analysis and stored component information of the PCB.

[0013] Other advantageous embodiments are provided in the dependent claims and are described below. Attached Figure Description

[0014] Figure 1 The diagram illustrates several steps in the PCB assembly manufacturing process, particularly the placement of electrical components and the soldering of PCB assemblies.

[0015] Figure 2 The diagram illustrates automated optical inspection during the PCB assembly manufacturing process and after electrical components are soldered onto the PCB.

[0016] Figure 3 Several steps in the manufacturing process of a PCB assembly according to a first embodiment are shown, wherein optical inspection is performed before electrical components are soldered to the PCB.

[0017] Figure 4 An image of a PCB assembly is shown, including the PCB and electrical components placed on the PCB.

[0018] Figure 5 The results of object detection analysis of the PCB assembly image are shown.

[0019] Figure 6 The system for inspecting PCB assemblies and the corresponding procedures are shown.

[0020] Figure 7 A system and corresponding steps for training a machine learning model to perform object detection on images of PCB components are shown.

[0021] Figure 8 The workflow for inspecting PCB assemblies and how to integrate the inspection into the production of PCB assemblies is illustrated. Detailed Implementation

[0022] Figure 1 Several steps in the manufacturing process of PCB assembly C are illustrated, particularly the placement of electrical components A1, A2, and SMDs, and the soldering of PCB assembly C. For the manufacturing of PCB assembly C, electrical components A1, A2, and SMDs are placed on PCB B. For example, through-hole devices A1, A2, and other electrical components A1, A2 (e.g., capacitors and / or integrated circuits) can be placed on PCB B. Alternatively, electrical components can be surface mount devices (SMDs), which can also be placed on PCB B.

[0023] Through-hole (THT) technology refers to a mounting scheme for electronic components A1 and A2. It involves using leads on the components, which are inserted into holes drilled in the PCB C via manual assembly (placement by hand) or automated insertion machines and then soldered to pads on the other side. Compared to surface mount technology, through-hole mounting provides a robust mechanical bond.

[0024] After placing electrical components A1 and A2 onto PCB B, the PCB assembly C undergoes a soldering process. For example, in... Figure 1 The diagram illustrates a wave soldering process, in which flux is first applied to PCB assembly C, followed by preheating. Finally, PCB assembly C is transported through a solder standing wave, where PCB B and components A1 and A2 come into contact with the solder.

[0025] See Figure 2 This illustrates an automated optical inspection process during the manufacturing of PCB assembly C. Typically, automated optical inspection is performed after one or more electrical components A1 and SMD are soldered to PCB B. For this purpose, image IM is captured by camera I on the bottom side of PCB assembly C.

[0026] As mentioned earlier, the increasing complexity and variety of electronic components have led to higher demands on the personnel assembling PCB C with electrical component A1. Depending on the specific circumstances, electrical component A1 may be forgotten, or the wrong component A1 may be placed on PCB B. In such cases, the inspection of PCB assembly C after soldering requires a significant amount of work to desolder PCB assembly C and remove incorrectly installed components, or, in the worst case, PCB assembly C may need to be scrapped.

[0027] Therefore, it is proposed to perform automated optical inspection of PCB assembly C after one or more electrical components A1 are placed on PCB B and before electrical components A1 are soldered to PCB B. Figure 3 The diagram illustrates several steps in the manufacturing process of PCB assembly C, where optical inspection is performed after one or more electrical components A1-A4 are soldered to PCB B.

[0028] PCB B can reach placement table 1, where workers can place electrical components A1-A4 onto PCB B. PCB B can be placed on or within tray Y for transport along the production line via conveyor F. Workers can retrieve electrical components A1-A4 from one or more shelves R1, R2 at the placement table and place components A1-A4 according to the type of PCB assembly C to be produced. Optionally, placement can be performed automatically, for example, by a robot.

[0029] Wave soldering station 3 may include a single wave (not shown). To transport components from placement station 1 or inspection station 2 to soldering station 3, a tower T for storing multiple PCB components may be provided. The tower can act as a buffer, for example, to reload the soldering machine if the placement of electrical components at the placement station takes too long. Now, before leaving the placement station or entering soldering station 3 for PCB component production, an automated optical inspection is performed at placement inspection station 2. For this purpose, images of the PCB components are acquired (e.g., using camera I). ​​The images are then analyzed, including object-based analysis of the images to identify at least one component placed on PCB B. Thus, based on a comparison between the findings of this object-based analysis and stored component information of PCB B, it is determined whether the at least one component should be placed on PCB B. The result of this comparison can be displayed to the personnel W at inspection station 2 and / or placement station 1 to replace incorrectly placed components A-A4 or to place one or more missing components A1-A4 on PCB B.

[0030] If object detection analysis confirms that all electrical components are correctly positioned, the PCB assembly can proceed to soldering station 3. For example, PCB assembly C can be placed in the tower T of the soldering equipment at soldering station 3.

[0031] However, if it is determined that not all electrical components A1-A4 are correctly positioned, the PCB assembly cannot proceed to further production steps; for example, it cannot be transported to soldering station 3.

[0032] It should be noted that for PCB assembly C to proceed with further production steps, automated optical inspection is a mandatory step; that is, all PCB assemblies C must be analyzed before production can continue. To initiate the optical inspection, the operator may need to press the button at inspection station 3.

[0033] See now Figure 4 Image IM shows PCB assembly C. PCB assembly C includes PCB B and electrical components A1-A3 placed (e.g., via THT) on PCB B. Image IM can be captured by a camera mounted at inspection table 2. Figure 4 As shown in the image IM, the top side of the PCB is where electrical components A1-A3 are located.

[0034] Object detection analysis can be performed on the image IM to identify at least one component A1-A3 placed on PCB B. Figure 5 The results of this object detection analysis are shown in the figure, where identified objects O1-O4 are highlighted in boxes. This analysis can assign a probability of correct identification to the identified objects O1-O4. If the probability is below a certain threshold, for example, below 75%, PCB assembly C and the corresponding electrical components A1-A3 may need to be checked before production of PCB assembly C can continue. This object detection analysis is a computer-implemented method for assigning at least one object O1-O4 to identified components A1-A3 on PCB B. This object detection analysis can be performed by, for example, by... Figure 5 The trained machine learning model ML described in the document is used to perform the operation.

[0035] exist Figure 6 Further details of the system used to inspect PCB assembly C and the corresponding procedures are shown in the diagram. The image IM of the PCB assembly can be obtained from, for example... Figure 3A camera is placed at inspection station 2 to capture images for object detection. The machine learning model (ML) can be hosted in a virtual machine within an operating system (such as Windows 10). The ML model itself can be part of a container (such as a Docker container) running on the virtual machine. The image IM can be processed by the ML model, overlaying the identified objects onto the acquired image IM, which is then displayed to, for example, the staff at the inspection station. Furthermore, a list of identified electrical components can be displayed to the staff on a monitor. This list of electrical components can be retrieved from a database (DB1) or a planning system (such as Teamcenter).

[0036] If the object detection identifies all electrical components to be placed on the PCB, the PCB assembly can proceed to the next production step. For this purpose, the inspection results can be written to tag G. For example, settings for one or more steps for subsequent soldering of the PCB assembly can be written to the tag. Tag G can be attached to the tray Y containing the PCB assembly. For example, tag G can be an RFID tag, including rewritable memory. In particular, the PCB type or PCB identifier can be written to the tag. The PCB assembly can then be transported to the soldering station, for example, as... Figure 3 As shown, the PCB type and / or PCB identifier can be read from the label here. Based on the settings on label G, the soldering equipment can adjust the soldering process. To control the soldering process, the soldering equipment may include a soldering program. The settings of the soldering program may include the temperature settings of a part of the soldering equipment (e.g., soldering iron, tweezers, micro tweezers, desoldering iron, and hot air, etc.).

[0037] If object detection fails to identify all electrical components to be placed on the PCB, the process is halted, and the PCB assembly is repaired, for example, by swapping one or more electrical components on the PCB or by placing one or more additional components on the PCB. After repairing the PCB assembly, a new image of the PCB assembly is acquired, and object detection is rerun for the repaired PCB assembly. In this case, the corresponding code can be written into a label or field to identify the PCB, and / or the corresponding settings for soldering can be (intentionally) left blank. Production can then be stopped, at least until the PCB assembly reaches the soldering station (e.g., where the label on the tray is read). In this case, a worker may need to remove the PCB assembly from the tray before the production process can continue.

[0038] Unlike the PCB assembly repair described above, object detection analysis may fail. In other words, the object detection analysis may identify one or more missing components or one or more incorrectly placed components on the PCB. In that case, workers can identify false errors by releasing the corresponding (virtual) button. Afterward, one or more settings for soldering the PCB assembly using soldering equipment can be written to a label.

[0039] Therefore, as mentioned above, object detection analysis is a computer-implemented method for image processing used to detect instances of one or more (semantic) objects of a certain class in one or more (digital) images. Machine learning (ML) models in the form of computer programs can be used for this object detection analysis. For example, this object detection analysis can be used to detect one or more components placed on a PCB. As a result, the object detection analysis can provide an identifier and coordinates representing each component detected on the PCB. The results of the object detection analysis can then be compared with the PCB's component information (e.g., a bill of materials (BOM)). The component information can be a list of components required to manufacture the PCB assembly. Therefore, by comparing the objects discovered by the object detection analysis with the component information, one or more missing components can be identified. Furthermore, it can be determined that one or more incorrectly placed components are on the PCB. Even further, it can be discovered that one or more components are incorrectly placed on the PCB.

[0040] For example, component information can be provided in the form of a file (such as an XML file), including a list of parts and coordinates associated with each part of the PCB assembly to be manufactured. Below is an exemplary excerpt of component information that can be stored as a file:

[0041]

[0042]

[0043] Here, components_1 and component_2 are part of the PCB assembly to be manufactured and are assigned corresponding coordinates by the bounding box “bnbdbox”. These coordinates represent the position of the respective component on the PCB (e.g., relative to a reference point on the PCB). For example, the PCB, and consequently an image of the PCB, may include one or more reference points. Such reference points are also called reference markers or marker points.

[0044] As a result of comparing the findings of object detection analysis with component information, the placement of components on the PCB can be verified. It is possible that, through comparison (e.g., via the object detection analysis), one or more components are missing, or one or more incorrectly placed components are identified on the PCB. This may occur when there is a discrepancy between the objects identified by the object-based analysis and the provided component information. In that case, the results of the comparison can be output (e.g., displayed) on the monitor of the inspection station. This output may include error information related to the missing or incorrect components or incorrectly placed components. For example, the identified objects may be overlaid on an image acquired and displayed to, for example, a worker at the inspection station. The error information may be in the form of a colored rectangle or box identifying the missing, misplaced, or incorrect component. The error information may be displayed on the monitor of the inspection station, for example, overlaid on an image of the printed circuit board. Optionally or additionally, the error information may identify the missing, misplaced, or incorrect component in the form of text (e.g., indicating "component_l").

[0045] In addition, markers indicating pass or fail can be displayed to personnel at, for example, inspection stations. Markers indicate incorrect component placement on PCB assemblies. Markers can be associated with images.

[0046] Then, the PCB assembly can be inspected by an operator (also shown as a worker) through visual inspection. This allows the operator to determine whether errors detected by object detection analysis are true or false errors. Input fields are set up on the display for this purpose. The operator can input the results of the visual inspection, preferably via the display on the inspection station, by releasing, i.e., pressing the corresponding (virtual) button on the inspection station.

[0047] If it is a false error, the image label can be changed from error to pass or false error. This allows PCB assembly manufacturing to continue. Therefore, subsequently, for example based on the results of a visual inspection, one or more settings for soldering the PCB assembly by soldering equipment are written to a label, for example, that can be attached to the tray containing the PCB assembly. Thus, the writing of one or more settings can be based on the results of a worker's visual inspection.

[0048] In cases where genuine errors are discovered by staff through visual inspection, misplacement can be corrected by the staff. PCB manufacturing can also continue by writing one or more settings for soldering PCB components using soldering equipment onto a label that can be attached to the tray containing the PCB component. Therefore, faulty or defective PCB components are not manufactured. Furthermore, the marking of images on PCB components has been improved.

[0049] Now, if a genuine error is discovered through visual inspection by staff, the object detection analysis can be performed again to provide feedback on whether the corrective action (e.g., repositioning of one or more parts) was successful. Thus, new findings or results from the object detection are obtained and displayed to the staff, who can then release the (virtual) button at the inspection station again, for example, to confirm that the part is now correctly positioned or that a false error has occurred again.

[0050] Additionally, for object detection, the acquired images of the PCB components can be stored in the database DB2. Images stored in DB2 can be used to (re)train the machine learning model ML for object detection analysis. Therefore, multiple images IM can be acquired during the production process of PCB component C to (re)train the machine learning model ML.

[0051] See Figure 7 This paper illustrates a system and corresponding steps for training a machine learning model to perform object detection on images of PCB assembly C. During the production of PCB assembly C, images IM1, IM2, and IM3 of PCB assembly C can be acquired and stored in a database DB2 for image data acquisition purposes. To effortlessly label images IM1, IM2, and IM3 and use them for training the machine learning model ML, images IM1, IM2, and IM3 can be loaded into or read by the automatic labeling tool ALT. The automatic labeling tool ALT labels images IM1, IM2, and IM3. Unlike manually labeling all acquired images IM1, IM2, and IM3, a one-time labeling method is used. For this purpose, templates are used to label images IM1, IM2, and IM3. Automatic labeling can be based on a template matching algorithm, which detects the offset of the PCB (of the template image) relative to the image boundary of each of images IM1, IM2, and IM3. By doing so, the labels defined in the template image are transformed from the image coordinate system to the PCB coordinate system of each image (of the template), thereby enabling the algorithm to automatically label each image in the database DB2 and subsequently use the automatically labeled images to train the ML object detection algorithm. For example, one or more reference points can be detected on each image in the image. A template can be set based on these reference points. In other words,

[0052] The template may include one or more predefined or preset coordinates used to identify one or more components. Similarly, as described above regarding component information, the template may be in the form of a file, such as an XML file. An excerpt of a template is shown below:

[0053]

[0054]

[0055] Now, to automatically match the template with each image in the image (thus labeling the parts in the image), the offset can be calculated separately using reference points for each image. The offset can be calculated based on the distance of one or more reference points in the image relative to one or more image boundaries, such as those of images IM1, IM2, and IM3. This determines the position, i.e., coordinates, of one or more parts in each image, thus performing automatic labeling of the parts in the image. It can be seen that a box or rectangle is defined by the four coordinates of each part, thereby identifying the position of each part in the image. Alternatively, the image boundaries can be adjusted to make them coincide with the reference points in the image. Adjustment of the template coordinates may be necessary due to the placement and position of the PCB in the tray. This occurs because the position of each PCB in its respective tray is different.

[0056] The template used for labeling can be an image that has already been manually labeled. Then, an automatic labeling tool converts the template's labels into one or more images IM1, IM2, IM3 previously stored in database DB2. Therefore, the acquired images IM1, IM2, IM3 do not need to be manually labeled; instead, images suitable for automatic labeling are selected and stored in database DB2. The selection of suitable images can be done automatically based on one or more predetermined criteria or manually by the user. Thus, labeling associates one or more objects detected in the images with one or more electrical components.

[0057] Once images IM1, IM2, and IM3 are labeled, i.e., the object or electrical component is identified, the machine learning model can be (re)trained based on the now labeled images IM1, IM2, and IM3.

[0058] After training the machine learning model (ML), it can be deployed on an industrial PC or integrated into a production system used to produce one or more PCB components, for example, by integrating it into existing infrastructure. For instance, the machine learning model (ML) can be deployed on the control unit of an inspection system, for example, to control the placement of inspection tables. The inspection system or inspection table itself can be integrated into the production system used to produce PCB components. The production system includes, for example, placement tables, inspection tables, and soldering tables, such as... Figure 3 As shown.

[0059] like Figure 7As shown, the automatic labeling tool ALT can obtain information (e.g., in list form) of electrical components for a specific PCB assembly or multiple PCB assemblies of a specific type from a database DB1 or a planning system (such as Teamcenter or NX). This information can be used by the automatic labeling tool ALT to label one or more images IM1, IM2, IM3 in the database DB1. It should be understood that the automatic labeling tool ALT is a software program that includes the database DB1, the database or planning system DB2, and a suitable interface, such as an API, depending on the specific circumstances, to the inspection and / or production system. Therefore, automatic labeling can be a computer program. In other words, automatic labeling is a computer-implemented method.

[0060] Therefore, for example, such as Figure 7 As shown, once deployed on the edge device EDGE, the machine learning model ML can receive images from the camera C at the inspection station, and can also receive, for example via an automatic labeling tool ALT, a bill of materials (BOM) associated with the PCB assembly C captured in the acquired images from a database or planning system DB1. The machine learning model ML can then identify one or more parts A1-A4 present in the BOM from the acquired images of the PCB assembly.

[0061] A bill of materials (BOM) or product structure (sometimes called a bill of materials (BOM) or related list) is a list of raw materials, sub-components, intermediate components, sub-parts, parts, and the quantity of each required to manufacture the final product. Generally, component information for PCB assemblies can be obtained using a machine learning (ML) model. For example, a list of components to be placed on a PCB can be stored within a production system, inspection station, or edge device.

[0062] Therefore, the machine learning model (ML) can infer whether the PCB components captured on the processed image are fully equipped or missing one or more electrical components, or whether the wrong electrical components are placed on the PCB.

[0063] See now Figure 8 This illustrates the workflow for inspecting PCB assemblies and how to integrate the inspection into the PCB assembly production (line).

[0064] This workflow can be implemented by one or more software program modules M1-M5. The first module M1 can run directly on the operating system and can be used to scan identifiers of PCB components. For example, the first module can be used to identify PCB components based on identifiers (e.g., 2D barcodes) set on the PCB components, wherein the identifiers are used to identify object-based analyzers from multiple object-based analyzers to identify at least one component placed on the PCB.

[0065] The identifier can then be transmitted to the second module M2. The second module M2 can then acquire images (capture frames) from the camera at the inspection station.

[0066] The identifier and image can then be transmitted to a third module M3, which, for example, retrieves a bill of materials or other component information for the PCB assembly to be assembled based on the identifier, preferably including electrical components to be placed on the PCB assembly.

[0067] Furthermore, the image and identifier can be transmitted to the fourth module M4. The fourth module can, for example, select a suitable machine learning model from multiple machine learning models based on the identifier, where each of the multiple machine learning models is adapted to a specific PCB assembly, i.e., a PCB assembly type, and is therefore trained to identify parts for that specific PCB assembly type. After selecting a suitable machine learning model, inference can be performed by the machine learning model. Inference may include object detection based on the received image. After object detection is completed and the corresponding electrical parts on the image are associated, the identified parts can be transmitted again to the third module M3, where the identified electrical parts are compared with the previously received bill of materials.

[0068] For visualization purposes, frames can be added to the objects detected on the image as previously processed using the fifth module M5. Furthermore, missing components can be visualized by adding frames to the portions of the PCB assembly image where missing components should be placed or where incorrect components are placed. The comparison result between the components identified by the object detection analysis and the component information from the third module M3 can be transmitted to the second module M2, and the result is forwarded from the second module to the first module M1. The comparison result can be, for example, pass or fail, i.e., in binary form.

[0069] It should be understood that modules M1-M5 can be combined with each other to form a single module, or functions can be split differently between modules, or the functions of a module can be combined into other modules.

[0070] Finally, the results can be displayed, for example, in a browser. Figure 8 As shown, the visualization of module M5 can be displayed to the host operating system.

[0071] The results of this comparison can then be used to control further production steps of the PCB assembly. In other words, as described above, settings or other information can be written to the label on the pallet transporting the PCB assembly based on the comparison results. These settings can be used to control further production steps of the PCB assembly. For example, the soldering of the PCB assembly can be controlled.

[0072] Further exemplary embodiments are described below:

[0073] According to a first embodiment, a method for inspecting a printed circuit board (PCB) assembly (C) is provided, the method comprising the steps of: acquiring an image (IM) of the PCB assembly (C) (e.g., using a camera) and analyzing the image (IM), wherein the analysis includes object-based analysis of the image (IM) to identify at least one component (A) placed on the PCB (B); and determining whether at least one component (A) is placed on the PCB (B) based on a comparison between the findings of the object-based analysis and stored component information of the PCB (B).

[0074] In a second embodiment, the method according to the first embodiment includes the following steps: writing one or more settings for soldering a PCB assembly (C) by a soldering device based on the result of the comparison, wherein preferably the settings include PCB type and / or PCB ID.

[0075] In a third embodiment, the method according to any of the foregoing embodiments includes the following steps: loading one or more settings for soldering the PCB assembly (C) by a soldering device based on the result of the comparison.

[0076] In a fourth embodiment, the method according to any of the foregoing embodiments includes the step of: preventing the writing of one or more settings for soldering a PCB assembly (C) by a soldering device based on the result of the comparison.

[0077] In the fifth embodiment, the method according to any of the foregoing embodiments includes the step of stopping the production of the PCB assembly (C) based on the result of the comparison.

[0078] In a sixth embodiment, the method according to any of the preceding embodiments includes the following steps: identifying at least one missing component on the PCB assembly (C) based on the comparison, and preferably repairing the PCB assembly (C) according to the identified missing component, and writing one or more settings for soldering the PCB assembly (C) by a soldering device based on the result of the comparison.

[0079] In the seventh embodiment, the method according to any of the preceding embodiments includes the steps of: identifying false errors based on the result of the comparison, and writing one or more settings for soldering the PCB assembly (C) by a soldering device based on the result of the comparison.

[0080] In the eighth embodiment, the method according to any of the preceding embodiments includes the step of: placing a PCB assembly (C) on a tray (Y), wherein the tray (Y) includes a rewritable memory (G) for storing one or more settings, such as RFID tags.

[0081] In the ninth embodiment, the method according to any of the preceding embodiments includes the following steps: identifying the PCB assembly based on an identifier (e.g., a 2D barcode) set on the PCB assembly, wherein the identifier is used to identify an object-based analysis program from a plurality of object-based analysis programs for identifying at least one component placed on the PCB (B).

[0082] In the tenth embodiment, according to any of the methods in the preceding embodiments, the object-based analysis procedure includes a trained machine learning model (ML).

[0083] In the eleventh embodiment, the method according to any of the foregoing embodiments includes the steps of: producing different types of PCB components (C) and loading an object-based analysis program based on the type of PCB component (C) identified by an identifier.

[0084] In the twelfth embodiment, the method according to any of the foregoing embodiments includes the following steps: receiving stored component information (e.g., in the form of a bill of materials) of PCB (B) from an engineering or planning system (e.g., TEAMCENTER) for the production of PCB assembly (C).

[0085] In a thirteenth embodiment, a method for training a machine learning model (ML) for an object-based analysis procedure includes the following steps: acquiring multiple images of a PCB assembly (C), preferably of different types of PCB assemblies (C), most preferably during the production process of the PCB assembly; selecting images suitable for training the machine learning model from the multiple images (IM1, IM2, IM3); automatically labeling the multiple images (IM1, IM2, IM3) based on a template for labeling the PCB assembly (C); and training the machine learning model (ML) based on the labeled images (IM1, IM2, IM3).

[0086] In the fourteenth embodiment, an inspection system (2) for inspecting printed circuit board (PCB) assemblies includes: a camera (I) for acquiring images (IM) of the PCB assembly (C) and a control unit for analyzing the images, wherein the analysis includes object-based analysis of the images (IM) to identify at least one component (A1-A4) placed on the PCB (B); the control unit is further configured to determine whether at least one component (A1-A4) is placed on the PCB (B) based on a comparison between the findings of the object-based analysis and stored component information of the PCB (B).

[0087] In the fifteenth embodiment, a production system (1, 2, 3) for producing printed circuit board assemblies (C) includes an inspection system (2) according to the foregoing embodiments and a soldering device (3) connected to the inspection system.

Claims

1. A method for inspecting a printed circuit board (PCB) assembly (C), comprising the following steps: Acquire an image (IM) of the PCB assembly (C) and analyze the image (IM), wherein the analysis includes object-based analysis of the image (IM) to identify at least one component (A) placed on the PCB (B). The object-based analysis is performed on top of an object-based analysis procedure, which includes a trained machine learning (ML) model. And the object-based analysis program determines whether the at least one component (A) is placed on the PCB (B) based on a comparison between the object-based analysis findings and the stored component information of the PCB (B); and If object detection analysis determines that one or more components are missing or misplaced, or that one or more incorrect components are placed on the PCB, then the object-based analysis program outputs an error. Displays an image of the PCB assembly and error information related to missing, incorrect, or misplaced components; The results of the visual inspection of the PCB assembly are input by the staff, and the results of the visual inspection indicate false errors in the object detection analysis; Write one or more settings for soldering the PCB assembly (C) by soldering equipment. The object detection analysis is performed by the trained machine learning model, and the training of the machine learning model is based on the labeled images of the PCB assembly. The trained machine learning model identifies one or more components in the acquired images of the PCB assembly that appear in the bill of materials. If the object detection analysis identifies all electrical components to be placed on the PCB or if a false error occurs, the production of the PCB assembly continues by writing one or more settings for soldering the PCB assembly by soldering equipment on a label attached to the tray where the PCB assembly is located.

2. The method according to claim 1, comprising the following steps: Based on the results of the comparison, one or more settings are written for soldering the PCB assembly (C) by soldering equipment, wherein the settings include PCB type and / or PCB ID.

3. The method according to any one of the preceding claims, comprising the following steps: Based on the results of the comparison, one or more settings are loaded for soldering the PCB assembly (C) by soldering equipment.

4. The method according to any one of claims 1-2, comprising the following steps: Based on the results of the comparison, writes to one or more settings for soldering the PCB assembly (C) by soldering equipment are blocked.

5. The method according to any one of claims 1-2, comprising the following steps: Based on the results of the comparison, production of the PCB assembly (C) is stopped.

6. The method according to any one of claims 1-2, comprising the following steps: Based on the comparison, at least one missing component on the PCB assembly (C) is identified, and the PCB assembly (C) is repaired according to the identified missing component. Based on the results of the comparison, one or more settings are written for soldering the PCB assembly (C) by soldering equipment.

7. The method according to any one of claims 1-2, comprising the following steps: Based on the results of the comparison, false errors are identified, and Based on the results of the comparison, one or more settings are written for soldering the PCB assembly (C) by soldering equipment.

8. The method according to any one of claims 1-2, comprising the following steps: The PCB assembly (C) is arranged on a tray (Y), wherein the tray (Y) includes a rewritable memory (G) for storing one or more settings.

9. The method according to any one of claims 1-2, comprising the following steps: The PCB assembly is identified based on an identifier disposed on the PCB assembly, wherein the identifier is used to identify an object-based analysis program from a plurality of object-based analysis programs for identifying at least one component placed on the PCB (B).

10. The method of claim 9, comprising the following steps: Produce different types of PCB components (C) and load an object-based analysis program based on the PCB component (C) type identified by the identifier.

11. The method according to any one of claims 1-2, comprising the following steps: Receive the stored component information of the PCB (B) from the engineering design or planning system used for the production of the PCB assembly (C).

12. A computer-implemented method for training a machine learning (ML) model for an object-based analytics program, comprising the following steps: Acquire multiple images of the PCB assembly (C); Select an image suitable for training the machine learning model from the plurality of images (IM1, IM2, IM3); The plurality of images (IM1, IM2, IM3) are automatically labeled based on a template used to label the PCB assembly (C). The machine learning model (ML) is trained based on labeled images (IM1, IM2, IM3) by adjusting one or more predetermined coordinates of the template based on one or more reference points for each image, wherein the coordinates are related to one or more components of the PCB assembly. The labeled images are used to train the machine learning model to perform object detection of images of the PCB assembly. The machine learning model is configured to determine one or more components in the acquired images of the PCB assembly that appear in the bill of materials. In the event that the object detection analysis identifies all electrical components to be placed on the PCB or in the event of a false error, production of the PCB assembly continues by writing one or more settings for soldering the PCB assembly by soldering equipment on a label attached to the tray where the PCB assembly is located.

13. An inspection system (2) for inspecting printed circuit board (PCB) assemblies, comprising: A camera (I) for acquiring images (IM) of the PCB assembly (C), and A control unit for analyzing the image, wherein the analysis includes object-based analysis of the image (IM) to identify at least one component (A1-A4) placed on the PCB (B). The control unit is further configured to determine whether the at least one component (A1-A4) is placed on the PCB (B) based on a comparison between the findings of the object-based analysis and the stored component information of the PCB (B), wherein the object-based analysis is performed on the basis of an object-based analysis program, wherein the object-based analysis program includes a trained machine learning model (ML). The control unit is also configured to: if object detection analysis determines that one or more components are missing or misplaced, or that one or more incorrect components are placed on the PCB, then the object-based analysis program outputs an error; The control unit is also used to display an image of the PCB assembly and error information related to missing or incorrect components or misplaced components; The control unit is also used to receive the results of a visual inspection of the PCB assembly by an operator, the results of which indicate false errors in the object detection analysis. The control unit is also configured to write one or more settings for soldering the PCB assembly (C) by soldering equipment. The object detection analysis is performed by the trained machine learning model, and the training of the machine learning model is based on the labeled images of the PCB assembly. The trained machine learning model identifies one or more components in the acquired images of the PCB assembly that appear in the bill of materials. If the object detection analysis identifies all electrical components to be placed on the PCB or if a false error occurs, the production of the PCB assembly continues by writing one or more settings for soldering the PCB assembly by soldering equipment on a label attached to the tray where the PCB assembly is located.

14. A production system (1, 2, 3) for producing printed circuit board assemblies (C), comprising an inspection system (2) according to claim 13 and a soldering device (3) connected to the inspection system.

Images