US-12626347-B2 - Visual inspection training board for artificial intelligence deep learning

Abstract

A computing system includes an automatic optical inspection (AOI) and a database that stores images of a plurality of visual inspection (VI) training boards. Each of the visual inspection (VI) training boards includes at least one intended defect. The AOI system includes an artificial intelligence (AI) algorithm implemented in an automatic optical inspection (AOI) system. The AOI system trains the AI algorithm to learn at least one target defect based on the at least one intended defect including in the plurality of VI training boards, analyzes a production printed circuit board (PCB), and determines either a non-defective PCB in response to the AI algorithm determining the production PCB excludes the at least one target defect, or a defective PCB in response to the AI algorithm determining the production PCB includes at least one target defect.

Inventors

• Yanlong Hou
• Weifeng Zhang
• Wei Wang
• Jiayu Zheng

Assignees

• INTERNATIONAL BUSINESS MACHINES CORPORATION

Dates

Publication Date

20260512

Application Date

20230117

Claims (18)

1. 1 . A computer-implemented method comprising: producing a plurality of visual inspection (VI) training boards, each of the visual inspection (VI) training boards including at least one intended defect; training an artificial intelligence (AI) algorithm implemented in an automatic optical inspection (AOI) system to learn at least one target defect based on the at least one intended defect included in the plurality of VI training boards; analyzing a production printed circuit board (PCB) using the AOI system; determining one of a non-defective PCB in response to the AI algorithm determining the production PCB excludes the at least one target defect, or a defective PCB in response to the AI algorithm determining the production PCB includes at least one target defect, wherein the at least one intended defect and the at least one target defect includes one or more of an incorrect component placement, a missing component, solder misregistrations, a solder open defect, and a solder short defect.
2. 2 . The computer-implemented method of claim 1 , wherein producing the plurality of visual inspection (VI) training boards comprises one or both of assembling a physical VI training board including the at least one intended defect and computer-generating a virtual VI training board including the at least one intended defect.
3. 3 . The computer-implemented method of claim 2 , further comprising generating, via the AOI system, an alert indicating a detection of the defective PCB.
4. 4 . The computer-implemented method of claim 2 , wherein the at least one target defect included in the production PCB is unintentionally formed.
5. 5 . The computer-implemented method of claim 4 , wherein training the AI algorithm comprises: capturing an image of each physical VI training board among the plurality of VI training boards; inputting the images of the plurality of VI training boards into a training engine; assigning a known label to the at least one intended defect included in a given physical VI training board among the plurality of VI physical boards; making a prediction for classifying the images of the plurality of VI physical training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.
6. 6 . The computer-implemented method of claim 4 , wherein training the AI algorithm comprises: inputting images of the virtual VI training board among the plurality of VI training boards into the training engine; assigning a known label to the at least one intended defect included in a given virtual VI training board among the plurality of VI training boards; making a prediction for classifying the images of the plurality of VI virtual training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.
7. 7 . A computing system comprising: a database configured to receive images of a plurality of visual inspection (VI) training boards, each of the visual inspection (VI) training boards including at least one intended defect; an automatic optical inspection (AOI) system including an artificial intelligence (AI) algorithm implemented in an automatic optical inspection (AOI) system, the automatic optical inspection (AOI) system configured to: train the artificial intelligence (AI) algorithm to learn at least one target defect based on the at least one intended defect including in the plurality of VI training boards; analyze a production printed circuit board (PCB); and determine either a non-defective PCB in response to the AI algorithm determining the production PCB excludes the at least one target defect, or a defective PCB in response to the AI algorithm determining the production PCB includes at least one target defect, wherein the at least one intended defect and the at least one target defect includes one or more of an incorrect component placement, a missing component, solder misregistrations, a solder open defect, and a solder short defect.
8. 8 . The computing system of claim 7 , wherein the plurality of visual inspection (VI) training boards comprises assembling a physical VI training board including the at least one intended defect and/or a computer-generated virtual VI training board including the at least one intended defect.
9. 9 . The computing system of claim 8 , wherein the AOI system generates an alert indicating a detection of the defective PCB.
10. 10 . The computing system of claim 9 , wherein the at least one target defect included in the production PCB is unintentionally formed.
11. 11 . The computing system of claim 10 , wherein the AOI system trains the AI algorithm by performing the operations of: capturing an image of each physical VI training board among the plurality of VI training boards; inputting the images of the plurality of VI training boards into a training engine; assigning a known label to the at least one intended defect included in a given physical VI training board among the plurality of VI physical boards; making a prediction for classifying the images of the plurality of VI physical training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.
12. 12 . The computing system of claim 10 , wherein the AOI system trains the AI algorithm by performing the operations of: inputting images of the virtual VI training board among the plurality of VI training boards into the training engine; assigning a known label to the at least one intended defect included in a given virtual VI training board among the plurality of VI training boards; making a prediction for classifying the images of the plurality of VI virtual training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.
13. 13 . A computer program product to control computing system to use visual inspection training boards to perform artificial intelligence deep learning, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by an electronic computer processor to the computing system to perform operations comprising: producing a plurality of visual inspection (VI) training boards, each of the visual inspection (VI) training boards including at least one intended defect; training an artificial intelligence (AI) algorithm implemented in an automatic optical inspection (AOI) system to learn at least one target defect based on the at least one intended defect included in the plurality of VI training boards; analyzing a production printed circuit board (PCB) using the AOI system; determining one of a non-defective PCB in response to the AI algorithm determining the production PCB excludes the at least one target defect, or a defective PCB in response to the AI algorithm determining the production PCB includes at least one target defect, wherein the at least one intended defect and the at least one target defect includes one or more of an incorrect component placement, a missing component, solder misregistrations, a solder open defect, and a solder short defect.
14. 14 . The computer program product of claim 13 , wherein producing the plurality of visual inspection (VI) training boards comprises one or both of assembling a physical VI training board including the at least one intended defect and computer-generating a virtual VI training board including the at least one intended defect.
15. 15 . The computer program product of claim 14 , further comprising generating, via the AOI system, an alert indicating a detection of the defective PCB.
16. 16 . The computer program product of claim 14 , wherein the at least one target defect included in the production PCB is unintentionally formed.
17. 17 . The computer program product of claim 16 , wherein training the AI algorithm comprises: capturing an image of each physical VI training board among the plurality of VI training boards; inputting the images of the plurality of VI training boards into a training engine; assigning a known label to the at least one intended defect included in a given physical VI training board among the plurality of VI physical boards; making a prediction for classifying the images of the plurality of VI physical training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.
18. 18 . The computer program product of claim 16 , wherein training the AI algorithm comprises: inputting images of the virtual VI training board among the plurality of VI training boards into the training engine; assigning a known label to the at least one intended defect included in a given virtual VI training board among the plurality of VI training boards; making a prediction for classifying the images of the plurality of VI virtual training boards based on the known label; comparing the prediction to a base model image of a PCB that excludes the at least one intended defect; and updating the AI algorithm based on results of the comparison.

Description

BACKGROUND The present invention generally relates to computer systems, and more specifically, to computer systems, computer-implemented methods, and computer program products for using visual inspection training boards to perform artificial intelligence deep learning. An automatic optical inspection (AOI) device is common tool used in electronics manufacturing for various printed circuit board (PCB) assembly tasks. The AOI device typically implements an AOI judgement algorithm which facilitates inspection of various PCB assembly errors including, but not limited to, incorrect component placement, missing components, solder misregistrations, solder open defects (also referred to as solder opens), and solder short defects (also referred to a solder shorts). The conventional AOI judgement algorithm implemented by traditional AOI tools provides no intelligence. Consequently, if the judgement criteria is loosened or relaxed, more defects go undetected. On the other, if the judgement criteria is tightened or more strict, all or almost all defects can be detected. These increased detections may also include an increase in false call points or “false positive.” These false positive result in the need to employ a second inspection stage that is performed by a human inspector. When performing the manual inspection for a common PCB board, the inspector may have to review and inspect hundreds of contact points, component placements and/or solder deposits, which is extremely time consuming and susceptible to human error allowing for a missed defect. SUMMARY One or more non-limiting embodiments of the present invention are directed to a computer-implemented method for using visual inspection training boards to perform artificial intelligence deep learning. The method includes producing a plurality of visual inspection (VI) training boards, where each of the visual inspection (VI) training boards including at least one intended defect, and training an artificial intelligence (AI) algorithm implemented in an automatic optical inspection (AOI) system to learn at least one target defect using the plurality of VI training boards. The method further includes analyzing a production printed circuit board (PCB) using the AOI system, and determining one of a non-defective PCB in response to the AI algorithm determining the production PCB excludes the at least one target defect, or a defective PCB in response to the AI algorithm determining the production PCB includes at least one target defect. According to one or more non-limiting embodiments, the method further includes wherein producing the plurality of visual inspection (VI) training boards comprises one or both of assembling a physical VI training board including the at least one intended defect and computer-generating a virtual VI training board including the at least one intended defect. According to one or more non-limiting embodiments, the method further includes generating, by the AOI system, an alert indicating a detection of the defective PCB. According to one or more non-limiting embodiments, the method further includes the at least one target defect included in the production PCB is unintentionally formed. According to one or more non-limiting embodiments, the method further includes the at least one intended defect and the at least one target defect includes one or more of an incorrect component placement, a missing component, solder misregistrations, a solder open defect, and a solder short defect. According to one or more non-limiting embodiments, the method of training the AI algorithm comprises capturing an image of each physical VI training board among the plurality of VI training boards, and inputting the images of the plurality of VI training boards into a training engine. The training method further includes assigning a known label to the at least one intended defect included in a given physical VI training board among the plurality of VI physical boards, and making a prediction for classifying the images of the plurality of VI physical training boards based on the known label. The training method further includes comparing the prediction to a base model image of a PCB that excludes the at least one intended defect, and updating the AI algorithm based on results of the comparison. According to one or more non-limiting embodiments, the method of training the AI algorithm includes inputting images of the virtual VI training board among the plurality of VI training boards into the training engine, and assigning a known label to the at least one intended defect included in a given virtual VI training board among the plurality of VI training boards. The training method further includes making a prediction for classifying the images of the plurality of VI virtual training boards based on the known label, comparing the prediction to a base model image of a PCB that excludes the at least one intended defect, and updating the AI algorithm based on results of the com