Hybrid deep learning systems and methods allow for detecting anomalies in objects, such as electrical printed circuit board (PCB) components, based on image data. In one or more embodiments, a hybrid deep learning model comprises a Graph Attention Network (GAT) that uses spatial properties of the PCB components to extract latent semantic information and generate an output set of hidden representations. The GAT treats each of the electrical components as a node and each connection between them as edges in a graph. The hybrid system further comprises a Convolutional Neural Network (CNN) that uses pixel data to obtain its own output set of hidden representations. The hybrid deep learning model concatenates both sets to detect anomalies that may be present on the PCB.

An image-recognition method is provided. The method includes the following steps: receiving structured data, wherein the structured data includes training-set data and testing-set data, and the structured data includes a plurality of groups, and each group includes one or more types, and each type includes a plurality of check-point images; training an artificial-intelligence (AI) model using the training-set data; inputting the testing-set data into the AI model to obtain a model evaluation of the AI model; and determining one or more first types with a lower overall recognition rate or a lower confidence level in the structured data, and deletes or corrects the check-point images in the one or more first types to update the structured data.

There is a need for more effective and efficient printed circuit board (PCB) design. This need can be addressed by, for example, solutions for performing automated PCB component estimation. In one example, a method includes identifying a plurality of initial component estimations for the PCB; performing a shadow detection segmentation using the plurality of initial component estimations, a non-direct-lighting image, and one or more direct-lighting images to generate a first set of detected PCB components; performing a super-pixel segmentation using the plurality of initial component estimations and the non-direct-lighting-image to generate a second set of detected PCB components; and generating a bill of materials for the PCB based at least in part on the first set of detected PCB components and the second set of detected PCB components.

An apparatus for determining the condition of a printed circuit board, including a stencil database for storing a plurality of stencil data files corresponding to stencils of different dimensions, an image capturing device having an area where the printed circuit board is laid thereon for a surface profile of the printed circuit board to be captured for generating a surface data file of the printed circuit board, and a condition determination module that compares data from the surface data file with data from one or more stencil data file as a nominal reference via a matching process to determine the condition of the printed circuit board based on a set of categories. The printed circuit board is sent for further processing (i.e., stencil printing) after being determined to be in a match-successful condition where data from the surface data file falls within a tolerance of the nominal reference.

A system classifies materials utilizing a vision system that implements an artificial intelligence system in order to identify or classify and then remove automotive airbag modules from a scrap stream, which may have been produced from a shredding of end-of-life vehicles. The sorting process may be designed so that live airbag modules are not activated, which may cause damage to equipment or persons.

A computer-implemented method of predicting a quality of a printed circuit board (PCB) assembly includes obtaining production data relating to production of the PCB assembly. The production data is mapped onto a latent vector of a latent space of a trained adaptive algorithm. The trained adaptive algorithm is trained on real X-ray images of PCB assemblies and/or serves for generating X-ray images of PCB assemblies. A subspace of the latent space related to the latent vector is determined. The subspace indicates a quality of the PCB assembly. Alternatively or additionally, an X-ray image of the PCB assembly is generated by the trained adaptive algorithm based on the latent vector in order to determine a quality of the PCB assembly.

The substrate processing management system for a substrate processing line including a defect tallying section to calculate a defect rate representing a rate of occurrence of substrates or electronic components determined to be defective with the substrate inspection machine, or obtain a number of defective products having substrates or electronic components determined to be defective; an attribute change recognition section to recognize when an attribute, among the attributes of the electronic component, having a possibility of affecting the inspection, has changed in the component mounting machine; a fault point estimation section to estimate, when the defect rate or the number of defective products exceeds a predetermined value, which of the component mounting machine or the substrate inspection machine is a fault point depending on whether an attribute has changed; and a countermeasure calling section configured to call for a defect countermeasure to the estimated fault point.

The present disclosure provides an inspection device for use in a mounting system including a mounting device for disposing a component on a board, including a control section configured to extract a mass area included in a captured image resulting from imaging a processing target object where a viscous fluid is formed at a predetermined part, obtain a center of gravity of the mass area so extracted, and determine whether the center of gravity is included in a normal range of the predetermined part as a reference of the captured image to thereby determine whether a bridge has occurred where the viscous fluid is formed over adjacent predetermined parts.

Machine vision systems and methods for automatically generating machine vision job(s) based on region of interests (ROls) of digital images are disclosed herein. The systems and methods comprise configuring a machine vision tool for capturing an image ID depicted in training images and labeling each training image to indicate a success or failure status of an object depicted by the training images. Candidate image feature(s) are extracted from the training images for generation of candidate ROI(s). A training set of ROls are selected from the candidate ROI(s) and are designated as an included or excluded ROls. The training set of ROls and the training images are used to train a vision learning model configured to output a machine vision job deployable to an imaging device that implements the machine vision job to detect the success or failure statuses of additional image(s) depicting the object.

In one embodiment, an X-ray inspection system may access a first set of X-ray images of one or more first samples that are labeled as being non-conforming. The system may adjust a classification algorithm based on the first set of X-ray images. The classification algorithm may classify samples into conforming or non-conforming categories based on an analysis of corresponding X-ray images. The system may analyze a second set of X-ray images of a number of second samples using the adjusted classification algorithm. The second samples may be previously inspected samples that have been classified as conforming by the classification algorithm during a previous analysis before the classification algorithm is adjusted. The system may identify one or more of the second samples from the second set of X-ray images. Each identified second sample may be classified as non-conforming by the adjusted classification algorithm.