• Use Cases

How Synthetic Data Can Facilitate Development of Automated Quality Assurance Systems for Manufacturing of Electronics

By: SKY ENGINE AI
scroll down ↓to find out more

Introduction

Considering how much the current economy relies on electronics goods, optimization and streamlining of component production becomes a paramount challenge for the manufacturers. One of the solutions is to increase automation of component manufacturing, product assembly and handling wherever possible. In combination with the increasing miniaturization of electronic goods and their increasing complexity, the need for automatic quality assurance (QA) and quality control (QC) processes becomes even more prominent. 

Synthetic data for training and validation of computer vision AI models for automated quality control in production of electronic goods

The ever-increasing push for process efficiency and product miniaturization drives the development of computer vision AI models to assist in quality assessment as well as handling of produced goods. Especially for the quality assurance tasks, the computer vision algorithms have to be trained on vast collections of images representing the broadest range of cases, including the rarest instances. Collecting such data in real life is often expensive and time-consuming and may even be unsuccessful, due to underrepresentation of some image classes. A resulting dataset is imbalanced and will not yield a well-trained and robust computer vision AI model.

Imagery and video datasets generated in the SKY ENGINE AI Synthetic Data Cloud for Vision AI circumvents all such limitations and adds additional perks, such as full control over the rendered scene, Physically-based simulation environment, and pixel-perfect ground truths. Furthermore, setting up an effective real-time object detection system involves additional parameters, such as the number of cameras, their types and capabilities, light source and its type, whether it is a still image analysis or video streaming system etc. All these elements have to be accounted for during the training of the ML algorithms for object detection. SKY ENGINE AI solves that problem as well with its customizable sensor simulations and optical setups.

How computer vision AI can benefit in the electronics industry

Printed Circuit Board (PCB) quality control and defect detection

Successful identification of defects on PCB plates and PCB assemblies with AI-powered automated quality control tools requires that the computer vision algorithms are trained on datasets with rich ground truths. Using synthetic data generated on the SKY ENGINE AI’s platform, which features 3D key points, bounding boxes, instance masks, and physically-based depth of field simulation, to train computer vision AI models ensures the quality and efficiency of manufacturing processes. Realistic renders of soldering issues, for example solder balling or bridging, enable elimination of faulty PCB plates and can help identify issues within the manufacturing process that cause those defects in the first place.

Product inspection and assembly verification

Object detection, identification, and object placement are enabled by the SKY ENGINE AI platform capabilities of providing synthetic data with pixel-perfect semantic maps, bounding boxes, and depth maps. With such synthetic data, computer vision AI developers can create solutions that accurately and precisely detect misassembled PCBA plates and other electronic goods.

Component identification and sorting

Streamlining the inventory management and manufacturing process with AI-powered computer vision tools may provide a competitive edge in large scale manufacturing operations. Computer vision AI models trained on synthetic data with pixel-perfect ground truths and appropriate representation of edge cases are necessary to build robust inventory management solutions based on machine vision. 

Counterfeit detection

The estimated value of the counterfeit goods market is USD 1 trillion annually, according to the International Chamber of Commerce report, with electronics accounting for a significant portion of that number. Only automated inspection and analysis of purchased components prior to assembly, can prevent electronic goods manufacturers from significant losses. Synthetic datasets can be designed specifically for the detection of counterfeit elements during automated quality assurance screening. Rich, pixel-perfect ground truths coupled with normal vector maps, depth maps and dataset balance customization assure development of effective machine vision solutions for the screening of electronic components pre-assembly. 

Testing and diagnostics of electronic devices

SKY ENGINE AI’s Synthetic Data Cloud enables the generation of highly realistic datasets for training and validating machine vision AI models, specifically tailored for automated troubleshooting and diagnostics of electronic goods. By simulating diverse scenarios—such as wear and tear, component failures, or user-induced damage—this synthetic data empowers models to identify and diagnose issues across a wide range of devices and environments. With precise annotations and control over variables like lighting, angles, and product variations, SKY ENGINE AI provides a powerful alternative to costly real-world data collection, accelerating development cycles and enhancing diagnostic accuracy for manufacturers.

SKY ENGINE AI reduces domain gap through the physically-based representation of Printed Circuit Board (PCB) defects

Using SKY ENGINE AI’s platform, data scientists can generate datasets with feature distribution characteristics required by their ML projects. For example, if a computer vision (CV) algorithm used for the detection of soldering issues on PCB plates does not recognize a certain class of problems (such as solder bridging or webbing) with the required accuracy, it is likely because of the insufficient representation of those classes within the training data. SKY ENGINE AI’s platform provides full control over the feature distribution within a dataset and allows for the iterative generation of training and testing datasets until the desired outcome of the algorithm training is reached. 

It is possible to simulate any optical setup that is required, even a defective one. 

SKY ENGINE AI platform allows the simulation of virtually any optical detection setup, as long as its parameters are known, e.g. optical path, numerical aperture, and magnification. What is more, it is possible to reproduce any aberration characteristic for the lens, such as chromatic aberration or cylindrical aberration. This feature pushes further the quality of the synthetic dataset, as it becomes even more representative of reality. 

Physically-based simulation of the depth of field (DOF)

Lenses used typically in automated quality control have a very shallow focus and accurate reproduction of this feature is crucial for uses such as PCB and PCBA computer vision-assisted automated quality control. SKY ENGINE AI’s custom code provides data scientists the means for factoring in such effects. We use full Fresnel equations for light simulation

Using full Fresnel equations is particularly important for the representation of materials that conduct electricity (e.g. copper). Other market solutions use the simplified Schlick’s approximation resulting in shorter times of image rendering, but by utilizing the two-part complex refractive index from Fresnel equations, we can realistically reproduce interactions of light with various surfaces. This approach guarantees the unrivaled technical quality of the dataset. What it means for the automated Quality Check (QC) of PCBs and PCBAs is that we can accurately replicate the interaction of light with the plate or assembly surface. Hence the data and images generated with SKY ENGINE AI are as nuanced as the real object. This translates to improved efficiency of the whole QC process and higher end-product quality. 

Summary 

SKY ENGINE AI’s Synthetic Data Cloud empowers developers of AI-driven automated quality control systems for electronics manufacturing with unparalleled precision and control. By offering physically-based, pixel-perfect datasets that accurately replicate real-world defects and environmental conditions, SKY ENGINE AI enables efficient training of machine vision models for tasks like PCB inspection, counterfeit detection, component sorting, and assembly verification. Unlike traditional data sources, SKY ENGINE AI provides consistent, balanced datasets with detailed control over sensor simulations, optical setups, and lighting conditions, equipping developers to tackle complex edge cases and refine their models iteratively. Embrace a new level of control, accelerate deployment cycles, and achieve superior diagnostic accuracy in quality control workflows with SKY ENGINE AI.

Let us know about your cases and get access to the SKY ENGINE AI Platform or get tailored AI models or synthetic datasets for your AI-powered automated quality control needs. As we support many more industries than just the electronics, a broad range of data customization is available even for specific sensors and environments.