How Can AI Improve Automatic Optical Inspection?

Inspection Methods in Industry - Automated Optical Inspection

In manufacturing, inspection is an essential function. Visual inspection guarantees that the product functions and looks as expected, and provides important benefits to manufacturers and customers. Quality assurance is provided by automated optical inspections, which can be communicated directly to customers through product labeling, or recorded within a manufacturing facility as part of their quality control process.

In addition, identifying any non-conforming items in the production process helps to determine whether the production process or steps need to be adjusted. The inspection results can help determine the cause of the failure, and the immediate identification of the defect can immediately stop the production and solve the problem. The sooner quality problems are identified, the lower the cost of solving them.

What is Automatic Optical Inspection (AOI) technology?

In the process of product manufacturing, due to various reasons, parts and components will inevitably have a variety of defects. These defects not only affect the performance of the product but even endanger life safety of the user in severe cases, causing huge losses to users.

With the rapid development of electronic technology, image sensing technology, computer technology, the use of automatic optical (vision) detection technology for surface defects based on optical image sensing, has gradually become the main method of surface defect detection, replacing manual visual detection of surface defects. Advantages of this method include automation, non-contact, high speed, high precision, and high stability. Automated optical inspection (AOI) technology is also known as machine vision inspection (MVI) technology or automated visual inspection (AVI) technology.

MVI is an emerging technology that integrates image sensing technology, data processing technology, and motion control technology to perform tasks such as measurement, detection, identification, and guidance in the process of industrial production. MVI uses optical imaging to simulate the visual imaging of the human eye. It uses a computer processing system instead of the human brain to perform data processing, and finally feeds back the results to the actuator to complete various prescribed tasks imitating the movements made by human hands.

From manual inspection to automated optical inspection (AOI):

Inspection is usually required for each product produced, and operators can be trained manually to inspect the processing or overall appearance of simple products. But as products become more sophisticated, some applications, such as printed circuit board components (PCBAs), may require scaled-up equipment whose minimum functional size is a challenge to inspectors' visual acuity. With the increase in product complexity, various types of equipment contain a large number of components. When inspecting and recording results, inspectors must overcome the dual challenges of vision and time requirements, which may lead to inaccurate manual inspections.

With increasing challenges in feature size, complexity, and throughput, automated optical inspection (AOI) is a practical way to ensure adequate inspection of every item. AOI includes image sensing, lighting, and computing subsystems that work together to capture and analyze images. The AOI system can compare the captured image to a reference image and then identify defects such as material surface defects, solder defects, or missing or misplaced components on the PCBA. Alternatively, some rule-based system measures feature dimensions to determine good or bad status. If a defect is detected, the machine equipment can isolate the defective item before continuing with subsequent inspections, or pause and warn the operator.

AOI can detect shortcomings of assembled circuit boards such as missing or skewed features, tombstone defects, wrong components, wrong polarity, defective soldering, solder bridges, and insufficient solder.

From traditional image processing to the application of AI technology:

The basic principle of image recognition is to digitize each captured image and apply various filters to detect patterns and features of objects. Edge detection filters are often used to detect objects in images, and algorithms that can identify people can apply slope detection to identify features such as arms, shoulders, legs, etc. It is also necessary to detect the orientation of these detected features relative to each other as a further defining criterion. The detection solder joint algorithm can use edge detection and color detection to identify solder joints and detect whether the fillet slope is within an acceptable range. The optical system can illuminate the unit under test from different angles using different colors.

Traditional image recognition faces many challenges, whether it is for people identification in applications such as security surveillance or vehicle-pedestrian detection, face recognition in social media, or defect detection in industrial inspection.

Defining rules and creating algorithms to detect and classify objects in digitized images is complex. In industrial inspection, developing reliable algorithms is expensive and time-consuming. When inspecting PCB components, solder joint quality is only one criterion to be tested. The presence of each component must also be verified, as well as its position and orientation relative to the solder mask, component coplanarity, and the presence of unwanted objects. Fine-tuning algorithms and adding more algorithms to cover more conditions is a never-ending task that requires constant software updates. Whenever a new product is used in the industry, new algorithms must be developed to detect it.

Artificial intelligence (AI) can imitate humans to a certain extent, applying the lessons learned to image recognition, and then being able to respond to the challenges posed by infinite changes. Among the various computing architectures covered under the general concept of AI, Convolutional Neural Networks (CNN) are commonly used for image recognition. These include artificial neurons connected and arranged in layers. They are usually deep neural networks that contain multiple inner or hidden layers between the input and output layers. The hidden layer performs specific, well-defined sampling pooling and convolution operations on the data received from the previous layer. The result is sent to the next layer, and finally to the output layer, which can indicate whether the sought object has been recognized. Before deploying a CNN, it needs to be trained to recognize specific objects. In this process, the importance or weight of each neuron is adjusted by whether each answer is correct or not. After many repeated operations, CNN can identify images with high accuracy.

The combination of AI technology and AOI equipment:

AI can bring advantages to AOI equipment suppliers and users. From a supplier perspective, if AI can determine the probability of finding a particular object, it can simplify algorithm development. Helps reduce time-to-market for new devices and reduce ongoing software support costs by reducing the need to define each item and corresponding acceptance criteria. For users, implementing enhanced AOI through AI can simplify inspection system settings, programs, and fine-tune quality judgment values.

The combination of AI technology and AOI equipment has higher accuracy and fewer false positives than traditional systems and can be quickly trained to detect new products or identify previously unknown defects. AI can automatically adjust multiple parameters faster than human experts and make decisions with a significantly reduced risk of error, enabling consistent detection results regardless of whether the AOI system is programmed by a beginner or an expert.

The AOI system is composed of a simple optical imaging and processing system integrated with general-purpose devices such as cameras, lenses, light sources, and computers. Such system characteristics underscore the advantages AI can bring to inspection applications in many areas, including security and retail. In applications where images need to be searched to detect objects and features or identify individuals, AI can simplify setup and programming, eliminate human error, minimize latency and enable better decision-making. To help developers get the most out of this technology, camera modules are now on the market with software support to simplify AI development.

AOI's future outlook:

AOI - which can operate at line matching rates, is already supporting manufacturers in various industries to improve quality assurance and productivity and to continuously improve production processes. Further improvement based on AI is the future development direction of AOI. Training algorithms for optical inspection applications can bring additional benefits such as higher decision-making capabilities, reducing operator involvement, simplifying procedures, and providing more powerful performance, which can improve defect detection while reducing false positives.