Machine vision plays a crucial role in smart factories, effectively increasing production capacity and improving product yield. When choosing a small machine vision system, traditional industrial smart cameras offer advantages such as small size, high integration, and ease of development and use; embedded machine vision systems, on the other hand, boast greater configuration flexibility, can be equipped with high-end CPU processors, support multi-channel cameras, and offer high scalability. Are there newer types of small machine vision systems that can combine the advantages of both, while also reducing costs and meeting the increasingly demanding and rapid requirements of vision applications?
The demand for machine vision in smart factories
Performance and processing power. System performance and throughput play a crucial role in improving productivity. For typical machine vision systems, high resolution and high frame rate (frames per second) are like fish and bear's paw—mutually exclusive. In practical applications, it's usually a combination of high resolution but low frame rate or low resolution but high frame rate. The only way to achieve both is to use a high-end CPU processor to compensate for the result of the multiplication of resolution and frame rate. How to achieve optimal processing performance at a reasonable cost is a key concern for system developers.
Production line environment. Factory environments are typically harsh. For example, in beverage packaging production lines, the system may be in direct contact with liquids. Machining environments, on the other hand, are harsh conditions filled with workpieces being cut. If a machine vision system must be deployed near such a demanding production line environment, then choosing a product with waterproof and dustproof capabilities is essential.
Multiple production workstations. In a factory environment, the process of a finished product reaching the market—from component manufacturing, semi-finished product handling, quality inspection to packaging—must pass through multiple different workstations. For example, a CNC machine is responsible for machining components; an industrial robot picks up the parts, and an industrial camera positions the workpiece before cutting begins; after cutting, the parts enter the inspection station for defect detection; and finished products that pass inspection have their shipping barcodes scanned in the packaging area. Ensuring easy integration and communication between these multiple production workstations is a key factor in achieving factory intelligence.
Software development environment. The ease of developing and integrating software solutions is a major concern for all engineers implementing intelligent systems, and it is often the most crucial factor determining the success or failure of a project. Shortening development time and reducing system development costs are critical priorities.
The Decisive Factor in Choosing a Small Machine Vision System
Processor computing performance. Traditional smart cameras, due to their small size and limited space, have restricted heat dissipation capabilities, thus limiting them to single-core Atom or ARM architecture processors. While these offer low power consumption, their limited performance restricts them to single-task image analysis processing, such as counting or barcode scanning. With the release of the Intel Atom™ 3840 processor series, processing performance is doubled compared to the previous generation, while maintaining low power consumption. This means that high performance can be achieved in a small form factor, enabling multi-task image processing. The new generation of small machine vision systems can simultaneously perform multiple tasks such as size measurement, counting, positioning, and QR code reading, offering the cost-effectiveness of one system replacing multiple systems.
Image sensor performance and image quality are crucial. The image sensor is the heart of a machine vision system, and its size directly reflects image quality. In the past, smart cameras were used for basic image inspection, and the relationship between sensor size and image quality wasn't readily apparent. However, if machine vision is to be applied to high-end, high-speed inspection applications, then sensor size becomes a critical factor to consider when selecting a system.
A comparison of rolling shutters and global shutters. The difference between rolling shutters and global shutters lies in the time difference of image exposure. A rolling shutter uses electronic signals to instruct the photosensitive element to expose images sequentially until the entire image is exposed. A global shutter, on the other hand, exposes the entire image "simultaneously" during the exposure. With improvements in system processing power, system performance will no longer be a bottleneck. For applications requiring the detection of high-speed moving objects, a global shutter sensor can capture accurate images without image ghosting.
Coprocessor. Image quality plays a crucial role in machine vision image acquisition and analysis. Optimizing the quality of acquired images before they enter the analysis process ensures accurate image analysis. In past applications, image data acquired by the system had to undergo computation and image quality optimization by the system processor. However, the amount of image data that could be processed was limited by CPU computing resources. But with the support of FPGAs, image matrix calculations can be filtered and optimized before reaching the CPU, significantly accelerating image processing performance and reducing CPU resource consumption. This allows system resources to be reserved for the core of the machine vision system—image algorithms—and enables more real-time processing of large amounts of image data, enabling high-speed and complex image processing and analysis. Preprocessing functions include lookup tables, regions of interest (ROI), and shading correction, among other image quality optimization functions.
GPU graphics and multimedia image processing performance. The new generation Intel Atom™ EM3840 processor offers approximately six times the computing performance of its predecessor, the Intel Atom™ DM2550 processor series. It can simultaneously handle multi-channel image compression and transmission via Intel HD Graphics 4000 technology. Through the improved CPU and GPU performance, image detection results can be recorded, stored, or used as raw data for further comparison and analysis, enabling more intelligent functions for factory information systems.
System display performance. In a factory environment, traditional smart cameras can only transmit data via Ethernet for monitoring at the central control station. If the machine vision system supports a VGA output interface, it can simultaneously output images via both VGA and Ethernet ports, connecting to HMIs or production line screens for real-time inspection and problem detection, effectively improving production line performance.
64-bit architecture. Image analysis software, due to the large volume of data it needs to process, is largely supported by mainstream applications on the market. Therefore, the choice of a machine vision system must also be based on 64-bit support to maximize the benefits of the application.
System storage capacity. The size of the storage capacity of a small machine vision system indicates that the user can store more image recognition and comparison samples, as well as detection data, or perform backups. This is highly beneficial to the overall stability of the system.
Total Cost of Ownership (TCO) Considerations. The total cost of ownership for a system purchase doesn't just consider the cost of the machine vision system itself. Users are savvy; the key is to help customers reduce costs from a TCO perspective.
ADLINK Technology's new generation of x86 smart cameras breaks the framework of traditional smart cameras and embedded machine vision systems. It boasts cross-industry advantages in performance, scalability, stability, development costs, and total cost of ownership, providing a new option for today's high-speed, high-end machine vision and image analysis system applications. It also offers users and integrators with flexible development capabilities a more cost-effective solution and a faster time-to-market.
