Dawn of the smart factory

Multiple solution options currently exist for machine vision applications. However, the convergence of high performance and low power consumption on new processors is providing a new alternative for machine vision applications.

What makes a factory smart?

To understand the elements critical to a machine vision system, one must first understand the many operating requirements of the smart factory.

High efficiency and throughput: In terms of conventional machine vision systems, high resolution and high frame rate are hard to achieve at the same time.

In reality, vision applications are often a marriage of high resolution with lower frame rate, or lower resolution with higher frame rate. To achieve both, a more advanced CPU is needed with costs raised commensurately. Striking the delicate balance necessary among these factors and achieving optimal efficiency with reasonable cost structure is an important issue faced by system developers on an ongoing basis.

Ruggedness and reliability: Operating environments of industrial production are often challenging for automatic systems. For example, a food and beverage production facility is likely to present damp conditions with extreme temperatures, while machine tooling environments are often dusty with metal or other intrusive particulates present. If the vision system is to be installed adjacent to production equipment, a higher degree of imperviousness to such elements is needed.

Integration with 3rd party equipment: A production line usually involves a series of operations from manufacturing, machining, pick and place, and inspection to packaging. Integration of and communication among the different systems involved is a challenge for all smart factories.

Faster development: Development of software solutions and related compatibility issues are critical factors, dictating success or failure of the implementation. Shortening development time and reducing system development costs are distinct challenges.

Types of vision systems

Conventionally, smart cameras have made use of a low power ARM-based or single-core microprocessor with limited memory, in order to meet requirements for size and ruggedness. This minimal memory limits the camera’s ability to process high-resolution images with sufficient speed for most industrial processes, and also impedes multi-tasking. Accordingly, conventional smart cameras are often single-purposed, dedicated to simpler image tasking—such as gaging, counting, alignment, or barcode scanning. Due to their minimal expandability, realization of additional functions requires installation of more system units.

When more complex, high performance machine vision applications are to be undertaken and expandability is demanded, users often turn to the other category of vision system, embedded vision systems, comprised of an industrial PC connected to high-resolution industrial cameras. Embedded vision systems typically feature a high-performance processor running a standard PC operating system with multiple vision channels supported and versatile I/O connectivity.

Embedded vision systems are, however, often more costly and complicated to deploy.

How to select the right solution for your machine vision application

The following are implementation factors to consider when selecting your machine vision solution:

CPU efficiency: As mentioned, conventional smart cameras usually run on a single-core Atom processor or ARM-based processor with considerations of size, power and heat dissipation. However, these conventional smart cameras have limited computing power and are often used only in simple image applications dealing with individual tasking of gaging, counting, alignment, or barcode scanning.

Image sensors: Image sensors are the eyes of the vision systems. Larger sensors can acquire more image information and deliver higher image quality. In the past, with conventional smart cameras focused on simple imaging tasks, the size of image sensors was not an issue. With, however, the implementation of high-end and high-speed applications, image sensor size becomes critical for image quality.

Rolling shutter vs. global shutter: Rolling shutters and global shutters differ in the way their pixels collect light. Rolling shutters collect light in sequential rows, with each row starting and finishing collection slightly different from each other. Global shuttering pixels start and end light collection during exactly the same period of time.

Conventional smart cameras, because of limited computing power insufficient to process large amounts of image data, have tended to adopt rolling shutter function. Even so, the inability of rolling shutters to remove residual signals when dealing with fast-moving objects (with attendant blur/skew/wobble/partial exposure effects), has excluded conventional smart cameras from use in high-speed industrial applications.

Currently, however, with the improved CPU efficiency of new generation processors, small form factor smart cameras are able to support global shutter deployment.

Co-processor: While image quality is critical for accurate automatic inspection and analysis, limits of optical conditions (light source or lens) frequently cause acquired images to exhibit inconsistent brightness, leading to misjudgment in analysis. If the vision system can automatically optimize acquired images before submission for analysis, accuracy of image analysis is significantly enhanced.

The use of an FPGA co-processor by new generation smart cameras greatly improves image processing efficiency by offloading image matrix operations from CPU to FPGA (image pre-processing), freeing CPU resources to carry out more advanced algorithmic operations. The FPGA co-processor can carry out image pre-processing tasks such as LUT (look up table), ROI (region of interest) and shading correction, with smaller vision systems accordingly realizing faster and more complex applications.

Graphics and media processing: New generation processors adopt a GPU driver, which offloads media processing tasks from the CPU, tripling graphic processing performance over previous generation processors. The GPU can process video encoding, compression and transmission across multiple channels simultaneously. This performance improvement empowers small vision systems to record, store and analyze media data, resulting in a "smarter” factory.

Display: Conventional smart cameras transmit data via only an Ethernet cable connected to the control center. If the vision system can also connect with HMI or a screen at the production line via VGA or Ethernet port and display image data simultaneously, operators can view inspection results and find problems earlier.

64-bit computing: As image analysis applications are required to manage large amounts of data, most mainstream software tools in this segment utilize 64-bit instructions. It is necessary, therefore, to deploy a vision system that supports 64-bit computing.

supports 64-bit memory addressing, benefiting large address space vision applications

System storage: System storage capacity can determine whether the vision system is able to run full PC OS and 3rd party APIs, in addition to the amount of image samples the system can store for matching and comparison.

TCO: TCO is not determined solely by the nominal price tag of the system, but rather a combination of factors, including space usage, peripheral support, system expandability and software development costs.

Conclusion

For modern mass production process, the implementation of automated inspection is crucial in guaranteeing manufacturing quality and productivity, a primary requirement in enhancement of corporate competiveness.

New smart cameras define a new category of vision system that singularly realizes high-performance, maximum integration, easy deployment, space efficiency and minimal total cost of ownership, well beyond what conventional systems can achieve.