What kind of “eyes” does smart manufacturing really need?

With the transition to Industry 4.0, the factories and production lines we were once familiar with have become test sites for new technologies. Any solution or technology that drives the informatization and intelligentization of manufacturing for improving production efficiency has the potential to effect great change. This includes machine vision.
Machine vision, when put to industrial use, is also known as “industrial vision,” which is generally understood as the use of machine vision technology in place of human eyes in the manufacturing process to complete specific jobs. It’s similar to equipping machinery with the ability to see to support its automated operations. This is why some consider industrial vision as the “eyes” of smart manufacturing.
Like other machine-substituted manual work, there are many reasons why industrial vision has become immensely popular. Firstly, machine vision is more precise than human vision, able to complete many jobs that cannot be achieved by the naked eye. Secondly, as machines do not experience fatigue and can work under extreme environments, they are more reliable. What’s more, machines can see beyond the visible spectrum (such as infrared light), giving them a wider scope of “vision.” Needless to say, machines have an obvious advantage in terms of efficiency.
Other than the technical reasons listed above, the economics of using machine vision is an important factor. This is because as populations around the world age and labor supplies shrink, labor costs for industrial manufacturing have been rising. No longer able to depend on their demographic dividends, many economies now consider it necessary to replace human labor with machines.
One example is the electronics sector where industrial vision enjoys the highest application rate. It is estimated that in China, 300 thousand people are required to check the surfaces of smartphone cover glasses, touch screens, and displays, incurring enormous labor costs. Replacing this type of repetitive human labor with industrial vision solutions can be both efficient and effective.
Market requirements have propelled the industrial vision market to rapid growth in recent years. According to estimations by Shenzhen Gaogong Industry Research Institute (GGII), the machine vision market, driven by China’s industrial sector, is expected to exceed 12 billion by 2020, with an annual growth rate of over 15% between 2017 and 2020. The electronics, automotive, medicine, food, and many other sectors will all be launch pads for machine vision applications.
Basic industrial vision functions employed during the manufacturing process consist mainly of the following elements:
- Object recognition: To process, analyze, and perceive objects or images through machine vision, identifying various targets and components.
- Inspection applications: To inspect visual targets such as colors and images. Mostly utilized for defect detection, as well as verification of object position and direction.
- Visual positioning: Machine vision systems can rapidly and accurately find parts awaiting inspection and determine their position. For instance, electronic installation equipment utilizes machine vision to obtain chip position, adjusting retrievers to pick up chips accurately.
- Object measurement: Machine vision-based non-contact measuring technology assesses targets with high precision and speed through non-contact and abrasion-free measuring.
The above functions can be mixed and matched into even more complex applications.
Compared with machine vision in other fields of application, industrial vision has several traits that are unique to the sector. Since it is for industrial use, it must perform faultlessly and exhibit high levels of reliability. As most industrial applications are highly vertical, specific application needs will also vary greatly. Industrial vision solutions, therefore, require outstanding scalability and flexibility.
So what kind of industrial vision solution is considered a “viable” solution? The following example is an analysis with Avnet’s object recognition embedded solution.
Figure 1. Avnet’s object recognition embedded solution (Image source: Avnet)
Regardless of type, machine vision systems have two core function modules: one is the visual information collection system, and the other is the visual processing system. The visual collection function for Avnet’s object recognition embedded solution is achieved through the PYTHON-1300 color image sensor from ON Semiconductor. PYTHON-1300 is a 1/2-inch SCGA CMOS image sensor with 1280x1024 HD resolution. To satisfy the needs of general industrial image sensing applications, the PYTHON-1300 also features global shutter, high speed and high sensitivity, as well as flexibility in configurations and resolution.
Figure 2. Color image sensor modules based on PYTHON-1300 (Image source: Avnet)
Within the solution, the visual processing system is connected with the image sensor module. This function is achieved by the MicroZed embedded vision development kit from Avnet, which consists of the MicroZed SOM and its accompanying video capturing card. MicroZed SOM provides the core visual processing and calculating properties, while the video capturing card integrates other peripheral circuit and extension functions required for video processing, such as HDMI input/output, camera image module connectors, and Power over Ethernet (PoE) interfaces.
Figure 3. Avnet’s MicroZed embedded vision development kit (Image source: Avnet)
The highlight of this solution is in the core board, MicroZed SOM. MicroZed SOM is a system-level module based on Xilinx’s Zynq-7000 all-programmable device. This means that, unlike other machine vision solutions, the object recognition embedded solution by Avnet utilizes a hybrid FPGA+Arm processing platform.
As a hybrid processing platform, Zynq incorporates both a system-on-chip (SoC) integrated processing system (PS) and a programmable logic (PL) unit. Embedded vision developers may divide calculation tasks between the two systems as needed to optimize the solution. They may also take advantage of the platform’s flexibility to deploy additional IOs and allot complex visual processing algorithms to the PL for acceleration. Zynq makes maximum performance possible at all times.
In addition, Zynq’s “hardware programmability” feature enables industrial vision solutions to offer greater flexibility. Whether for testing new algorithms or for expanding to new industrial applications, this solution delivers effortlessly for applications like robotic positioning, image searching, and video monitoring.
Figure 4. MicroZed SOM based on the Zynq hybrid processing platform (Image source: Avnet)
With current smart manufacturing developments, industrial vision is destined to become a prevalent application. Technological advancements are also moving this momentum forward in our pursuit of even keener industrial "eyes".

