An industrial fanless PC for visions must be sized based on the cameras, the volume of images to be processed, the algorithms used, and the machine’s constraints. The CPU, GPU, memory, interfaces, PoE, and local acquisition and processing must therefore be considered as a whole.
In a machine vision architecture, the PC does more than just receive images. It can handle image acquisition, preprocessing, the execution of vision or artificial intelligence algorithms, communication with the PLC, and sometimes the control of lighting or actuators.
The choice of hardware must therefore answer a simple question: What computing power and which interfaces are actually needed to process the images within the time constraints imposed by the industrial process?
⇒ Key takeaway: An industrial fanless PC for machine vision must be sized based on the entire system: cameras, data acquisition, CPU, GPU, memory, interfaces, processing time, and the industrial environment. Solutions such as Neousys NUVIS or Advantech architectures with GPU expansion show that it is possible to integrate acquisition, computing, and edge processing into a single platform.
The right choice, therefore, is not necessarily the most powerful PC, but rather the configuration capable of processing images within the required timeframe, with the appropriate interfaces and a design compatible with production constraints.
TABLE OF CONTENTS:
1.What role does a fanless PC play in a machine vision system?
2.CPU or GPU: What level of computing power is needed for machine vision?
3.Machine vision: Why are latency and processing time critical?
4.PoE, GigE vision, USB vision: which interfaces should you choose?
5.Image acquisition: a factor to consider from the start
6.Memory, storage, and bandwidth: criteria often overlooked
7.Why choose a fanless PC for machine vision?
8.Industrial environment: The protection rating matters just as much as performance
9.Local processing: Why is Edge computing relevant for machine vision?
10.Machine vision and predictive maintenance: two complementary applications
11.Machine vision on an existing machine: Consider a retrofit
12.What criteria should you consider when choosing a fanless vision PC?
FAQ
A machine vision architecture typically combines cameras, optics, lighting, sensors, vision software, a PLC, and a computing unit. The fanless industrial PC serves as the computational core of this architecture. Depending on the application, it can perform the following functions:
This architecture brings computing closer to the field and enables decisions to be made directly at the machine level.
To learn more about Edge AI architectures and local processing, see our article “Industrial fanless PCs and Edge computing: what role does AI play?”
The choice of processor depends directly on the complexity of the processing tasks.
A sufficiently powerful CPU can be suitable for many applications: dimensional inspection, code reading, presence detection, shape comparison, or rule-based inspection.
For more demanding processing tasks, the number of cores, clock speed, available memory, and scalability become critical factors.
Depending on requirements, an architecture can thus be based on Intel Core, Intel Xeon, or AMD Ryzen processors, with a memory configuration tailored to the vision software and the number of data streams.
The GPU becomes particularly relevant when the application relies on extensive parallel computations: neural networks, object detection, image classification, segmentation, or the simultaneous analysis of multiple streams.
However, the GPU should not be considered mandatory for every computer vision application. Its usefulness depends on the software used, the AI models, their size, the number of streams, and the expected processing time.
The industrial architectures available from manufacturers clearly demonstrate this capability. For example, the Neousys NUVIS-5306RT can accommodate an NVIDIA card with up to 75 W and combines a CPU with camera interfaces and I/O dedicated to vision.
In anindustrial application, performance is not measured solely by computing power. The system must be able to keep pace with the production cycle: image acquisition, transfer, processing, decision-making, and, if necessary, actuator control must all be completed within the available time.
Several factors influence this latency:
It is therefore necessary to consider the entire chain.
Let’s consider a cell equipped with four industrial cameras, each producing several images per second, with automated analysis designed to detect defects in a part.
The PC must therefore be capable of:
1. Simultaneously receive the video streams from all four cameras;
In this situation, simply increasing CPU power is not necessarily the best solution. It may be more appropriate to add PoE interfaces, increase memory, integrate a GPU, or install a dedicated acquisition card. It is this comprehensive approach that allows for proper sizing of the PC.
Connectivity is a key factor in selecting an industrial vision. Industrial cameras use various standards, so the PC must have interfaces suited to the number of cameras and their data rates.
GigE Vision cameras transmit their images via Ethernet. This architecture makes it possible, in particular, to connect cameras located far from the PC and to build flexible vision systems.
The number of Ethernet ports must be sized according to the number of cameras, their data rates, and other devices on the network.
Power over Ethernet (PoE) allows data and power to be transmitted over the same network cable. This architecture can simplify the cabling of a vision cell and reduce the number of connections required.
The NUVIS-5306RT from Neousys Technology is a concrete example: its manufacturer’s specifications list 4 Gigabit PoE+ IEEE 802.3at ports, supplemented by 4 USB 3.0 ports. The platform also includes camera trigger, lighting control, and encoder input functions.
Factory Systems currently offers the NT-NUVIS-5306RT and NT-NUVIS-534RT models, two fanless Neousys PCs specifically designed for machine vision and equipped with PoE.
For USB cameras, the PC must have a sufficient number of USB 3.x ports and bandwidth compatible with the data streams to be processed. Some applications also require capture cards or specialized interfaces, particularly for Camera Link or CoaXPress.
The NUVIS-5306RT, for example, includes a PCIe slot that allows for the integration of a Camera Link or CoaXPress card. The PC should therefore be selected based on the interfaces actually used by the cameras.
The quality of processing also depends on the system’s ability to acquire images correctly. Depending on the application, it may be necessary to handle:
These functions can be integrated into the PC or added via expansion cards. The NUVIS-5306RT exemplifies this approach:Neousys specifies, in particular, 4 camera trigger outputs, 4 lighting control channels, an encoder input, 8 digital inputs, and 8 digital outputs. The manufacturer also highlights I/O management specifically designed for machine vision, with real-time control at the microsecond level.
These features demonstrate why it is important not to choose a PC based solely on its processor.
A computer vision application can generate a large volume of data.
Memory capacity must be sized based on:
The storage system must be capable of retaining, as needed, the images, inspection results, logs, and data required for traceability. An industrial SSD may be the preferred choice for applications requiring frequent access and continuous operation.
Bandwidth must be consistent with the number of cameras, their resolution, and their frame rate. It is also important to distinguish between the bandwidth available on the network and the PC’s actual capacity to process the received data.
PCIe slots may be required to integrate:
For example, the Advantech ARK-3532 is a fanless platform with multiple PCIe/PCI expansion slots designed specifically for machine vision, data acquisition, and AOI inspection applications. Factory Systems carries this platform in its Advantech product lineup.
A vision system can be installed directly on a machine, in a workshop, or near a production line. The hardware must therefore be designed to withstand the actual environment, including dust, vibrations, temperature, humidity, and integration constraints.
Fanless cooling relies on passive heat dissipation and eliminates the fan as a moving mechanical part. This can be beneficial in environments where you want to limit air intake and circulation inside the chassis.
However, choosing a fanless PC does not mean that just any model will be suitable for a machine vision application. As CPU/GPU power increases, the system’s thermal design becomes increasingly important.
The selection must therefore take into account computing power, ambient temperature, the cooling method, and the hardware’s thermal envelope simultaneously.
-> For more on this topic, see our comparison : Fanless PC vs. Fan-cooled PC, which one Should you choose?
A vision system may be technically advanced but ill-suited to its environment. Before selecting a PC, you must therefore determine:
Factory Systemes’ fanless PC product lines cover various protection levels and integration constraints, including solutions ranging from IP30 to IP69K depending on the model.
-> To learn more: Fanless PCs in industrial environments: what are the criteria?
In a computer vision architecture, transmitting the entire video stream to a remote server isn’t always practical. Local processing, or edge computing, allows images to be analyzed directly on the industrial PC.
The architecture can then follow this principle:
Camera → acquisition → preprocessing → computer vision / AI → decision → PLC or actuator
The system can thus transmit the inspection results to the higher level rather than the entire video stream. This approach can reduce network traffic and, above all, bring processing closer to the industrial process.
The Advantech MIC-770 architectures, for example, are positioned by the manufacturer for Edge AI, inference, and automated inspection applications, with options for GPU expansion. Local processing may also be appropriate when connectivity to a remote infrastructure is limited or when a decision must be made quickly at the machine level.
The images produced by a vision system are not used solely for quality control. They can alsosupport a predictive maintenance approach: detecting wear, tracking the progression of a defect, inspecting a component, or visually monitoring equipment.
For example, Factory Systems has documented the use of AI-powered vision to detect defects on wind turbine blades in real time, with local image processing and identification of defects suchas cracks or fractures. The industrial PC thus serves as a common processing hub for image acquisition, AI, and data analysis.
-> To learn more about this topic: Fanless PCs and predictive maintenance: What role does machine data processing play?
A new vision architecture does not necessarily require completely replacing a machine. As part of an industrial retrofit, a fanless PC can be integrated into an existing setup to add an inspection feature, modernize image acquisition, or introduce an AI feature without modifying the entire production system.
The design must therefore take into account the interfaces already available on the machine, the integration space, the power supply, and existing communication protocols. This approach allows for the gradual upgrading of a system while retaining a significant portion of the existing infrastructure.
-> Learn more: Fanless industrial PCs and retrofits: modernizing an existing system.
The choice should be based on the overall architecture, not on a single component.
The goal, therefore, is not to choose the most powerful PC, but rather the one whose performance, interfaces, and data acquisition capabilities are precisely matched to the vision system.
Factory Systemes offers several architectures designed to meet industrial vision needs. At Neousys Technology, the NT-NUVIS-5306RT and NT-NUVIS-534RT models are specifically marketed as fanless, ruggedized PCs for vision applications, featuring PoE and dedicated I/O.
The NUVIS-5306RT is a particularly representative example of an integrated architecture: Intel Core CPU, 4 PoE+ ports, USB 3.0, camera trigger, lighting control, encoder, digital I/O, and the option to integrate an NVIDIA card or a Camera Link/CoaXPress card.
Advantech also offers fanless architectures designed for Edge AI and vision. The MIC-770 V3, for example, can be paired with MXM GPUs, while the ARK-3532 offers several expansion options for image acquisition, processing, and inspection applications.
The final choice, however, depends on the vision software, cameras, number of streams, computing power required, necessary interfaces, and environmental constraints. For vision applications, the choice of hardware must then be refined based on computing power, interfaces, acquisition capabilities, and the production environment.
⇒ Learn more: To understand the general criteria for sizing a fanless PC, see our guide How to choose an industrial fanless PC for your application.
Can a fanless PC handle multiple industrial cameras?
Yes, provided that the processor, memory, bandwidth, and interfaces are sufficient for the number of cameras and their frame rates.
Is a GPU required for machine vision?
No. A CPU may be sufficient for many standard vision applications. A GPU becomes particularly useful for AI, deep learning, and highly parallelized processing.
Why use PoE with industrial cameras?
PoE allows data and power to be transmitted over the same Ethernet cable. This can simplify the cabling of a vision cell.
What is the difference between GigE Vision and USB Vision?
GigE Vision is based on Ethernet and is particularly useful for connecting remote cameras. USB Vision relies on a USB connection and is particularly well-suited for setups where cameras are located close to the PC and where high USB bandwidth is available.
Can a fanless PC perform real-time image processing?
Yes, if the configuration is properly scaled. However, processing time must be evaluated based on the number of cameras, the resolution, the frames per second, the algorithm used, and the machine cycle.
Which fanless PC should you choose for a machine vision application?
The choice depends primarily on the cameras, the number of streams, the vision software, the required computing power, the necessary interfaces, the expected processing time, and environmental constraints. An analysis of the entire architecture will help you select the appropriate configuration.