Today’s modern production environments face the challenge of digitizing various manual processes. This is the only way to remain competitive and survive in the market. To digitize processes, various prerequisites are needed: network-capable devices, the network itself, and edge devices that can record, exchange, and process data.
Artificial intelligence (AI), based on powerful hardware and software solutions, plays a crucial role in modern factories. It helps with process optimization, detects product or component faults in real time, and can support predictive maintenance. One example of a digitized process is Automated Optical Inspection (AOI). It uses high-resolution cameras and optical sensors to check components and products for defects, deviations or irregularities with the help of AI.
AI-powered AOI helps detect errors
Let’s stick to an example: If circuit boards in electronics production are to be checked for quality using a manual process, this leads to errors, for example, due to employee fatigue or limits in acuity. This causes costly rework, rejects, and warranty returns. (Fig. 1)
This is where AOI comes into play: AI algorithms are trained to recognize complex defect patterns that are difficult to detect using conventional methods. The AI system learns from the data over time, improving its ability to detect subtle defects and reduce scrap. Instead of programming specific defect types, the AI learns what “normal” looks like and flags anything unusual to detect new defects, not foreseen in the training data.
AI also helps to improve the quality of products with additional functions, including:
- Decision support through probability assessments for different types of defects.
- Process correlation, which can identify the causes of quality problems.
- Transfer learning reduces training data requirements for new components or products.
AOI system with SECO- and Axelera AI-powered hardware
SECO and Axelera AI offer integrated hardware and software solutions that enable developers to implement projects quickly and easily. Specifically, SECO provides System-on-Modules (SoMs), also known as Computer-on-Modules (CoMs), which allow shortened time-to-market, reduced development costs, and integrated AI acceleration.
All these benefits can be seen in SECO’s SOM-COMe-BT6-RK3588. This COM Express Type 6 module pairs a Rockchip RK3588 processor, which incorporates four Cortex-A76 cores, four Cortex-A55 cores, and the Mali G610 MC4 GPU, with the AI acceleration of Axelera AI’s Metis AI Processing Unit (AIPU) soldered directly onto the module. (Fig. 2)
The AIPU offers AI performance of 214 tera-operations per second (TOPS) with INT8 precision at an outstanding energy efficiency of 15 TOPS/W. The processors are supported by 32 GB LPDDR5 memory for the CPU, 2 GB LPDDR4x memory for the AIPU, and extensive high-performance interfaces, including:
- 1x Gigabit Ethernet (GbE)
- 2x USB-C with DP-Alt-Mode
- 4x USB 5 Gbps
- 1x USB High-Speed
- 2x PCIe Gen3 lanes
- 2x PCIe Gen2 lanes
Due to the high performance of the module and its broad scalability, it can be easily and quickly integrated into AOI systems for modern production facilities. For example, the module:
- Enables the connection of multiple high-resolution MIPI-CSI, Ethernet, or USB cameras
- Runs AI models for defect patterns, anomalies, and quality deviations, without an off-module GPU or AI accelerator
- Reduces latency by processing images directly on the module instead of sending them over a network to another internal network device or a cloud
Highly integrated software stack
In addition to powerful hardware, appropriate software integration is required to analyze the recorded data. This is where the SECO IoT software suite, Clea, comes into play. Clea consists of cloud and edge software modules, orchestrating data analysis and communications (via Clea Astarte) and managing fleets of devices at multiple levels of granularity (via Clea Edgehog). An entire IoT installation is managed via web-based access via the Clea Portal. The Clea design maximizes scalability, supporting everything from a minimal number of local nodes to very large, geographically dispersed networks—all managed at user-defined levels of granularity.
Clea OS, an embedded Linux distribution based on the Yocto Project, can facilitate the recording and processing of video streams from multiple cameras connected via Ethernet or USB. The OS also makes it possible to localize decision-making thanks to its ability to handle complex data stream analysis and management on edge nodes. For example, it can be used to automate action on detected errors and trends, such as whether an anomaly is within a given tolerance. Clea OS additionally facilitates communication with an IoT cloud, whether based on Clea or other software packages that use industry standards.
With Clea, developers have a single platform on which to operate, manage, and control their AOI systems—within an assembly line, across multiple assembly lines within a factory, or across factories widely dispersed, even worldwide. (Fig. 3)
Additionally, Clea OS integrates the Axelera AI Voyager SDK, which supports intelligent edge AI solutions. It provides easy software integration for AI inference at the edge, with built-in tools to quickly evaluate Metis performance, accuracy, and energy consumption. It also enables fast end-to-end integration with optimized pipelines as well as direct access to inference hardware via low-level APIs.
This hardware and software infrastructure provides the optimal platform for developers to design end-to-end AI solutions that can detect, classify, and respond to defects in real-time, and collect data for AI-based workloads.
Conclusion
SECO and Axelera AI provide developers with outstanding hardware and software solutions, ready to create powerful, future-proof, and scalable AI-based computer vision solutions such as AOI systems to digitize manual processes in modern production environments.
Ready to start your AOI design? Connect with SECO experts now!