In automotive manufacturing, assembly verification has always been the core of quality assurance. Previously, many car manufacturers without robotic arms and AOI (Automated Optical Inspection) real-time inference systems commonly faced the following challenges: as car models diversified and assembly parts became more complex, it became difficult to distinguish between good and defective products. Furthermore, fragmented verification data collection couldn’t provide strong support for subsequent quality analysis and process optimization. However, with the maturity of AI edge computing and advanced machine vision technologies, the combination of high-resolution industrial cameras and embedded AI real-time inference systems can not only complete hundreds of inspection items within minutes but also improve the yield rate of inspection tasks. This transformation shifts production lines from traditional “experience-based inspection” to “data-driven intelligent verification,” effectively enhancing overall quality management and production efficiency.
To meet the high precision, real-time performance, and traceability requirements of automotive assembly verification, a dual-vision inspection architecture incorporating the PSYS-508-Q670 and PJAI-100-OX is introduced to create a complete intelligent quality inspection environment. The PSYS-508-Q670 training server, equipped with 14th/13th/12th generation Intel® Core™ i processors, supports 128GB DDR5 memory and 7 SATA 6.0Gb/s storage interfaces. It can rapidly process vast image data from the production line and accelerate the deep training of AI models for appearance inspection and assembly verification using GPU acceleration. The trained models are then securely deployed to the edge-based PJAI-100-OX inference system via HTTPS/SFTP secure protocols. The PJAI-100-OX serves as the intelligent inference core. This system, powered by the NVIDIA® Jetson Orin NX SOM, boasts 100 TOPS of inference performance, enabling complex image analysis to be completed within milliseconds. Its configuration includes 2 Gigabit LAN ports, 3 USB 3.2 Gen1 interfaces, and multiple M.2 expansion options (E-Key for WiFi/BT, M-Key for NVMe storage), ensuring high-speed connection and stable communication with industrial cameras, robot arm control boards, and other equipment.
In practical applications, the entire image data flow begins at the sensing layer. During the assembly verification confirmation process, 3D vision inspection cameras capture subtle differences on surfaces such as car door trim parts and headlamp covers, and instantly infer whether their flushness and gap meet design specifications. These inspections not only ensure aesthetic quality but also directly impact airtightness and safety. Through AI inference, the system real-time analyzes images of the car body surface, gaps, logos, lamps, and other features, quickly identifying minor defects and assembly deviations, automatically classifying good and defective products, and continuously learning abnormal patterns to reduce batch defects.
When the system infers an anomaly (e.g., out-of-tolerance flushness, incomplete adhesion), the PJAI-100-OX can send the abnormal image and structured data back to the PSYS-508-Q670, initiating AI model fine-tuning and retraining to continuously optimize inference accuracy. Furthermore, the overall architecture supports EtherCAT and Profinet industrial communication protocols, ensuring data synchronization and real-time interoperability between cameras, robotic arms, lighting systems, HMIs, and other equipment, forming a highly collaborative and intelligently aware manufacturing environment. Through the perfect pairing of the PSYS-508-Q670 and PJAI-100-OX, these two systems work in concert to create a cycle of model training → model deployment → real-time inference → anomaly feedback → retraining, fully meeting the stringent requirements of “high-accuracy defect identification” and “high-reliability operational stability” in automotive manufacturing production lines. This solution, by instantly establishing a complete data traceability mechanism, automatically records and feeds back every inspection image and judgment result to the car manufacturer’s traceable quality database, supporting subsequent analysis and process optimization, and significantly enhancing the basis for future assembly technology or work decisions.
In addition to standardized product platforms, Portwell leverages its years of accumulated DMS (Design and Manufacturing Services) expertise to further assist customers in tailoring intelligent manufacturing solutions that best adapt to market changes. DMS services provide one-stop support from initial requirement analysis and customized hardware design to firmware optimization, system integration, certification (CE/FCC/UL, etc.), and mass production delivery. Particularly addressing the automotive production line’s requirements for environmental resistance, long lifecycles, and high reliability, DMS can customize designs based on application scenarios, such as IP-rated protection, wide-temperature operation, and platforms with a guaranteed supply of over ten years, ensuring worry-free customer deployment. In today’s increasing prevalence of AI edge computing, machine vision, and smart production line integration, Portwell is not just an embedded hardware supplier but also the best accelerator for automotive industry customers to achieve intelligent manufacturing upgrades, working hand-in-hand with customers to create future competitiveness.