LUSTER CTO Yan Zhao: Industrial Large Vision Models and Intelligent Inspection Jointly Shape the Future of Industry
Recently, the BOE Innovation Partner Conference 2024 (BOE IPC 2024) opened at the Beijing Zhongguancun International Innovation Center. Dr. Yan Zhao, CTO of LUSTER, was invited to attend the forum and deliver a keynote speech titled "Industrial Large Vision Models and Intelligent Inspection Shaping the Future of Industry," sharing LUSTER's technological innovations and successful practices in general Large Vision Models and intelligent inspection.
Dr. Zhao stated that with the promotion of national strategies and the development of AI technology, intelligent inspection equipment is becoming a core driving force for the development of new productive forces. This presents both opportunities and unprecedented challenges for the industry. How to set product quality benchmarks to achieve quantifiable inspection results and how to fully utilize production data for intelligence remain key issues in the transformation and upgrading of the intelligent inspection industry.
Global leading companies like Apple and CATL are setting industry quality benchmarks by building UPI (Universal, Proactive, Inspection) super factories, pursuing an ultimate standard of almost zero defects. Dr. Zhao emphasized that lean operations and data mining are the means to achieve this goal, while universal proactive inspection is the core of UPI super factories. "Universal" means "one product, one quality standard," ensuring consistency of product quality across the global supply chain. "Proactive" represents the shift from passively receiving defective products to actively analyzing defects and optimizing production.
To achieve the goal of a super factory, the key lies in building and utilizing three core capabilities: implementing universal intelligence in the industrial field, achieving proactive defect detection and analysis, and implementing closed-loop quality management.
Universal Industrial Intelligence: Affordable and Effective
In terms of universal industrial intelligence, LUSTER has launched the LuserLVM industrial universal vision large model, unleashing universal intelligent inspection capabilities. Through layered design, this model meets the universality requirements of basic large models while optimizing for different industries and application scenarios. It demonstrates excellent performance in defect generation, assisted annotation, and defect prompting, greatly improving the efficiency and accuracy of industrial quality inspection without relying on massive computing power. It can be quickly deployed in more than 10 industrial vision inspection environments such as 3C and lithium batteries.
For example, using the industrial AIGC defect generation industry large model, we can generate over 10,000 simulated defect images from a single sample image in just one hour, improving defect generation efficiency by 80%. Utilizing the assisted annotation large model, segmentation efficiency is significantly increased by 400%. The industrial defect prompt detection large model can achieve cross-scenario application detection in one minute based on existing information.
Proactive Inspection: AI-Empowered Prevention
For proactive inspection, LUSTER adopts the integration of data, vision, and AI to comprehensively enhance the intelligence level of proactive inspection. On one hand, it uses the LuserLVM model for real-time accurate classification of defect detection; on the other hand, it employs language large models for in-depth process analysis of defect classification results, improving sudden defect response by 95%, reducing defect losses by 75%, improving analysis efficiency by 98%, and reducing manual analysis costs by 90%.
In practical applications, this large model has achieved significant results in scenarios such as lithium battery electrode coating defect detection, PVD mobile phone middle frame appearance defect detection, and display screen module appearance inspection, with classification accuracies reaching 99.25%, 98.2%, and 95% respectively.
Closed-Loop Management: 60-Second Deployment Without Downtime
In terms of quality closed-loop, LUSTER has built a cloud-edge-end integrated AI solution based on large model AI capabilities and existing quality management software, achieving full-process automation from data collection to model adjustment and deployment. It can complete rapid updates and deployment in just one minute without stopping production, bringing customers significant benefits in cost savings, production efficiency improvements, and product quality enhancements.
For example, in an application for a leading customer in the solar cell industry, through the implementation of a full-process AI+AOI closed loop, the over-inspection and miss-inspection rates were controlled to within 1%. In a collaboration with a customer in the display screen industry, through closed-loop management, rapid capacity ramp-up was achieved, saving over 9 million yuan annually. In the past two years, applications have been implemented across multiple industries, supporting a production value of nearly 300 million yuan and serving 39 customers.
Quality drives industry, intelligence leads to victory. As large model technology continues to develop and mature, LUSTER will work together with industry colleagues in the intelligent inspection industry to explore new applications of AI in smart manufacturing, connecting thousands of segmented scenarios, and continuously promoting new momentum in the intelligent inspection industry.