Industrial Intelligence Core R&D
Multi-field research and development directions
Industrial Large Model R&D
• Develop vertical-specific lightweight large models to overcome the pain points of general-purpose models: “insufficient data, low accuracy, difficult deployment”
• Achieve efficient output of core capabilities such as equipment status prediction and process optimization
• Covering industrial manufacturing, with average accuracy ≥90%
Knowledge Distillation Optimization
• Transfer core capabilities from complex large models to lightweight small models for real-time adaptation on edge devices, solving issues of “limited computing power and high energy sensitivity”
• After distillation, model volume is reduced by 70%, inference speed increased by 50%, maintaining ≥90% accuracy
AI Self-Training Platform Evolution
• Automate the entire AI process: “data cleaning – annotation assistance – model tuning – performance evaluation”
• Generate custom models with minimal scenario data; a new algorithm training cycle takes only 4-8 weeks
• Compatible with domestic CPUs/GPUs/NPUs, fully meeting project requirements for secure and trustworthy IT systems (XinChuang)
Multi-Scenario Research Areas
“
Abstract: GSmart focuses on core technology research in industrial AI, exploring innovative applications of artificial intelligence in scenarios such as ports, mines, and manufacturing, from multi-modal perception and edge computing to self-learning models. Below are our selected cutting-edge research achievements and industry insights.
Academic Collaboration / Technology Evolution Zone
GSmart is collaborating with multiple universities, research institutes, and AI labs to advance research and industrial application in the fields of industrial vision algorithms, edge computing architectures, and secure AI.
In industrial vision, we are jointly developing high-precision defect detection and intelligent sorting recognition algorithms to help manufacturing improve quality and efficiency. In edge computing, we are exploring lightweight architectures and low-latency deployment solutions to tackle the challenge of real-time industrial data processing. In secure AI, we are building multi-dimensional risk identification models to strengthen network and device security. Through resource sharing and complementary expertise, we continuously drive the translation of research outcomes into industrial applications, injecting momentum into innovation in related fields.
