The Chinese textile industry is undergoing a quiet efficiency revolution. While most companies struggle with data silos between ERP and MES systems, early adopters are leveraging AI for direct production decisions. On July 27, Wenge launched the 'Industrial Decision Brain' in Keqiao, Shaoxing, integrating perception, analysis, and execution into a closed loop, aiming to solve the long-standing dilemma of 'rich data but poor decisions' in textile factories.
Background: Three Gaps in Textile Intelligence
Textile factories are not short of data. Over the past decade, systems like ERP, MES, and PLC have been widely deployed, accumulating multi-dimensional information on orders, equipment, processes, and quality. However, industry data shows that less than 20% of this data is used for optimizing scheduling, controlling quality, or aiding decision-making. Zeng Dajun, deputy director of the Institute of Automation at the Chinese Academy of Sciences, noted that decision-making is shifting from human-centric to human-agent collaboration, posing new challenges in data processing and human-machine decision-making.
Professor Liu Yuan from Zhejiang University highlighted a deeper issue: AI's technical capabilities do not automatically translate into industrial value. Data, computing power, and algorithms must be deeply integrated with specific industry scenarios, business processes, and institutional environments. This means textile companies need more than another general-purpose large model; they need a system that understands process mechanisms, adapts to complex modal data, and structures the tacit knowledge of senior technicians.
Technical Path: From Data to Decision
The core architecture of the Industrial Decision Brain is DOMA (Data-Ontology-Models-Agents) and the 'Base-Pivot-Core-Brain-Edge' decision AI tree. The bottom layer uses Haihui and DIP to organize industrial data and knowledge, the middle layer uses Yayi, Panshi large models, and decision engines for prediction and simulation, and the top layer uses Dragon Worker agents and industrial smart equipment for execution. The system does not require companies to scrap existing systems but connects order, equipment, material, process, and quality data through a unified business language, generating executable plans under constraints of delivery time, capacity, cost, and quality.
CTO Cao Jia summarized three barriers to industrial AI deployment: data that is hard for AI to understand, models that are hard to adapt, and expertise that is hard to preserve. The Industrial Decision Brain addresses these by transforming data into AI-readable structures and using agents to call business systems and tools, forming a 'perception-decision-execution-feedback' loop. For example, in dyeing color matching, the system integrates historical recipes, spectral data, and fabric information to generate suggestions, reducing sample rounds from 7-8 to 3-4 and cutting costs by 30%-50%, while continuously capturing expert knowledge.
Results: From Point Tools to Full-Process Decisions
The Industrial Decision Brain has achieved phased results in key textile processes. In trade order processing, verification time dropped from 30-60 minutes to tens of seconds, with key element extraction accuracy exceeding 95%. In weaving scheduling, AI integrates constraints of equipment, capacity, processes, inventory, and delivery, compressing scheduling from hours to minutes and enabling real-time response to rush orders and equipment anomalies. In dyeing vat scheduling, time was reduced from 2 hours to 2 minutes. In order delivery, cross-department collaboration was cut from 1.5-3 hours to 5 minutes.
Notably, the system extends to physical production via embodied intelligent robots from Zhongke Juxi. The inspection robot identifies over 20 types of textile defects with accuracy over 95%; the yarn loading robot operates at heights of 0.5-3 meters with ±0.1 mm repeat positioning accuracy; the cloth unloading robot has a rated load of 300 kg. These robots serve as physical terminals, converting system decisions into actions like stopping machines, loading yarn, and transporting goods.