Projects | Youcheng Zong

Digital-Twin and LLM Decision Support for Thickening–Dewatering

Sun, 01 Mar 2026 00:00:00 +0000

This project developed an operational decision-support workflow for a thickening–dewatering line. It joins field measurements, condition recognition, a process digital twin, and an LLM layer with retrieval, Text2SQL, and controlled tool use. The public description omits the industrial partner, site, and operating values.

System scope

process-data access and online condition recognition;
a digital representation of the thickening–dewatering equipment;
retrieval over technical documents and structured production data;
tool-mediated analysis with the operator retained in the decision loop.

Research connection

The project supplied an application setting for work on specialist-model adaptation under operating shifts. The associated preprint presents the method and evaluation protocol without exposing production identifiers.

Multimodal LLM-Assisted Analysis for Oral Mucosal Imaging

Sun, 01 Feb 2026 00:00:00 +0000

This ongoing project studies a multimodal workflow for oral mucosal dermoscopy. Images, patient reports, and medical knowledge are combined for structured analysis and clinician review. It is decision support and does not replace clinical judgment.

Research components

oral microvessel and capillary analysis from dermoscopy images;
retrieval of relevant medical knowledge and case context;
multimodal reasoning over images and structured patient information;
evidence presentation and clinician review before any final interpretation.

Related work includes zero-shot capillary segmentation with SAM2 and connectivity-preserving slender-structure segmentation. Only figures from public research outputs are used on this page.

Visual Perception and Energy-Saving Control for LF Refining

Thu, 01 Jun 2023 00:00:00 +0000

This project linked visual perception with process control for ladle-furnace refining. It combines furnace imaging and slag-state analysis with a bottom-blowing control model connected to the plant PLC. Partner and site identifiers are omitted.

System scope

process imaging at the LF furnace;
visual analysis of slag state;
bottom-blowing mode selection and control logic;
PLC communication and integration with production software.

The public page reports the system boundary and completed integration only; confidential operating settings and plant-specific control rules are not disclosed.

Intelligent Transformation and State Recognition for Blast-Furnace Iron Tapping

Thu, 01 Dec 2022 00:00:00 +0000

The system combines field instrumentation and communication with molten-iron level tracking, Bi-GRU state recognition, post-processing, and an operator interface. The industrial partner and furnace identifiers are omitted from the public version.

Technical focus

continuous extraction of level-related process signals;
temporal state recognition using a Bi-GRU composite model;
post-processing rules for stable tapping-state decisions;
integration of algorithm services with industrial data and visualization.

Public outputs

The work led to two first-author journal articles and the first-inventor patent CN116678473B, which covers a method for judging the tapping state of a blast-furnace hot-metal ladle.