IN Brief:
- Five Zhen Ding and NTHU research projects have been selected for presentation at IEEE CASE 2026.
- Work covers production optimisation, smart logistics, supply resilience, materials risk, and high-mix, low-volume manufacturing.
- An RF anomaly-detection system has already been validated in production and identifies abnormal frequencies within seconds.
Zhen Ding Technology and National Tsing Hua University have developed five smart-manufacturing research projects spanning PCB and semiconductor production, with work covering process optimisation, logistics scheduling, supply-chain resilience, raw-material risk, and AI-assisted management of high-mix, low-volume factories.
The projects have been selected for presentation at the 22nd IEEE International Conference on Automation Science and Engineering, being held in Shenyang from 17 to 21 August. They have emerged from the Zhen-Ding & NTHU Joint Research Center, established in 2020 and now entering the second five-year phase of its industry-academia programme.
Total R&D funding across the two phases exceeds NT$100 million. Current research extends beyond individual machine optimisation into higher-level manufacturing problems including high-end IC substrates, heterogeneous integration, materials availability, and the coordination of complex production flows.
One project has already moved beyond academic evaluation. An online hierarchical RF signal anomaly-detection and localisation system developed by NTHU master’s student Chen-Yi Kuo with Zhen Ding engineers has been validated in an actual production environment.
The system combines multi-scale partitioning, weighted voting, and AI agents to identify defects and locate abnormal frequencies within seconds. Zhen Ding says the technology is intended to improve anomaly diagnosis and production efficiency, with its deployment providing a useful distinction between a research algorithm and one that has been exposed to live manufacturing data.
That distinction matters in PCB production because process-monitoring systems have to separate genuine faults from ordinary variation. Electrical signals, machine states, material changes, tooling condition, and product mix can all alter the data generated by a line, and a diagnostic method that works against a clean research dataset can become considerably less reliable when exposed to routine production noise.
RF-domain monitoring adds another layer because abnormal behaviour may not appear as one obvious threshold excursion. Changes can be distributed across several frequencies or emerge only as the interaction between multiple signals. Hierarchical analysis gives the detection system a way to divide that information into different scales before combining the results into a diagnosis.
The other four research areas address problems that sit around the process equipment itself. Smart logistics scheduling deals with the movement and timing of materials and work-in-progress, while supply-chain resilience and raw-material risk affect whether a planned production sequence can proceed at all. High-mix, low-volume manufacturing adds frequent product changes, smaller batch sizes, and more combinations of recipes, materials, and inspection requirements.
Those conditions make fixed production rules harder to maintain. A scheduling model might improve equipment utilisation but create additional material movements, changeovers, or downstream queues; an inventory policy can reduce working capital while increasing exposure to an unavailable laminate or chemical. Useful optimisation therefore has to work across several manufacturing constraints rather than maximise one measurement in isolation.
The inclusion of raw-material risk is particularly relevant to advanced PCB and IC-substrate production. Laminate systems, copper foils, process chemicals, drilling and imaging consumables, and specialist materials can require lengthy customer qualification. A nominally equivalent substitute may still alter electrical, thermal, mechanical, or process behaviour, making material availability an engineering issue rather than a purchasing decision alone.
Zhen Ding’s research programme provides a route for testing academic work against those factory constraints. The joint centre brings university researchers into contact with real process data and operating environments, while the manufacturer gains access to methods that can be assessed before wider deployment across production sites.
The five papers demonstrate the breadth of that collaboration, but the strongest evidence comes from the project already working on a production line. For the remaining research, comparable deployment results — shorter diagnosis times, lower process variation, improved scheduling, reduced material exposure, or measurable yield gains — will determine whether conference work becomes repeatable manufacturing capability.
