AI-Driven Process Modeling and Intelligent Control

Modern manufacturing is generating unprecedented volumes of process data while facing increasing demands for higher product quality, greater operational efficiency, sustainability, and autonomous operation. Artificial intelligence (AI) and machine learning have become key enabling technologies for transforming these data into actionable intelligence, empowering the next generation of smart manufacturing.

Our research develops next-generation data-centric and physics-informed AI methodologies for process modeling, monitoring, optimization, diagnosis, and intelligent control. A major focus of our work is latent variable (LV) learning, which provides powerful representations of high-dimensional industrial data by separating essential process information from noise and uncertainty. These latent representations enable accurate process understanding, predictive analytics, virtual sensing, fault diagnosis, and decision support for complex industrial systems.

Building upon classical latent variable methods, our research advances modern AI technologies through deep learning, foundation models, reinforcement learning, probabilistic modeling, tensor learning, and explainable AI. We develop scalable algorithms capable of handling nonlinear, dynamic, multimodal, and large-scale industrial processes while maintaining robustness, interpretability, and physical consistency.

Our technologies have broad applications in semiconductor manufacturing, chemical and petrochemical industries, advanced materials, biotechnology, energy systems, and intelligent manufacturing, helping industries achieve higher productivity, improved product quality, enhanced safety, reduced energy consumption, and accelerated digital transformation.

[Strategic Research Thrusts]

• Next-Generation Multi-Resolution Latent Variable Learning for Smart Manufacturing (新世代多解析度潛變數學習（Multi-Resolution Latent Variable Learning)：建構高效能智慧製造分析模型，提升製程監控與決策能力)

• Tensor Learning and Foundation Models for High-Dimensional Process Intelligence (高維製程智慧之張量學習（Tensor Learning）與基礎模型（Foundation Models）：發展適用於複雜製程的大規模AI模型，提高資料分析與知識萃取能力)

• Attention-Based Deep Learning for Autonomous Process Systems (注意力機制深度學習（Attention-based Deep Learning）：打造具自主感知與決策能力的智慧製程系統)

• AI-Driven Data Reconciliation and Fault Diagnosis with Reinforcement Learning (強化學習於資料調和與故障診斷：結合AI自主學習能力，提升製程可靠度及異常診斷準確率)

• Physics-Informed and Explainable AI for Process Systems Engineering (融合物理知識與可解釋AI（Physics-informed & Explainable AI）：建立兼具高準確度、可解釋性與可信賴性的智慧製程模型。)

• Safe Reinforcement Learning for Autonomous Process Control (安全自主製程控制（Control-aware Reinforcement Learning）：發展兼顧安全性、自主性與最佳化效能的智慧控制技術)

• LLM–Machine Learning Integration for AI-Assisted Process Design, Intelligent Control, and Engineering Decision Support (大型語言模型（LLM）與機器學習整合：打造AI輔助製程設計、智慧控制及決策支援系統，提升工程效率與決策品質)

• Continuous-Time Neural Operators for Nonlinear Dynamic Process Modeling and Predictive Control (連續時間神經運算子（Continuous-time Neural Operators）：發展高精度非線性動態製程建模與控制技術，提升複雜工業系統的即時控制能力)

Advanced Topics in Smart Process Systems Engineering

• Intelligent Process Monitoring & Fault Diagnosis for Smart Manufacturing; (智慧製造之製程監控與故障診斷技術)

Modern industrial plants are built upon highly integrated and complex processes, where multiple systems operate simultaneously to achieve optimal performance. As operational complexity increases, so does the risk of process deviations, equipment failures, and safety hazards, all of which can significantly impact productivity, product quality, and operational reliability. Our research focuses on the development and application of advanced data analytics and intelligent monitoring technologies to detect, diagnose, and predict process abnormalities in real time. By integrating cutting-edge methodologies—including multivariate statistical analysis, wavelet transforms, fractal encoding, artificial neural networks, and Hidden Markov Models—we enable more accurate process monitoring, early fault detection, and intelligent decision support. These innovative solutions have been successfully applied across a wide range of industrial sectors, including ceramic glaze manufacturing, stainless steel surface quality inspection, industrial combustion systems, and plasma etching reactors. By bridging advanced research with practical industrial applications, we help manufacturers enhance operational efficiency, improve product quality, reduce downtime, and strengthen process safety in the era of smart manufacturing.

(現代化工業製程由多種高度整合且複雜的系統與單元所組成，各製程環節彼此緊密連結，對生產效率、產品品質與設備穩定性具有關鍵影響。然而，隨著製程複雜度日益提升，任何異常或設備故障都可能導致製程效能下降、產品品質劣化，甚至引發重大安全風險。 本研究致力於開發與應用先進的資料分析及智慧監控技術，透過即時監測、異常偵測與故障診斷，提升工業製程的可靠性與智慧化管理能力。我們結合多變量統計分析、小波轉換、碎形編碼、人工智慧類神經網路（Artificial Neural Networks）及隱藏式馬可夫模型（Hidden Markov Models）等先進方法，建立高精度的製程監控與預測模型，協助企業及早發現潛在問題，降低停機風險，並提升決策效率。 本研究成果已成功應用於多項實際工業案例，包括陶瓷釉料製程、不鏽鋼鋼板表面品質檢測、工業燃燒系統以及電漿蝕刻反應器等領域。透過將學術研究與產業需求緊密結合，我們持續推動智慧製造技術發展，協助企業提升生產效率、優化產品品質、降低營運成本，並強化整體製程安全與競爭力。)

The next generation of smart manufacturing demands production systems that are intelligent, adaptive, and data-driven. As product lifecycles shorten and market demands evolve rapidly, manufacturing facilities must frequently adjust operating conditions and perform product grade transitions while consistently maintaining high product quality and operational efficiency. A major challenge across many industries is that critical product quality characteristics—such as melt index in polymer manufacturing, film thickness and critical dimensions in semiconductor fabrication, and other key process variables—cannot be measured continuously during production. Instead, quality inspection often relies on offline laboratory analysis or sampling-based metrology, resulting in delayed feedback, limited process visibility, and slower responses to equipment or process abnormalities. These limitations may lead to prolonged product grade transitions, process instability, excessive off-specification products, increased manufacturing costs, and reduced productivity. Our research develops advanced process control, data-driven soft sensing (virtual sensing), and virtual metrology technologies that enable real-time quality prediction and intelligent process optimization. By integrating process modeling, multivariate data analytics, machine learning, and artificial intelligence, we create predictive models capable of accurately estimating product quality immediately after manufacturing, eliminating the need to wait for conventional measurements. In semiconductor manufacturing, where wafer fabrication involves hundreds of highly sophisticated process steps, our virtual metrology technologies provide immediate estimation of wafer quality after each critical process. Compared with conventional sampling-based inspection, these methods enable rapid detection of equipment degradation, process drift, and abnormal operating conditions, significantly reducing defective production, improving yield, and lowering inspection costs. Complementing these capabilities, we develop advanced control strategies that optimize product grade transitions through automated control policies and optimal operating trajectories. These intelligent control systems minimize transition time, reduce process overshoot, decrease off-grade production, and enhance process stability and manufacturing efficiency. By combining intelligent process control with real-time quality prediction, our research enables predictive manufacturing, autonomous process optimization, and digital quality assurance. These technologies help industries improve product consistency, maximize production yield, reduce waste and operating costs, and accelerate the transformation toward Industry 4.0 and AI-enabled smart manufacturing.

(智慧製造的發展趨勢要求生產系統具備高度智慧化、彈性化與資料驅動能力，以因應快速變化的市場需求、多樣化產品規格及頻繁的製程切換，同時兼顧產品品質、生產效率與製程穩定性。然而，在許多工業製程中，關鍵產品品質指標，例如聚合製程中的熔融指數（Melt Index）、半導體製程中的薄膜厚度、關鍵尺寸（Critical Dimension）及其他重要品質特性，通常無法於製程中即時量測，而必須依賴離線實驗室分析或抽樣量測。此類傳統方式容易造成品質資訊延遲、製程可視性不足，無法即時掌握設備劣化或製程漂移，進而導致產品切換時間增加、製程不穩定、不良品產生及整體生產成本上升。本研究致力於發展先進製程控制、資料驅動軟感測（Soft Sensor／Virtual Sensor）以及虛擬量測（Virtual Metrology, VM）技術，結合製程建模、多變量資料分析、機器學習與人工智慧，建立高精度的產品品質預測模型，使產品在完成製程後即可立即獲得品質估測結果，而無須等待傳統量測程序。在半導體製造中，一片晶圓通常需經過數百道精密製程，每一道製程皆影響最終元件品質。我們所開發的虛擬量測技術能於關鍵製程完成後即時預測晶圓品質，相較於傳統抽樣檢測方式，可更快速偵測設備異常、製程漂移與潛在缺陷，大幅降低不良晶圓產生、提升產品良率，並有效降低量測成本。此外，我們亦發展智慧化先進製程控制技術，建立最佳化產品切換策略與自動化控制軌跡，使製程能快速、平穩地完成不同產品規格的切換，有效縮短穩定時間、降低製程超越（Overshoot）、減少不符合規格產品（Off-grade Products）的產生，並提升整體製程效率與穩定性。 透過整合先進製程控制、軟感測及虛擬量測技術，本研究打造兼具即時品質預測、自主決策與智慧最佳化能力的新世代製造系統，協助產業提升產品一致性、提高生產良率、降低原料浪費與營運成本，並加速邁向 AI 與 Industry 4.0 驅動的智慧製造新世代。)

• Intelligent Batch Process Control and AI-Driven Quality Optimization (智慧批次製程控制與 AI 驅動品質最佳化)

As global markets become increasingly competitive, manufacturers face growing pressure to shorten product development cycles while delivering superior quality, greater product customization, and faster time-to-market. These challenges are particularly significant in high-value manufacturing sectors, including pharmaceuticals, bioprocessing, specialty chemicals, and semiconductor fabrication, where batch processes are characterized by complex dynamics and stringent quality requirements. A fundamental challenge in batch manufacturing is that critical product quality information is often available only after batch completion. The lack of real-time quality feedback makes it difficult to optimize operating conditions during production, frequently resulting in quality variability, reduced productivity, and increased manufacturing costs. Our research focuses on developing next-generation data-driven control and intelligent decision-making technologies that enable autonomous quality optimization throughout the entire batch process. By integrating multivariate statistical analysis, iterative learning control, reinforcement learning (RL), and advanced artificial intelligence, we design adaptive control systems capable of continuously improving process performance based on historical production knowledge. A key innovation of our research is the development of the Quality-Aware Reward Transformer (QaRT), a novel reinforcement learning framework that overcomes the challenge of sparse quality feedback in industrial batch processes. QaRT combines mechanism-informed pseudo rewards with final product quality information to generate informative step-wise learning signals, enabling AI agents to understand how operational decisions throughout the batch influence final product quality. This dual-scale learning strategy captures both local process dynamics and global quality optimization objectives, significantly improving learning efficiency and control performance. To further address the diversity of modern manufacturing, we develop Multi-Task Inverse Reinforcement Learning (Multi-Task IRL) for intelligent multi-mode process control. By learning from historical closed-loop operating data, our framework automatically identifies different operating modes, infers the underlying control objectives, and constructs mode-specific control policies that adapt to varying production conditions without requiring extensive manual tuning. The integration of reinforcement learning, inverse reinforcement learning, and explainable AI provides a powerful framework for autonomous process optimization, adaptive control, and intelligent manufacturing. Our technologies have been applied to semiconductor manufacturing and complex batch chemical processes, demonstrating significant potential for improving product quality, enhancing process robustness, reducing operational costs, and accelerating the realization of AI-enabled smart manufacturing and Industry 4.0.

(在全球市場競爭日益激烈的環境下，製造業必須同時滿足更短的產品開發週期、更高的產品品質、更高程度的客製化需求，以及更快速的上市時程。此一挑戰在製藥、生物製程、特用化學品及半導體製造等高附加價值產業中尤為顯著，因其批次製程具有高度非線性、複雜動態及嚴格的品質要求。 然而，批次製程中的關鍵產品品質資訊通常必須待整個生產批次完成後才能取得，無法於製程進行期間即時回饋。此一特性使得製程操作難以即時最佳化，容易造成品質波動、生產效率下降及製造成本增加。本研究致力於發展新一代資料驅動製程控制與智慧決策技術，透過多變量統計分析、迭代學習控制（Iterative Learning Control, ILC）、強化學習（Reinforcement Learning, RL）及人工智慧等方法，建立具備自主學習與持續最佳化能力的智慧控制系統，使製程能夠根據歷史生產經驗持續改善操作策略，提升產品品質與製程效能。本研究的重要創新之一為提出品質感知獎勵轉換器（Quality-Aware Reward Transformer, QaRT）。此方法針對工業批次製程中最終品質回饋稀疏的問題，結合機制導向的偽獎勵（Mechanism-informed Pseudo Rewards）與批次最終品質資訊，建立具有時間序列特性的逐步獎勵模型，使強化學習演算法能有效理解整個製程中各階段操作對最終產品品質的影響。透過兼顧局部製程動態與整體品質目標的雙層監督學習機制，大幅提升學習效率、控制精度及製程最佳化能力。此外，為因應現代製造系統多產品、多操作模式的需求，本研究進一步發展多任務逆向強化學習（Multi-Task Inverse Reinforcement Learning, Multi-Task IRL）技術。透過分析大量歷史閉迴路製程資料，自動辨識不同操作模式，推估隱含的控制目標，並建立適用於各種生產模式的智慧控制策略，使控制系統能根據不同製程條件自主調整，而無需大量人工建模與參數 整定。透過整合強化學習、逆向強化學習及可解釋人工智慧技術，本研究建構具備自主學習、自適應控制與智慧決策能力的新世代批次製程控制架構，並成功應用於半導體製造及化工批次製程。研究成果可有效提升產品品質一致性、增強製程韌性、降低生產成本，並推動 AI 與 Industry 4.0 智慧製造技術的創新發展。)

• Intelligent Data Reconciliation for Smart Process Systems (智慧製程之智慧型資料調和技術)

Reliable process data are the foundation of safe, efficient, and intelligent process operations. However, measurements collected from industrial processes are often affected by instrument imperfections, random noise, and occasional gross errors, making accurate process analysis and decision-making challenging. Our research focuses on intelligent data reconciliation, which estimates the true values of process variables from imperfect measurements, thereby enhancing data quality for process monitoring, optimization, and control. Our team has developed several advanced AI-driven data reconciliation technologies:

• Neural Network–Enhanced Extended Kalman Filter (EKF): A data-driven dynamic data reconciliation framework that combines neural networks with the Extended Kalman Filter to accurately estimate dynamic process states.

• Conditional Dynamic Variational Autoencoder with Particle Filter (CDVAE-PF): An innovative deep learning framework that integrates a Conditional Dynamic Variational Autoencoder with a Particle Filter for simultaneous data reconciliation and intelligent process monitoring.

• Next-Generation Intelligent Data Reconciliation: Our latest framework further advances intelligent process analytics by integrating the CDVAE-PF model with a dual-agent reinforcement learning system. This adaptive architecture significantly accelerates the reconciliation process while maintaining high estimation accuracy, enabling more efficient real-time industrial applications.

These technologies provide a new generation of intelligent, adaptive, and computationally efficient solutions for smart manufacturing and Industry 4.0, supporting reliable process monitoring, optimization, and autonomous operation.

可靠的製程資料是實現安全、高效率與智慧化製程操作的基礎。然而，工業製程量測資料常受到儀器誤差、隨機雜訊及偶發性粗大誤差（Gross Errors）的影響，使得製程分析、監測與決策的準確性受到限制。本研究致力於發展智慧型資料調和（Intelligent Data Reconciliation）技術，透過先進人工智慧演算法，由受雜訊干擾的量測資料估測製程變數的真實值，大幅提升資料品質，進而強化製程監測、最佳化與控制能力。我們已建立多項創新的智慧資料調和方法，包括 ：

• 神經網路結合延伸卡爾曼濾波器（Extended Kalman Filter, EKF）：建構資料驅動的動態資料調和架構，有效提升動態製程狀態估測的精度。

• 條件式動態變分自編碼器結合粒子濾波器（Conditional Dynamic Variational Autoencoder–Particle Filter, CDVAE-PF）：整合深度生成式模型與粒子濾波技術，同時實現高精度資料調和與智慧製程監測。

• 新世代智慧型資料調和架構：最新研究進一步將 CDVAE-PF 與雙代理人強化學習（Dual-Agent Reinforcement Learning）結合，建立具自適應能力的智慧型資料調和系統，在維持高估測精度的同時，大幅降低計算時間，提升即時工業應用的可行性。

本研究成果為智慧製造與工業4.0提供兼具智慧化、自適應與高效率的新世代資料調和技術，促進製程監測、最佳化及自主運轉能力的全面提升。

• Advancing Intelligent Engineering through Large Language Models and Data-Driven Intelligence (融合大型語言模型與資料驅動智慧技術，推動下一代智慧工程發展智慧製造之先進製程式控制制、軟感測與虛擬量測技術)

The integration of Large Language Models (LLMs) with data-driven modeling represents a transformative direction for next-generation intelligent engineering systems. While conventional data-driven approaches demonstrate strong capabilities in capturing complex patterns from numerical and sensor data, LLMs provide complementary strengths through domain knowledge understanding, contextual reasoning, and knowledge extraction from unstructured sources, including technical documents, maintenance records, and experimental reports. By synergistically combining these capabilities, hybrid LLM–data-driven frameworks can significantly enhance predictive accuracy, improve engineering decision-making, reduce reliance on extensive labeled datasets, and enable more efficient, reliable, and adaptive design optimization. Such approaches open new possibilities for intelligent monitoring, diagnosis, control, and lifecycle management of complex industrial systems. Several innovative methodologies have been developed to advance this research direction, including:

• A generative and generalized zero-shot industrial fault diagnosis framework based on LLM-driven attribute extraction, enabling fault identification under previously unseen operating conditions without requiring extensive fault-specific training data.

• An LLM-enhanced control valve stiction detection framework that integrates time-series representation learning with the discriminative modeling capabilities of pre-trained LLMs, achieving more accurate and robust detection of nonlinear industrial equipment behaviors.

• A cross-modal industrial health monitoring framework that combines LLM-based knowledge representation with temporal feature learning for comprehensive equipment health-state prediction and intelligent maintenance decision support.

These developments demonstrate the potential of LLM-enabled engineering intelligence to bridge the gap between data-driven analytics and human-like domain reasoning, paving the way toward more autonomous, resilient, and sustainable industrial systems.

(以大型語言模型融合資料驅動智慧技術，開創新世代智慧工程 大型語言模型（Large Language Models, LLMs）與資料驅動模型的深度融合，正逐漸成為推動下一代智慧工程系統發展的重要技術方向。傳統資料驅動方法擅長從數值資料、感測訊號及歷史運轉紀錄中學習複雜特徵與潛在規律；然而，大型語言模型則進一步具備領域知識理解、語境推理，以及從技術文件、實驗報告、維護紀錄等非結構化資訊中萃取關鍵知識的能力。透過整合兩類模型的互補優勢，LLM–資料驅動混合智慧架構可有效提升預測準確度、強化工程決策能力、降低對大量標註資料的依賴，並實現更高效率、更可靠且具自適應能力的工程設計最佳化。此類創新方法為複雜工業系統之智慧監測、故障診斷、製程控制及全生命週期管理提供全新的技術途徑。針對此研究方向，已發展多項創新技術，包括：

• 基於大型語言模型屬性特徵擷取之生成式廣義零樣本工業故障診斷架構 藉由 LLM 自動萃取設備故障特徵與語意資訊，使系統能在缺乏大量故障樣本的情況下，有效辨識未知工況及未曾觀測的新型故障模式。

• 融合時間序列表徵學習與預訓練大型語言模型判別能力之控制閥卡澀（stiction）偵測方法 結合深度時間序列分析與 LLM 的高階語意表徵能力，提升複雜工業設備非線性異常行為偵測之準確性、穩健性與泛化能力。

•  結合大型語言模型知識表徵與時間序列特徵學習之跨模態工業設備健康狀態預測架構 整合多來源異質資訊，實現更全面的設備健康評估、剩餘壽命預測及智慧維護決策支援。

上述研究成果展現大型語言模型賦能工程智慧化的高度潛力，透過融合資料驅動分析能力與接近人類專家推理模式的領域知識理解能力，突破傳統智慧系統的限制，為建立更自主、更韌性、更高效率且永續發展的未來智慧工業系統提供關鍵技術基礎。)

• AI-Driven and Data-Driven Control Loop Performance Assessment (AI驅動的控制迴路效能評估與智慧診斷估)

Modern industrial plants often operate continuously for years, while hidden control-loop performance degradation remains unnoticed. Undetected deterioration in controller performance can lead to reduced product quality, unstable process operation, increased energy consumption, accelerated equipment wear, and significant losses in operational profitability. Our research develops an AI-driven, data-driven framework for intelligent control loop performance assessment and diagnostics. By learning directly from process and control data, the proposed approach continuously evaluates controller health without requiring complex first-principles process models. The framework automatically detects performance degradation, identifies root causes, and recommends corrective actions to restore optimal control performance. Integrating artificial intelligence, machine learning, intelligent diagnostic trees, statistical learning, and advanced process control, our technology enables continuous monitoring, early anomaly detection, predictive diagnostics, and performance optimization for industrial control systems. The result is a scalable and reliable solution that improves process stability, product quality, equipment reliability, and energy efficiency while accelerating the adoption of Smart Manufacturing, Digital Transformation, and Industry 4.0.  

(在現代智慧製造環境中，許多工業製程已持續運轉多年，但控制迴路的效能衰退往往在日常操作中難以察覺。當控制器性能逐漸惡化卻未被及時診斷，不僅會降低產品品質與製程穩定性，更可能增加能源消耗、設備磨損及維護成本，進而影響企業整體營運效益。我們提出結合資料驅動（Data-Driven）分析與人工智慧（AI）的控制迴路效能評估技術，透過控制訊號與製程運轉資料，自動分析控制器的健康狀態，無需建立複雜的製程數學模型，即可快速評估控制效能、診斷性能衰退的根本原因，並提供最佳化建議。本系統融合機器學習、智慧診斷樹、統計分析及先進控制理論，能持續監測控制迴路狀態，及早發現異常、預測潛在故障，協助企業提升控制品質、設備可靠性、能源效率與生產力，為智慧製造、數位轉型及工業4.0提供高效且可信賴的AI解決方案。 )

• AI-Powered Image-Based Process Monitoring and Intelligent Visual Analytics (AI影像式製程監測與智慧視覺分析)

Advances in optical sensing, high-speed imaging, and industrial vision systems have transformed process images into one of the richest sources of information for modern process monitoring. However, converting massive streams of visual data into reliable and actionable insights remains a major challenge for intelligent manufacturing. Our research develops AI-powered, image-based monitoring technologies that combine computer vision, deep learning, and data-driven analytics to automatically interpret complex process images without manual feature engineering. By learning hierarchical visual representations directly from image data, deep neural networks can identify subtle patterns, detect anomalies, classify operating conditions, and predict process behavior with unprecedented accuracy. Leveraging state-of-the-art deep learning architectures, our intelligent visual analytics framework transforms large-scale image data into real-time decision support for process monitoring, fault diagnosis, quality assurance, and predictive maintenance. The technology significantly enhances process reliability, operational safety, and production efficiency while enabling autonomous monitoring in next-generation smart factories. Our AI-based image analytics have been successfully applied to a wide range of industrial and research problems, including flame image-based combustion monitoring, ultrasonic time-domain reflectometry (UTDR) for membrane filtration fault diagnosis, in-situ crystal size measurement and morphology analysis, and other vision-enabled process monitoring applications. These technologies bridge artificial intelligence and process engineering to realize intelligent sensing for Smart Manufacturing and Industry 4.0.

(隨著光學感測、高速影像擷取及機器視覺技術的快速發展，製程影像已成為智慧製造中最具價值的資訊來源之一。然而，如何從大量且複雜的影像資料中，自動萃取有用資訊並轉化為決策依據，仍是現代製程監測的重要挑戰。我們致力於發展結合人工智慧、深度學習、電腦視覺及資料驅動分析的影像式智慧監測技術，使系統能直接從大量製程影像中自動學習多層次特徵，無需人工設計特徵，即可精確辨識製程狀態、偵測異常、診斷故障，並預測製程發展趨勢。透過先進深度神經網路與智慧影像分析技術，我們建立即時且自主的視覺監測平台，提供高精度的製程監控、品質檢測、故障診斷及預知維護能力，有效提升製程穩定性、生產品質、設備可靠度及生產效率，實現智慧工廠與工業4.0所需的自主感知能力。本研究成果已成功應用於多項工業與學術案例，包括火焰影像燃燒監測、膜過濾系統之超音波時域反射（UTDR）故障診斷、晶體三維尺寸與形貌之原位量測，以及其他AI影像式製程監測技術，展現人工智慧與製程工程深度融合的創新價值。 )

• AI for Intelligent Process Integration, Sustainable Design, and Energy-Efficient Optimization

The transition toward net-zero emissions, sustainable manufacturing, and smart industries demands a new generation of process systems that are not only energy-efficient but also intelligent, adaptive, and environmentally responsible. Our research integrates Artificial Intelligence (AI), data-driven modeling, process systems engineering, optimization, digital twins, and advanced process control to develop innovative solutions for next-generation chemical and environmental processes. By combining physics-based knowledge with machine learning and optimization algorithms, we design intelligent operation strategies that minimize energy consumption, reduce carbon emissions, maximize resource utilization, and enhance process reliability. Our research bridges AI and process engineering to enable autonomous decision-making, optimal process integration, and sustainable manufacturing for Industry 4.0.

Representative Research Areas

• AI-Enabled Solar-Driven Membrane Distillation: We develop intelligent membrane distillation systems powered by renewable solar energy for sustainable water purification and desalination. AI-assisted operational optimization enables robust performance under fluctuating solar irradiation and changing environmental conditions, providing energy-efficient and decentralized water treatment solutions.

• AI for CO₂ Capture, Process Integration, and Decarbonization: Reducing industrial carbon emissions is one of the most critical challenges for sustainable manufacturing. Our research applies AI-driven process synthesis, superstructure optimization, and intelligent decision support to design economically viable CO₂ capture and emission reduction technologies for existing industrial plants while simultaneously minimizing SO₂, NOₓ, and overall environmental impacts.

• Intelligent Control and Optimization of Membrane Filtration Systems: Membrane fouling remains the primary limitation affecting filtration efficiency and operating cost. We develop AI-enhanced Iterative Learning Control (ILC) and adaptive optimization strategies that continuously learn from cyclic filtration operations to improve productivity, reduce fouling, extend membrane lifetime, and minimize energy consumption.

• Hybrid Membrane–Distillation Systems for Energy-Efficient Separation:  Conventional distillation is one of the most energy-intensive operations in the chemical industry. Our research combines membrane technology with distillation through AI-assisted process integration and optimal design to overcome thermodynamic limitations such as azeotropes and close-boiling mixtures, significantly improving separation efficiency while reducing energy demand and operating costs.

• AI-Driven Sustainable Biofuel Production : We design intelligent integrated reactor–membrane systems for high-efficiency biodiesel production. By combining reaction engineering, membrane separation, and AI-based operational optimization, our technologies increase biodiesel purity, improve conversion efficiency, reduce downstream purification requirements, and support the development of renewable energy systems.
• Smart Adsorption and Advanced Separation Processes: Adsorption remains one of the most important technologies for chemical purification and pharmaceutical production. We develop AI-assisted adsorption process design and optimization that enhance separation performance, improve productivity, and enable intelligent operation for large-scale industrial manufacturing as well as high-value specialty chemical and pharmaceutical applications.

Research Vision

Our long-term vision is to create AI-native process systems that seamlessly integrate sensing, modeling, optimization, and autonomous control into a unified intelligent platform. Through the convergence of Artificial Intelligence, Process Systems Engineering, Digital Twins, Sustainable Energy Technologies, and Green Manufacturing, we aim to develop transformative technologies that accelerate industrial decarbonization, improve energy efficiency, and contribute to a cleaner and more sustainable future.