本研究探討新建半導體廠燃燒濕式洗滌器（burn wet scrubber）之削減率（Destruction and Removal Efficiency， DRE）最佳化，並採用先進污染減排技術。為提升 DRE削減率之表現，本研究提出 FMEA-RSM-KANGA 框架。在此框架中，首先應用失效模式與效應分析（Failure Mode and Effect Analysis， FMEA）來識別失效模式，並評估影響 DRE 的關鍵作業參數。接著，利用部分因子設計（Fractional Factorial Design， FFD）及變異數分析（Analysis of Variance， ANOVA）確定主要影響因子，包括 CH4 flow、Burn CDA flow及Hot N₂ flow。透過反應曲面法（Response Surface Methodology， RSM）建立二階模型，以進一步優化操作設定。最後，本研究採用 Kolmogorov-Arnold Network（KAN）模型與基因演算法（Genetic Algorithm， GA）來尋找全域最優解，確定最佳操作條件為 CH4 flow = 23 slm、Burn CDA flow = 37.9 slm及Hot N2 flow = 17 slm.。此方法顯著提升 DRE 率，從 41.38% 提高至 95.15%，達到產業目標標準（95%）。本研究所提出的框架能有效優化操作條件，透過數據驅動建模及 AI 最佳化技術，大幅減少有害氣體排放並提升處理效率。;This study investigates the Destruction and Removal Efficiency (DRE) optimization of burn wet scrubbers for a newly constructed semiconductor facility， leveraging advanced pollution reduction technologies. A novel FMEA-RSM-KANGA framework is proposed to enhance DRE performance. Developing this framework， initially Failure Mode and Effect Analysis (FMEA) was applied to identify failure modes and evaluate critical operational parameters influencing DRE. Fractional factorial design (FFD) and ANOVA (Analysis of Variance) were employed to determine the key factors of CH4 flow， Burn CDA flow， and Hot N2 flow. A second-order model was developed using Response Surface Methodology (RSM) to refine operational settings. Subsequently， a Kolmogorov-Arnold Network (KAN) model and Genetic Algorithm (GA) were utilized to identify the global optimum， achieving optimal operational conditions at CH4 flow = 23 slm， Burn CDA flow = 37.9 slm， and Hot N2 flow = 17 slm. This approach greatly improved the DRE rate from 41.38% to 95.15%， meeting the targeted industry standard of 95%. The proposed framework effectively optimizes operational conditions， achieving a substantial reduction in harmful gas emissions and an increase in efficiency through data-driven modeling and AI-based optimization.
