https://ir.lib.ncu.edu.tw/handle/987654321/81 Search the Channel s https://ir.lib.ncu.edu.tw/simple-search https://ir.lib.ncu.edu.tw/handle/987654321/111372 title: 生命週期評估對 AI Slop 影片生成過程中的環境衝擊分析;生命週期評估對 AI Slop 影片生成過程中的環境衝擊分析 abstract: 近年來生成式人工智慧（Generative Artificial Intelligence， Generative AI）技術快速發展，帶動文字、圖片、影音等內容自動生成應用蓬勃成長。其中，以大量自動化產製、內容品質較低且重複性高之影音內容，逐漸形成所謂「AI Slop」現象。由於生成式影音模型需仰賴高效能圖形處理器（Graphics Processing Unit， GPU）、大型資料中心及冷卻系統支援，其背後伴隨可觀之能源消耗、水資源使用及溫室氣體排放。然而，目前相關研究多聚焦於人工智慧技術效能與商業應用，對於AI Slop內容生成所衍生之環境衝擊探討仍相對有限。 本研究以生命週期評估（Life Cycle Assessment， LCA）方法為基礎，探討AI Slop影片生成過程之能源消耗與環境衝擊。研究以生成30秒AI Slop影片作為功能單位（Functional Unit），建立涵蓋上游能源供應、中游模型運算與資料中心運作，以及下游資料傳輸與內容發布之系統邊界，並蒐集公開文獻、產業報告及資料庫資訊建構生命週期盤查（Life Cycle Inventory， LCI）資料。透過LCA軟體進行環境衝擊分析，評估其全球暖化潛勢（Global Warming Potential， GWP）、能源消耗、水資源使用及其他相關環境衝擊指標，並進一步執行敏感度分析與不確定性分析，以驗證研究結果之穩健性。 研究結果顯示，以30秒AI Slop影片為功能單位，其碳排放量約為0.35 kg CO₂e，於18項ReCiPe環境衝擊指標中，以氣候變遷（Climate Change）、化石資源耗竭（Fossil Resource Scarcity）及人類毒性（Human Toxicity）等類別較為顯著。於18項ReCiPe環境衝擊指標中，以氣候變遷（Climate Change）、化石資源耗竭（Fossil Resource Scarcity）及人類毒性（Human Toxicity）等類別較為顯著。 熱點分析結果顯示，GPU運算為AI Slop影片生成過程中最主要之環境熱點，其環境衝擊貢獻約占整體系統之58%；其次為資料中心冷卻系統，約占整體環境衝擊之24%，顯示運算需求與能源供應效率為影響環境衝擊之關鍵因素。 敏感度分析結果顯示，AI模型運算量為影響環境衝擊結果之最主要因子。當模型運算需求增加20%時，碳排放量約增加26%；反之，當運算需求降低20%時，碳排放量約下降22%。此外，電網排放係數及GPU能耗亦對最終結果具有顯著影響。 不確定性分析結果顯示，30秒AI Slop影片之碳排放量呈近似常態分布，主要集中於0.25至0.47 kg CO₂e之間，平均值約為0.36 kg CO₂e，顯示本研究評估結果具有良好穩定性與可信度。 本研究證實AI Slop內容生成雖具有便利性與低成本特性，但其背後仍伴隨能源消耗及環境負荷。未來研究可進一步納入不同AI模型、資料中心區域配置及設備生命週期等因素，以提升人工智慧環境衝擊評估之完整性與準確性。;In recent years， the rapid advancement of Generative Artificial Intelligence (Generative AI) has significantly accelerated the development of automated content creation， including text， images， audio， and videos. Among these applications， a phenomenon known as “AI Slop” has emerged， referring to large quantities of low-quality， repetitive， and automatically generated content. The generation of AI-based video content relies heavily on high-performance Graphics Processing Units (GPUs)， large-scale data centers， and cooling systems， resulting in considerable energy consumption， water usage， and greenhouse gas emissions. However， existing studies have primarily focused on the technical performance and commercial applications of AI， while limited attention has been paid to the environmental impacts associated with AI-generated content， particularly AI Slop videos. This study applies the Life Cycle Assessment (LCA) methodology to evaluate the energy consumption and environmental impacts of AI Slop video generation. A 30-second AI Slop video was defined as the functional unit (FU). The system boundary includes upstream energy supply， core processes involving AI model inference and data center operations， as well as downstream data transmission and content distribution. Life Cycle Inventory (LCI) data were compiled from publicly available literature， industry reports， and environmental databases. Environmental impacts were assessed using LCA software， including Global Warming Potential (GWP)， energy consumption， water use， and other environmental impact indicators. Sensitivity analysis and uncertainty analysis were further conducted to evaluate the robustness of the assessment results. The results indicate that the environmental impacts of AI Slop video generation are primarily associated with AI model inference and data center operations. GPU computation and cooling systems were identified as the major environmental hotspots. Although the environmental burden of a single AI-generated video is relatively limited， the cumulative impacts could become significant due to the large-scale generation and dissemination of AI content. In addition， factors such as electricity generation mix， data center energy efficiency， and AI model complexity were found to substantially influence the overall environmental performance. This study establishes a quantitative assessment framework for evaluating the environmental impacts of AI Slop video generation and provides a foundation for future research on sustainable AI development. The findings may serve as a reference for AI service providers， data center operators， and policymakers in developing energy management strategies， carbon disclosure mechanisms， and sustainability-related policies， thereby promoting a balance between technological innovation and environmental sustainability. <br> https://ir.lib.ncu.edu.tw/handle/987654321/111369 title: 隱形眼鏡製造廠溫室氣體盤查及減量效益之評估研究;隱形眼鏡製造廠溫室氣體盤查及減量效益之評估研究 abstract: 本研究以位於臺灣北部之A公司為研究對象，主要從事軟式隱形眼鏡 製造及醫療光學產品研發。依據ISO 14064-1:2018建立組織層級溫室氣體盤查制度，分析2021年至2024年間之溫室氣體排放結構、排放趨勢及減量效益，並結合同業公開揭露資料，比較隱形眼鏡製造業之排放特性及減碳作法，以作為企業推動溫室氣體管理及減碳策略之參考。 研究結果顯示，A公司總溫室氣體排放量由2021年55，059.3160公噸 CO₂e增加至2023年61，990.3931公噸CO₂e，增幅約12.6%，2024年則下降至59，576.7160公噸CO₂e，顯示整體排放變化與產量提升、產線擴增及營運規模具有密切關聯。其中，一廠為主要生產據點，研究期間排放量長期占公司總排放量約九成；二廠則因包裝線擴增及供應鏈活動增加，使排放量持續成長。排放結構分析結果顯示，類別二「輸入能源」平均約占年度溫室氣體排放總量68%，為最主要排放來源；其次為類別四「使用產品」之間接排放，平均約占30%，且有逐年增加趨勢。由同業公開資料比較可知，能源使用為隱形眼鏡製造業共同之主要排放來源，而供應鏈相關排放則因產品組成、原物料使用及營運模式不同而呈現差異。 進一步評估產品碳排放強度由A公司2021年度之1.095公噸 CO₂e／仟盒下降至2022年度之0.942公噸CO₂e／仟盒，顯示設備更新、能源管理及製程改善已提升單位產品生產效率，並具有一定減量成效。由於產量持續增加及營運規模擴大，總排放量改善幅度有限，顯示現行減碳措施雖已提升能源使用效率，但仍不足以完全抵銷營運成長所增加之排放量。 整體結果顯示，隱形眼鏡製造業之溫室氣體排放主要為能源使用，供應鏈相關排放的重要性逐漸提高。未來企業除持續推動節能改善外，亦應逐步提高再生能源使用比例，強化供應鏈碳資訊管理及低碳原物料應用，並將溫室氣體管理納入產能規劃及營運決策，以提升整體減量效益。;This study examines Company A， located in northern Taiwan， which is primarily engaged in the manufacture of soft contact lenses and the research and development of medical optical products. An organizational-level greenhouse gas (GHG) inventory was established in accordance with ISO 14064-1:2018 to analyze the company’s GHG emission profile， emission trends， and emission reduction performance from 2021 to 2024. Publicly disclosed information from peer companies was also reviewed to compare the GHG emission characteristics and carbon reduction practices of the contact lens manufacturing industry. The findings are intended to provide a reference for corporate GHG management and the development of carbon reduction strategies. The results show that Company A’s total GHG emissions increased from 55，059.3160 tCO₂e in 2021 to 61，990.3931 tCO₂e in 2023， representing an increase of approximately 12.6%， before decreasing to 59，576.7160 tCO₂e in 2024. These changes were closely associated with increased production output， production line expansion， and growth in the scale of operations. Plant 1， the company’s primary production site， consistently accounted for approximately 90% of the company’s total GHG emissions during the study period. In contrast， emissions from Plant 2 continued to increase due to the expansion of packaging lines and increased supply chain activities. Analysis of the emission profile indicates that Category 2， “Imported Energy，” was the largest source of emissions， accounting for an average of approximately 68% of annual total GHG emissions. Category 4， “Use of Products，” was the second-largest source， accounting for an average of approximately 30%， with its share showing an increasing trend over the study period. A comparison with publicly disclosed information from peer companies indicates that energy use is a major source of GHG emissions across the contact lens manufacturing industry， whereas supply chain-related emissions vary depending on product composition， raw material use， and operating models. Further analysis shows that Company A’s GHG emission intensity decreased from 1.095 tCO₂e per thousand boxes in 2021 to 0.942 tCO₂e per thousand boxes in 2024. This decrease indicates that equipment upgrades， energy management， and process improvements contributed to greater production efficiency and lower GHG emissions per unit of product. However， as production output and the scale of operations continued to increase， the reduction in total GHG emissions remained limited. These findings suggest that although the existing carbon reduction measures have improved energy efficiency， they have not been sufficient to fully offset the additional emissions associated with business growth. Overall， the results indicate that GHG emissions from the contact lens manufacturing industry are primarily associated with energy use， while the importance of supply chain-related emissions is gradually increasing. In addition to continuing to improve energy efficiency， companies should progressively increase the use of renewable energy， strengthen the management of supply chain carbon information， promote the use of low-carbon raw materials， and integrate GHG management into production capacity planning and operational decision-making to enhance overall emission reduction performance. <br> https://ir.lib.ncu.edu.tw/handle/987654321/111366 title: 電熱式 RTO 爐與天然氣 TO 爐對半導體製程之 VOCs 去除效率與碳排放強度探討;Comparative Analysis of VOCs Removal Efficiency and Carbon Emission Intensity between Electric-Heated RTO and Natural Gas-Fired TO Systems in Semiconductor Manufacturing abstract: 隨著淨零排放與能源轉型政策之推動，揮發性有機物（Volatile Organic Compounds， VOCs）防制系統之評估面向已由污染削減效能延伸至能源消耗、碳排放及操作成本。本研究以半導體實廠為案例，比較沸石轉輪分別結合電熱式蓄熱式氧化爐（Regenerative Thermal Oxidizer， RTO）與天然氣直燃式熱氧化爐（Thermal Oxidizer， TO）之 VOCs 防制效能、操作能源、碳排放及操作成本。 本研究採用 2024 年整年度實廠資料，VOCs 入口與出口濃度以每日監測值進行分析。因未取得入口端獨立流量與質量流率，年度入口量、去除量及質量去除效率係在假設入口與出口標準體積流量相同之前提下進行推估。RTO 主加熱器及兩套系統風機用電量採工程估算，TO 天然氣使用量則依流量計紀錄計算；操作能源碳排放範圍限於外購電力及天然氣燃燒所造成之操作階段排放。研究結果顯示，RTO 與 TO 推估年度 VOCs 去除量分別約為 109.3 與 109.6 t-VOC/yr，差異不大；每日濃度去除效率之平均值分別為 93.7% 與 97.5%，以 TO 較高。RTO 推估用電量約為 TO 的 1.6 倍，惟納入天然氣後，TO 之終端能源投入較高。依本研究採用之排放係數估算，RTO 與 TO 年度操作能源碳排放量分別約為 1，161.1 與 998.8 t-CO₂e/yr；在既定成本邊界下，TO 之年度操作成本及單位 VOCs 去除操作成本亦較低。綜合而言，兩套系統之 VOCs 去除量相近；TO 具有較高之濃度去除效率、較低之操作能源碳排放及操作成本，RTO 則具有較低之操作終端能源投入。本結果僅反映本案例、2024 年資料、工程估算參數及既定系統邊界，不宜直接外推為所有 RTO 與 TO 技術之一般性優劣。;With the implementation of net-zero emissions and energy transition policies， the evaluation of volatile organic compound (VOC) control systems has expanded beyond pollutant removal performance to include energy consumption， carbon emissions， and operating costs. This study used an actual semiconductor manufacturing facility as a case study to compare two integrated VOC control systems: a zeolite rotor coupled with an electrically heated regenerative thermal oxidizer (RTO) and a zeolite rotor coupled with a natural gas-fired thermal oxidizer (TO). The two systems were evaluated in terms of VOC removal performance， operating energy use， carbon emissions， and operating costs. Full-year operational data from 2024 were analyzed using daily monitoring data for VOC concentrations at the inlet and outlet. Because independent inlet flow rates and inlet VOC mass flow rates were unavailable， annual VOC inlet mass， VOC mass removed， and mass removal efficiency were estimated under the assumption of equal standard volumetric flow rates at the inlet and outlet. Electricity consumption by the RTO main heater and the fans of both systems was estimated using engineering calculations， whereas natural gas consumption by the TO was determined from flowmeter records. The carbon emission boundary was limited to operational emissions associated with purchased electricity and on-site natural gas combustion. The results showed that the estimated annual amounts of VOC removed by the RTO and TO systems were approximately 109.3 and 109.6 t-VOC/yr， respectively， indicating little difference between the two systems. The mean daily concentration removal efficiencies were 93.7% and 97.5%， respectively， with the TO achieving the higher efficiency. The estimated electricity consumption of the RTO was approximately 1.6 times that of the TO; however， when natural gas consumption was included， the TO had a higher total final energy input. Based on the emission factors adopted in this study， annual operational energy-related carbon emissions were estimated at approximately 1，161.1 t-CO₂e/yr for the RTO and 998.8 t-CO₂e/yr for the TO. Under the defined cost boundary， the TO also had lower annual operating costs and lower operating costs per unit of VOC removed. Overall， the two systems achieved similar annual VOC removal amounts. The TO had a higher concentration removal efficiency and lower operational energy-related carbon emissions and operating costs， whereas the RTO had a lower total final energy input. These findings are specific to the case conditions， 2024 operational data， engineering estimation parameters， and defined system boundaries and should not be generalized to all RTO and TO systems. <br> https://ir.lib.ncu.edu.tw/handle/987654321/111363 title: 空氣污染防制設備組合對柴油車排氣中粒狀污染物數目（PN）濃度影響之探討;Effects of Air Pollution Control Device Configurations on Particle Number (PN) Concentrations in Diesel Vehicle Exhaust abstract: 本研究以裝設柴油碳微粒過濾器（Diesel Particulate Filter， DPF）之柴油車為研究對象，蒐集國內粒狀污染物數目（Particle Number， PN）檢測資料，共納入6，352輛柴油車，分析不同排放期別、空氣污染防制設備組合及車輛特性對PN濃度之影響，並探討PN檢測應用於柴油車排放管理及DPF功能狀態評估之可行性。 研究結果顯示我國四期、五期及六期柴油車之PN濃度具有明顯差異，其中PN濃度中位數分別為3.08 × 10⁶、5.22 × 10⁵及1.40 × 10⁴ #/cm³，高排放比率則分別為70.42%、23.24%及3.65%。不同空氣污染防制設備組合之PN濃度、高排放車輛比率及低於檢出下限（Below Detection Limit， BDL）比率皆呈現顯著差異，其中配置DPF、選擇性觸媒還原系統（Selective Catalytic Reduction， SCR）及廢氣再循環系統（Exhaust Gas Recirculation， EGR）之車輛整體PN濃度及高排放比率均較低，顯示不同空氣污染防制設備組合具有不同之PN排放特性。進一步依排放期別比較後，各設備組合間之差異明顯縮小，顯示整體設備組合分析所呈現之PN排放差異可能部分受到各組排放期別組成差異影響，且不同排放期別車輛之PN排放特性仍呈現明顯差異。 所有受測車輛均已配置DPF，惟仍有19.88%車輛之PN濃度達1 × 10⁶ #/cm³以上；六期車輛雖多數PN濃度低於儀器檢出下限，亦有少數車輛超過高排放判定門檻。 本研究以1 × 10⁶ #/cm³作為高排放判定門檻，用以辨識PN濃度相對較高之車輛，並比較不同排放期別及車輛特性之高排放情形。研究結果亦顯示PN檢測可反映不同車輛之粒子排放特性，並可作為辨識異常高排放車輛及輔助評估DPF功能狀態之工具。整體而言，PN檢測除可作為柴油車排放管理工具外，亦可作為辨識異常高排放車輛及建立未來柴油車排放管理策略之重要參考。;This study investigated diesel vehicles equipped with diesel particulate filters (DPFs) using particle number (PN) measurement data collected in Taiwan. A total of 6，352 diesel vehicles were included to analyze the effects of different emission stages， combinations of air pollution control devices， and vehicle characteristics on PN concentrations， and to assess the feasibility of applying PN measurements to diesel vehicle emission management and the evaluation of DPF functional status. The results showed clear differences in PN concentrations among Taiwan Phase IV， Phase V， and Phase VI diesel vehicles. The median PN concentrations were 3.08 × 10⁶， 5.22 × 10⁵， and 1.40 × 10⁴ #/cm³， respectively， while the corresponding high-emission vehicle ratios were 70.42%， 23.24%， and 3.65%. Significant differences were also observed among different combinations of air pollution control devices in terms of PN concentrations， high-emission vehicle ratios， and below detection limit (BDL) ratios. Vehicles equipped with DPF， selective catalytic reduction (SCR)， and exhaust gas recirculation (EGR) systems exhibited lower overall PN concentrations and high-emission vehicle ratios， indicating that different combinations of air pollution control devices were associated with different PN emission characteristics. After further comparison by emission stage， the differences among device combinations were substantially reduced， suggesting that the differences observed in the overall analysis may have been partly influenced by differences in the emission-stage composition of each group， while PN emission characteristics remained clearly different among emission stages. Although all tested vehicles were equipped with DPFs， 19.88% of the vehicles had PN concentrations of 1 × 10⁶ #/cm³ or higher. Although most Phase VI vehicles had PN concentrations below the instrument detection limit， a small proportion still exceeded the high-emission threshold. In this study， 1 × 10⁶ #/cm³ was adopted as the high-emission threshold to identify vehicles with relatively high PN concentrations and to compare high-emission conditions among different emission stages and vehicle characteristics. The results also demonstrated that PN measurements can reflect differences in particle emission characteristics among vehicles and can serve as a tool for identifying abnormally high-emitting vehicles and assisting in the evaluation of DPF functional status. Overall， PN measurement can serve not only as a tool for diesel vehicle emission management but also as an important reference for identifying abnormally high-emitting vehicles and developing future diesel vehicle emission management strategies. <br>