DSpace community: 工業管理研究所碩士在職專班

結合檢索強化生成技術之智慧問答系統研究 -以D公司內部知識應用為例

title: 結合檢索強化生成技術之智慧問答系統研究 -以D公司內部知識應用為例 abstract: 近年來，隨著人工智慧技術的快速發展，大型語言模型（LLMs）在自然語言理解與生成方面展現出卓越能力，但其生成內容仍可能出現「幻覺」（hallucination）現象，導致資訊不準確或不一致，限制其在企業知識管理與決策支援中的實用性。本研究以 D 公司內部知識應用為背景，探討檢索強化生成（Retrieval-Augmented Generation，RAG）技術在智慧問答系統中的應用，並分析企業內部資訊特性與查詢模式，以作為 RAG 技術應用的理論基礎。研究重點在於探討 RAG 技術如何結合語言模型與知識檢索機制，改善資訊落差、提升語意理解能力，並提供即時且可靠的回應。透過理論分析與模擬查詢情境的評估，本研究說明 RAG 技術在提升查詢準確率、改善使用者信任及優化知識管理效率上的潛力，為企業導入 AI 智慧問答系統提供可行的理論與實務參考。關鍵詞：檢索強化生成、語言模型、智慧問答系統、向量資料庫、企業知識管理、D 公司內部知識 ;In recent years，with the rapid development of artificial intelligence，large language models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However，these models may produce “hallucinations,” resulting in inaccurate or inconsistent information，which limits their applicability in enterprise knowledge management and decision support. This study focuses on internal knowledge applications within D Company，investigating the use of Retrieval-Augmented Generation (RAG) technology in intelligent question-answering systems and analyzing the characteristics of enterprise knowledge and query patterns to provide a theoretical foundation for RAG application. The research emphasizes how RAG technology integrates language models with knowledge retrieval mechanisms to reduce information gaps，enhance semantic understanding，and deliver timely and reliable responses. Through theoretical analysis and evaluation in simulated query scenarios，this study demonstrates the potential of RAG technology to improve query accuracy，strengthen user trust，and optimize knowledge management efficiency，offering practical and theoretical guidance for enterprises adopting AI-based question-answering systems. Keywords: Retrieval-Augmented Generation，Language Model，Question-Answering System，Vector Database，Enterprise Knowledge Management，D Company Internal Knowledge

基於機器學習與集成方法預測半自動化生產線直接人員工時需求;Forecasting Direct Labor Hour Demand in Semi-Automated Production Lines via a Two-Stage Ensemble Learning Approach

title: 基於機器學習與集成方法預測半自動化生產線直接人員工時需求;Forecasting Direct Labor Hour Demand in Semi-Automated Production Lines via a Two-Stage Ensemble Learning Approach abstract: 在半自動化生產環境中，直接人員工時需求受到產能變動、設備狀態與人力管理因素影響，若僅依賴經驗法則或歷史平均進行規劃，容易造成工時預測偏誤，進而影響生產效率與營運穩定性。此外，實務資料往往來自不同管理或作業層級，其資料生成頻率與分析單位不一致，使工時需求預測在模型建構與結果穩定性上面臨挑戰，特別是在可用樣本有限的情境下。本研究以某LED封裝製造公司的半自動化生產線為研究對象，整合生產、人力與設備等多來源資料，提出一套適用於資料顆粒度不一致情境的人員工時需求預測架構。研究採比較式設計，先建構單層模型作為基準，再提出分層模型，其中第一層以站別層級資料建立線性迴歸模型，第二層則以月度層級管理變數校正預測殘差。為因應小樣本特性，本研究結合主成分分析法與自助法檢驗模型穩定性，並以判定係數與均方根誤差作為評估指標。實證結果顯示，分層模型在預測精準度與穩定性上皆顯著優於單層模型，且在多次重抽樣下能維持一致表現。此外，透過移動平均法推估未來輸入變數，使模型具備跨月度預測能力。在模型解讀方面，敏感度分析結果顯示站別結構差異為影響工時需求的主要因素，而設備叫修次數在連續型變數中具有顯著影響，並可辨識具有管理意義的轉折區間。整體而言，本研究所提出的分層建模策略，能有效處理資料顆粒度不一致與小樣本問題，並提供具實務價值的人力規劃與改善決策依據。;In semi-automated manufacturing environments, direct labor hour demand is influenced by production variability, equipment conditions, and workforce management factors. When planning relies solely on experiential rules or historical averages, biased labor hour forecasts are likely to occur, thereby affecting production efficiency and operational stability. Moreover, practical manufacturing data are often generated across different managerial and operational levels, resulting in heterogeneous data granularity and inconsistent analytical units. These characteristics pose additional challenges to labor hour demand forecasting, particularly under limited sample conditions. This study investigates a semi-automated production line of an LED packaging manufacturer and integrates multi-source production, manpower, and equipment data to develop a labor hour demand forecasting framework suitable for heterogeneous data granularity. A comparative research design is adopted by first constructing a single-layer model as a baseline, followed by a proposed two-stage ensemble learning framework. In the proposed framework, the first stage employs station-level data to build a linear regression model that captures operational differences among workstations, while the second stage utilizes month-level managerial variables to correct the prediction residuals. To address the small-sample characteristic, principal component analysis and bootstrapping are employed to evaluate model stability, with model performance assessed using the coefficient of determination and root mean squared error. Empirical results indicate that the proposed two-stage ensemble model significantly outperforms the single-layer baseline model in terms of both predictive accuracy and stability, and maintains consistent performance across repeated resampling. In addition, a moving average approach is applied to estimate future input variables, enabling multi-period labor hour forecasting. From a model interpretation perspective, sensitivity analysis reveals that structural differences among workstations are the primary drivers of labor hour demand, while machine repair frequency exhibits a significant impact among continuous variables and helps identify management-relevant turning point ranges. Overall, the proposed two-stage ensemble learning approach, which follows a hierarchical modeling rationale, effectively addresses heterogeneous data granularity and small-sample challenges, and provides practical insights for workforce planning and decision-making for operational improvement.

零組件發料流程之影響生產總工時研究 -以L公司為例

title: 零組件發料流程之影響生產總工時研究 -以L公司為例 abstract: 本研究旨在探討半導體設備製造中，零組件發料流程對生產總工時之影響。由於蝕刻設備具備少量多樣、工序複雜與高度相依特性，物料發放安排若未依工序邏輯進行，將導致搬運、拆裝、等待等非增值作業增加，使生產工時易產生波動並造成產能利用率下降。隨著半導體設備組裝難度提升與產線規模持續擴大，如何透過流程設計優化發料順序，以提升工時穩定性與縮短組裝工時，已成為提升競爭力的重要課題。在此背景下，本研究以L公司蝕刻設備產線為研究對象，針對影響工時的關鍵流程環節進行分析，包括大型件發放時序、外觀件多次入線造成之重複拆裝，以及主、副生產線作業不同步等問題。本研究結合2024–2025年之實際工時資料、AHS（Activity Hierarchy Structure）活動拆解、現場Gemba Walk觀察，以及半結構式訪談等方法，辨識生產流程中的非增值活動來源，並據以提出三項改善方案，分別聚焦於大型件群組化發料、外觀件批次整合，以及主副線作業同步化。研究結果顯示，流程改善對生產工時具有正面效益。主生產線工時自改善前之57.6小時逐步下降至方案C之53.5小時；副生產線因外觀件批次化策略使重工與搬運行為減少，工時變異亦顯著改善；測試線則因上游流程穩定度提升而呈現idle time減少。此外，整體Cycle Time亦由原先約27天縮短至26.57天，顯示發料策略與工序配置之調整能有效提升生產節奏一致性與流程可預測性。本研究結果除可作為L公司後續跨產品線改善參考外，亦對半導體設備製造中物料管理、流程設計與工時控制提供具體管理意涵。研究亦顯示，透過資料驅動的流程診斷與ECRS改善原則之應用，可有效消除非增值活動，減少生產工時並強化製造系統穩定度。 ;This study investigates the impact of component material issuance processes on the total production hours in semiconductor equipment manufacturing. Due to the characteristics of etch equipment—high product mix, complex assembly sequences, and strong interdependence across workstations—improperly arranged material issuance timing may lead to non-value-added activities such as repeated handling, reassembly, and waiting. These inefficiencies in turn increase variability in production hours, reduce capacity utilization, and undermine the stability of delivery performance. As equipment complexity continues to rise and global manufacturing sites expand, optimizing material issuance logic to enhance production stability has become a critical element of operational competitiveness. Using the etch equipment assembly lines of Company L as the research case, this study analyzes key process bottlenecks affecting production hours, including the timing of large-component issuance, repeated re-entry of appearance parts, and asynchronous operations between the main and sub assembly lines. Multiple data sources and analytical approaches were employed, including 2024–2025 production-hour datasets, Activity Hierarchy Structure (AHS) analysis, on-site Gemba Walk observations, and semi-structured interviews with production personnel. Based on the identified bottlenecks, three improvement strategies were proposed: grouped issuance for large components, batch consolidation for appearance parts, and synchronization of main and sub assembly lines. The empirical results demonstrate that all three improvement strategies contribute positively to reducing production hours and stabilizing process performance. Main-line production hours decreased from 57.6 hours before improvement to 53.5 hours under Strategy C. The consolidation of appearance parts significantly reduced rework and unnecessary handling on the sub line, thereby decreasing variability. The test line also benefited from reduced idle time due to more stable upstream processing. The overall Cycle Time was shortened from approximately 27 days to 26.57 days, indicating that better material issuance strategy and process coordination can enhance flow consistency and improve system predictability. The findings of this study provide actionable insights for Company L to extend improvements across multiple product families. More broadly, the results contribute to understanding how material management, process design, and work-hour control can be optimized in semiconductor equipment manufacturing. The study highlights that data-driven diagnosis combined with ECRS principles can effectively eliminate non-value-added activities, improve production efficiency, and enhance the stability of manufacturing systems.

運用 ECRS 原則與動線分析於產線空間效率之改善研究

title: 運用 ECRS 原則與動線分析於產線空間效率之改善研究 abstract: 近年因全球生產基地移轉與廠房租金上升，企業面臨產線空間縮減的挑戰，如何在有限空間內維持原有產能與物流流暢度，成為製造業重要課題。本研究以某電子設備組裝產線為對象，針對其現行作業流程中存在的動線冗長、物料配置不佳與作業順序未優化等問題，採用 ECRS（Eliminate、Combine、Rearrange、Simplify）原則及動線分析進行改善研究。研究首先蒐集產線現況資料，包含人員作業步驟、物料搬運動線與工位佈置等資訊，並透過人、機、料、法、環架構診斷產線瓶頸。接著以 ECRS 原則重新檢視作業流程，透過刪除不必要動作、整併重複步驟、調整作業順序與簡化操作方式，使流程更符合實際工作動線。此外，本研究以動線分析比對改善前後人員移動距離，並重新規劃物料補給策略，由原先一次性搬入大量物料，改為依作業階段分批進料，以降低空間占用並提升物料可視性。透過改善後的流程與佈置，本研究在不增加設備投資的前提下，有效減少人員移動距離，降低物料存放空間需求，同時提升工位間銜接效率。實證結果顯示，改善方案可縮短整體作業時間並維持產能水準，證明 ECRS 原則結合動線分析可作為提升產線空間效率的有效方法。本研究成果可提供廠房空間受限或正面臨產線縮減需求的製造企業做為佈置規劃與流程改善之參考。;In recent years, global production relocation and rising facility costs have forced manufacturers to operate within reduced factory space. Maintaining production capacity and efficient material flow under such constraints has become a critical challenge. This study examines an electronics assembly line with inefficiencies caused by lengthy operator movement,suboptimal material placement, and non-ideal task sequencing. To address these issues, the ECRS principles（Eliminate, Combine, Rearrange, Simplify）and flow analysis are applied toredesign the workflow and improve space utilization. The research begins with on-site data collection, including operator tasks, material handling routes, and workstation layout. Using the Man–Machine–Material–Environment–Method framework, bottlenecks are identified. ECRS principles are then used to streamline operations by removing unnecessary actions, consolidating repetitive tasks, reorganizing process sequences, and simplifying procedures to match natural movement. A revised material supply strategy is also introduced: materials are delivered in staged batches rather than all at once, reducing space usage and improving material visibility. The improvements reduce operator travel distance, lower storage requirements, and enhance workstation connectivity without additional equipment investment. Results show shorter overall processing time while maintaining production capacity, confirming that combining ECRS with flow analysis is an effective method for improving space efficiency. These findings offer practical reference for manufacturers facing space limitations or seeking to optimize layout and workflow.