- 學位論文
以本體論為基礎進行模糊推論流程之規劃與執行
Ontology-Based Planning and Execution of Fuzzy Inference Flow
摘要
在21世紀知識經濟體系下，知識被看待為創新、進步以及競爭的重大力量。此環境下企業面臨的問題在於如何整合散落知識，並發展一系統模擬人類智慧給予決策，使系統能夠應用於各個組織，提升系統價值與減少企業資源浪費。因此一個以專家系統為背景的架構被提出，稱做以本體論為基礎之模糊推論(Ontology-Based Fuzzy Inference, OFI)系統，也就是第一代OFI系統。 本研究提出第二代OFI系統架構，取模糊規則認知圖(Rule Based Fuzzy Cognitive Maps, RBFCM)作為推論方法，並發展動態多重模糊推論滿足不同起始資訊情況的半導體錯誤診斷問題。第二代OFI系統以第一代OFI系統為發展基礎，致力於多種領域知識建模，檢視原始系統物件對於動態多重模糊推論方法的支援性，有效提昇系統彈性與應用層面。 第二代OFI系統主要分作兩大部份，整合本體論模型與可組建系統物件。利用網路本體語言(Web Ontology Language, OWL)建構與整合產業領域、模糊理論以及推論方法之知識，使知識具備分享與再利用特性，供各組織加以運用，故稱作整合本體論模型；可組建系統物件是利用Java程式語言開發之推論引擎，主要特色在於操控知識庫中推論方法知識，讓使用者針對不同問題選擇適合之推論方法，或者自行組裝運用。在整合本體論模型部份，延續第一代OFI系統，稍微更動知識概念；可組建系統物件是本研究改革重點，將焦點放於各個物件之彈性檢視，並希望系統物件可支援動態多重模糊推論。 因此提出三項問題：(1) 系統功能物件無法重複使用：由於設計方式不佳，使得相同計算邏輯的概念無法以一物件涵蓋，無形中造成資源的浪費。(2) 系統資料物件生成方式缺乏彈性：資料本體論與資料物件無法同步更新，一旦資料本體論中有所變動，資料物件也必須更改，造成資訊承接錯誤或是引發系統缺乏彈性問題。 (3)無法動態規劃多次模糊推論與執行：第一代OFI系統也許可透過設定流程知識完成RBFCM推論，但只要因素類型改變或因素相互關係改變，推論流程知識也必須重新人為規劃，造成時間上的耗費。 針對上述問題提出三項設計來進行改善，提昇系統彈性並可執行不同情況下的RBFCM推論，三項設計分別為(1) 彈性功能物件：相同概念之功能只用一個功能物件與之對應，減少重複概念的功能物件，提昇系統穩定與彈性程度。(2)通用資料物件：利用Vector的二維陣列作為資料物件基本型態，讓資料本體論與資料物件同步更新，提昇系統彈性化程度。 (3)多重OFI流程規劃者：使系統可自動規劃並執行多次模糊推論流程，並且當評估因素數量或因素關係改變時，系統依然可以自動執行，不需要重新設定推論流程。 落實上述三種設計並整合系統後，將其套用在五種情境模式：兩因素類型-單一路徑、三因素類型-單一路徑、三因素類型-多路徑、因素類型數目對時間的影響以及單一類型因素數目對時間的影響。利用前三大情境驗證系統的可行性與可擴充性；得知當擴充因素或路徑後，仍然能夠得到正確的規劃與執行結果，證實系統的確能支援動態多重模糊推論的規劃與執行，並且其擴充能力也是足夠的。藉由後兩大情境檢視因素類型數目與單一類型因素數目對系統執行時間的影響，得知系統執行時間的多寡是取決於因素數目；。綜合上述，企業能夠將此系統運用於各個組織，以分享與再利用特性減少資源浪費，降低系統開發成本。
並列摘要
“Knowledge” is the main driving force for innovation, advancement, and competitiveness in the 21st century knowledge-based economy system. The challenges of which the industry will have to face in this economy system include, integrating scattered know-how, developing a simulated AI system for decision making, applying the system to individual organizations, raising the value of the system, and limiting a waste of corporate resources. As a result, a system with an expert system as the backbone infrastructure was proposed, known as Ontology-Based Fuzzy Inference (OFI) system, which is also the first generation OFI system. This thesis proposes a second generation OFI system infrastructure using the Rule Based Fuzzy Cognitive Maps (RBFCM) as the method of inference, and further expands to include multiple dynamic fuzzy inferences for the different initial conditions of incorrectly diagnosed semiconductor failures. The second generation OFI system is formulated around the foundation built by the first generation OFI system and focuses on constructing the knowledge base module for different application fields. This research also inspects the amount of support on the methods of multiple dynamic fuzzy inferences provided by the original system objects, for effective improvements on system application and flexibility aspects. The development of second generation OFI system is two-fold: Integrated OFI Ontology and Configurable System Object. The purpose of Integrated OFI Ontology is for knowledge sharing and reuse, which is enabled by using Web Ontology Language (OWL) to construct and integrate the knowledge from different fields of the industry, fuzzy theories, and inference methods. Configurable System Object is for the manipulation of various inference methods from the knowledge database. Users will be able to select the most appropriate inference method for different problems, or they can fabricate their own solution. This research aims at designing formal specifications of Integrated OFI Ontology and developing Configurable System Object to support multiple dynamic fuzzy inferences. Three main issues of the OFI system are first addressed: (1) System Function Object is not reusable: improper design leading to identical logical concepts that cannot be covered by a single object, resulting in the waste of system resources; (2) System Data Object is not flexible during declaration: Data Ontology and Data Object cannot be updated in sync, once data ontology has been modified, the data object will also have to be changed, thereby resulting in information disconnect error or issues with system flexibility; (3) Unable to support the dynamical planning and execution of multiple OFI instances: even though the first generation OFI system might be able to complete the RBFCM inference through configuration of the process knowledge base, changes to the element type or dependencies will require manual re-planning of the inference process knowledge, of which the entire process is time consuming. Three design improvements are proposed to counteract the problems pointed out above and to improve system flexibility on the execution of the RBFCM inference based on different conditions. The three designs are (1) Flexible Function Object: functions with identical concepts correspond to a single function object, thereby reducing function objects of duplicate concepts, improving system reliability and flexibility; (2) Generic Data Object: a 2D matrix using vectors is defined as the generic form of data objects, where Data Ontology and Data Object may be updated in sync thereby improve system flexibility; (3) Multiple OFI Flow Planner: the system will be able to automatically plan and execute multiple fuzzy inferences, as well as evaluate the changes in number or dependencies of factors without the need to reconfigure the inference process flow. Implementing the three design changes mentioned above to the system and applying to the following five scenarios: two factor types – single path, three factor types – single path, three factor types – multiple path, influence of quantity of factor types on time, and influence of quantity of single factor types on time. The former three scenarios may be used to verify the practicability and expandability of the system. If the correct planning and execution results may be expected after the expansion of factors or paths, then the system is proved to be capable of supporting the planning and execution of multiple dynamic fuzzy inferences, as well as proving the expandability of the system. The latter two scenarios may be used to inspect the influence of quantity of factor types and single factor types on system execution time, where system execution time will see a rise with an increase of factors. In summary, enterprises will be able to adopt this system to each of their individual operating organizations for sharing and reusing of the knowledge database. Valuable corporate resources will no longer be wasted and system development costs can also be reduced.