使用相似性指標辨識萃智解答模型:以相關趨勢辨識為例

Using Similarity Measures to Identify TRIZ Model of Solutions: Examples of Relevant Trends Identification

邱聖家

指導教授 ： 許棟樑

國立清華大學/工學院/工業工程與工程管理學系/碩士(2013年)

https://doi.org/10.6843/NTHU.2013.00430

摘要

基於「類似問題有類似解答」之概念，問題若具有相似的特性，通常亦含有相似的解答特徵。因此，本研究將過去解題案例建構為案例資料庫，問題的求解即可依據問題特徵陣列之相似性比對，判斷與問題最相似的過去解題案例並將此案例解答模型作為問題的參考解答。
本研究使用相似性指標發展演化趨勢辨識系統，而於演化趨勢辨識過程中，首先為問題與趨勢階段屬性特徵之比對，如兩者比對相符則可再自此趨勢階段跳躍至具有問題所需功能的後續階段，並且該趨勢階段就能作為問題的觸發解。經由演化趨勢辨識軟體的建立，即可自動、快速地辨識問題相關趨勢，而不需再仰賴專家經驗或知識。最後本研究以124個案例與51個演化趨勢進行K-fold cross-validation驗證，測試此數理方法的有效性，而其結果證明本研究的方法所推薦前10名趨勢觸發解可100%命中案例原本解答，遠超過隨機挑選10個趨勢觸發解的9%案例解答命中率。
本研究的貢獻包含：1）以數理方式發展客觀且快速辨識解答模型之方法，開拓一個萃智科學研究之方向; 2）配合電腦化工具，建立一個電腦輔助的解題工具，經由過去案例把專家經驗引入系統中，提供解題者一個快速、客觀且穩定之電腦化系統。

關鍵字

功能屬性分析 ； 技術演化趨勢 ； 相似性比對 ； 萃智 ； 電腦輔助創新解題

並列摘要

Based on the concept of “Like Problem, like solutions” in TRIZ theory, the problems with similar problem characteristics are likely to have similar solution characteristics. With a set of known solved problems and their corresponding solutions as a casebase, solving a problem becomes a matter of identifying highly similar known problems in terms of Problem Characteristic Array similarity and integrating the Solution Arrays of the corresponding similar problems to form the set of solution models.

In this research, relevant trends identification system using similarity measures were developed. In identifying relevant trend solutions, characteristic attributes of the problem are compared against the characteristic attributes of certain earlier stage of a trend first. If they match, the ensuing stages of the same trend can imply model of solutions as jumping into that stage can provide functions needed to solve the problem. By encoding the 'knowledge' embedded in the trends, a piece of software is written to identify the relevant trends for problem solving quickly and objectively without needing to rely on expert experience and knowledge. K-fold validity verification was used to verify the effectiveness of this method. With 124 known cases and 51 trend examples, the results showed that the solutions recommended by the 10 most likely trends achieved 100% coverage of problem known solutions and is significantly better than randomly selected 10 solutions which covered less than 9% of the known solutions.

The contributions of this research include: 1) Opening up a new branch of TRIZ research using mathematical methods to objectively identify model of solutions. 2) Establishing a computer aided problem-solving tool that can automatically and quickly identify the relevant trends for problem solving.

並列關鍵字

Computer-aided Problem Solving ； Function and Attribute Analysis, ； Similarity ； TRIZ ； Technical Trends ； Trends of Engineering Systems Evolution

參考文獻

4.李子平(2008)，電腦輔助質場分析解題助手，國立清華大學，工業工程與工程管理系碩士論文。

5.沈穎廷(2009)，運用田口法、模糊理論以及TRIZ於產品之重新設計，國立臺灣大學，工學院機械工程學系碩士論文。

16.黃健錡(2011)，提升音樂允許容錯比對能力之研究，朝陽科技大學，資訊管理學系碩士論文。

20.蔡美慧(2010)，以偏好導向為相似性量測的協同過濾式推薦，朝陽科技大學，資訊管理學系碩士論文。

19.廖坤祈(2011)，整合TRIZ與品質機能展開法評估產品未來演化趨勢，國立成功大學，機械工程學系碩士論文。