個人電腦的中央處理器基座(CPU Socket)端子是一個小且精密的元件,在正向力及永久位移二項都有個嚴格的設計規範。若是設計不當,將導致過大的正向力或永久位移量,無法符合嚴格的設計規範要求。本論文結合類神經網路及基因演算法,先進行實驗規劃,然後使用商用軟體ABAQUS進行分析,最後利用類神經網路(Artificial Neural Network, ANN)建構CPU Socket 內部端子之正向力及永久位移量與幾何形狀之間關係的數學預測模式。接著利用數學預測模式及基因演算法進行最佳化搜尋,尋找最小永久位移量及適當的正向力大小的最佳幾何設計。將最佳設計製作成實物,並利用自動插拔力機進行實際測試,驗證正向力及位移量皆能達到設計規範的要求。本論文所使用的方法能快速及精準地找出此端子之最佳設計,降低產品開發的錯誤率,加速生產流程,大幅減低生產成本。
The CPU socket is a small and precision component. There are high requirements of normal contact force and permanent deformation in design specification. Improper design of geometry of socket pin will create large permanent deformation or unsuitable normal force and it will not fit the design requirements. Neural network and genetic algorithms were integrated for geometry optimization of socket pin. Design an experiment for geometry of socket pin. The commercial program, ABAQUS, is used for analysis. A back-propagation neural network(BP) is used to build two predicting models for normal contact force and permanent deformation of CPU Socket pin, respectively. Then the models are applied in genetic algorithms for searching optimum geometry for socket pin with minimum permanent deformation and appropriate normal contact force. The framework of this study can help the designer find the optimum geometry of socket pin fitting design requirements accurately and quickly. Beside it also can increase the productivity and reduce unnecessary errors and cost effectively in product development.