頁籤選單縮合
題名 | Prediction of Laser Cutting Qualities for QFN Strips by Using Levenberg-Marquardt-based Neural Network=應用LM學習法與神經網路於QFN封裝之雷射切割品質預測 |
---|---|
作者姓名(中文) | 蔡明忠; 李振豪; | 書刊名 | 中國機械工程學刊 |
卷期 | 31:4 2010.08[民99.08] |
頁次 | 頁273-280 |
分類號 | 446.893 |
關鍵詞 | 雷射切割; QFN封裝; Quad flat non-lead strip; QFN strip; Laser cutting; Back-propagation neural network; BPNN; Levenberg-marquardt; LM; |
語文 | 英文(English) |
中文摘要 | 本文提出以Levenberg-Marquardt (LM)學習法為基礎之倒傳遞神經網路來建構QFN封裝之雷射切割品質預測模型,主要以3個雷射切割參數為輸入參數及6個切割品質為輸出參數,來建立QFN封裝之雷射切割品質預測模型。在訓練部分,使用27組的不同切割參數組合之實驗數據來進行多組不同神經元數目等參數之模型訓練,並找出訓練誤差最低之最好模型作為預測用模型,實驗結果顯示最好模型之平均訓練誤差為0.51%;在預測應用部分,經選取10組以上進行驗證,得到平均誤差為1.665%。經上述的實際驗證,由本文所建立的預測模型,確實可有效、精準地預測出QFN封裝之雷射切割品質。 |
英文摘要 | In this paper, a back-propagation neural network (BPNN) with Levenberg-Marquardt (LM) algorithm is proposed to build a predicting model of laser cutting quality characteristics for QFN (Quad Flat Non-lead) strips. The inputs of LM-BPNN model include three main laser parameters and its outputs include six cutting quality characteristic. Several BPNN models with different configuration such as the number of neurons were trained with 27 set of experimental data. Then, a best LM-BPNN model with least training error was obtained and then examined with over 10 sets of experimental data. From the experimental results, the average training error and predicting error by using the best LM-BPNN model are 0.51% and 1.665%, respectively. The results show that the proposed LM-BPNN model has the reliable predicting ability to estimate six laser cutting quality characteristics of QFN strips. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。