目前微機電系統在半導體產業所占的地位已經愈來愈重要,這些產品在各項電子產品的終端應用上,佔有一席之地。因此,提高產出、降低生產成本就成為各家設計公司以及生產工廠進入微機電系統市場最重要的課題。本研究利用5百萬筆的原始測試數據進行比較與驗證分析,再透過機器學習(ML)中隨機森林法來調控測試機台參數並達到預期最佳UPH(Units per hour)並藉由streamlit智慧介面同步顯示,優化生產機台參數的最佳設定值以及UPH的等級。結果顯示隨機森林演算法所預測的震動盤的圓震及直震的最佳震動頻率,可讓每小時產品測試數量(UPH)有效從約86顆提升至111顆。
At present, the position of MEMS in the semiconductor industry has become more and more important, and these products have a place in the terminal applications of various electronic products. Therefore, increasing output and reducing production costs have become the most important issues for design companies and production plants to enter the MEMS market. This research uses 5 million original test data for comparison and verification analysis, and then uses the random forest method in machine learning (ML) to adjust the parameters of the test machine and achieve the expected best UPH (Units per hour) and rely on the wisdom of streamlit. The interface displays synchronously to optimize the best setting value of the production machine parameters and the level of UPH. The results show that the optimal vibration frequency of the circular and direct vibrations of the vibration plate predicted by the random forest algorithm can effectively increase the number of product tests per hour (UPH) from about 86 to 111.