近年伺服器機殼隨著GPU算力與模組重量大幅提升，底板需承受20-25kg 以上集中載重，傳統多以增加板厚或改用高階材料來提高剛性，卻伴隨材料用量、碳排與成本的上升。為回應企業ESG與循環經濟需求，本研究以環保回收鍍鋅鋼板燁輝 RC95為研究材料，探討在不增加板厚與材料等級之前提下，是否能透過深抽幾何設計，使1.2mm回收鍍鋅鋼板滿足高剛性GPU伺服器機殼底板之荷重需求。 本研究依實際GPU底板尺寸與支撐邊界條件，在ANSYS Workbench中建立有限元素模型，以25.45kg等效載重與最大垂直變形量小於2.0mm為設計要求，評估不同深抽幾何配置下之最大變形量。方法上採用田口法 L₁₆(4⁴) 直交表規劃，抽引角度、抽引方向、筋條形狀及筋條組數四項變化因子，並假設四項因子並不會相互影響下，先完成16組基礎模擬，並透過主效應、信噪比與變異數分析辨識主要控制因子；再針對高潛力區域進行16組局部探討，並額外進行邊界效益的評估，篩選出最適合的幾何組合。其後製作深抽試樣，並以U envelope系統量測未載重與載重後之底板撓度分布，以驗證有限元素模型之準確性。 分析結果顯示，筋條形狀為影響最大變形量之主導因子，其次為抽引方向，筋條組數與抽引角度則扮演次要調整角色。在本研究條件下，以70°抽引角度、向下且延伸至兩側側牆之抽引方向、一字形筋條與9組筋條配置之組合表現最合適，在理想化雙固定平面邊界條件下，可將最大變形量由3.655mm 降至1.089mm；若改採較貼近實務之16軸支撐邊界條件，對應之模擬最大變形量約為1.409mm，整體改善約達60–70%，並滿足2.0mm最大垂直變形量限制。實體試樣量測結果顯示，載重後最大下撓量約為1.47mm，且變形分布與有限元素分析法法趨勢一致，證明本研究所建立之有限元素分析法法結合田口法與實驗驗證流程，能有效用於回收鍍鋅鋼板深抽幾何設計之前期評估，亦顯示1.2mm RC95回收鍍鋅鋼板具備應用於高載重GPU伺服器機殼底板之工程可行性;This study addresses the structural challenge that modern GPU server chassis base plates must sustain concentrated loads of 20–25 kg or more as GPU computing power and module weight continue to increase， while conventional reinforcement strategies rely on thicker sheets or higher‑grade materials， which in turn raise material usage， embodied carbon， and cost. To respond to corporate ESG and circular‑economy requirements， this work employs eco‑friendly recycled galvanized steel Yieh Phui RC95 as the base material and investigates whether appropriate deep drawing geometry design can enable a 1.2 mm recycled galvanized steel sheet to meet the load‑bearing requirements of a high‑stiffness GPU server chassis base plate without increasing sheet thickness or material grade. A finite element model was established in ANSYS Workbench based on the actual dimensions and support boundary conditions of the GPU base plate， with an equivalent load of 25.45 kg and a design requirement that the maximum vertical deflection be less than 2.0 mm. A Taguchi L₁₆(4⁴) orthogonal array was adopted to plan four varying factors—drawing angle， drawing direction， rib shape， and rib quantity—under the assumption that there is no interaction among factors， leading to 16 baseline simulations. Main‑effect plots， signal‑to‑noise ratios， and analysis of variance were used to identify the dominant control factors， after which 16 additional local exploration cases were conducted in the high‑potential region， along with an assessment of the marginal effect of increasing rib quantity， to determine the most suitable geometric configuration. A deep‑drawn specimen was then fabricated according to the selected design， and the U‑envelope measurement system was used to obtain full‑field surface deflection data before and after loading， thereby validating the accuracy of the finite element model. Analysis results show that rib shape is the dominant factor affecting the maximum deformation， followed by drawing direction， while rib quantity and drawing angle play secondary tuning roles. Under the conditions considered in this study， the combination of a 70° drawing angle， a downward drawing direction extended to both side walls， straight‑line ribs， and 9 rib sets exhibits the best performance; with idealized dual fixed‑plane boundary conditions， this configuration reduces the maximum deformation from 3.655 mm to 1.089 mm， whereas under more realistic 16‑pin support boundary conditions， the corresponding maximum deformation is approximately 1.409 mm， yielding an overall improvement of about 60–70% while satisfying the 2.0 mm maximum vertical deflection limit. Experimental measurements on the deep‑drawn RC95 specimen show that the maximum downward deflection under the 25.45 kg equivalent load is about 1.47 mm， and the deformation pattern is consistent with the finite element predictions， confirming that the proposed “finite element analysis combined with Taguchi method and experimental validation” framework is effective for the preliminary design of deep drawing geometries using recycled galvanized steel sheets， and demonstrating that 1.2 mm RC95 recycled galvanized steel is technically feasible for high‑load GPU server chassis base plate applications.
