6. ConclusionOn the methodological aspect, the case study established thatthe statistical models usually encountered in the literature related to statistical cost modelling—namely MLR and ANNs—are super- seded in terms of performance by more recent techniques from the fields of Data Mining and Machine Learning, notably Support Vector Regression and Gradient Boosted Trees. It also appears that the various statistical techniques yield complementary perspec- tives on the cost data and should thus be used concurrently. Moreover, ensemble methods may be a worthwhile solution to optimally average the cost estimates from several models. Theanalysis also generated valuable engineering insights. First of all,MLR showed that the cost of material accounts for a small portion of total manufacturing cost. Moreover, the bigger the part, the higher the cost of material represents as a percentage of total manufacturing cost. Semiparametric GAM and assessment of variable influence in GBTs demonstrated that the manufacturingcost seems to follow a logarithm function α − β log(Q + γ) of the accumulated production volume with α being related to the set-up and equipment preparation operations, β as a scaling or damping effect related to production volume and γ related to the initialrunning cost of the equipment. Finally, Gradient Boosted Trees indicate that the accumulated production volume exerts the big- gest influence on manufacturing cost, followed by span, machin- ability, chord and finally the part category.Some perspectives and future work can be identified after this initial investigation. First of all, it might be worthwhile to apply the same statistical analyses to other jet engine parts, or even to the whole engine, to verify whether the conclusions remain valid in other situations. It might also be particularly interesting to compare the results of the statistical models to the manufacturing costs estimated by cost engineers manually, based on their ex- perience and different methods (analytical and analogous cost models). On the medium term, the cost estimations could be part of a decision system used by production engineers and pro- gramme managers to identify outlier components in term of unit cost and launch cost reduction initiatives; a cost prediction system could also be used by supply chain experts to identify the best price amongst manufacturers. On the longer term, the statistical estimation of manufacturing cost could be generalized during the design phase, embedded in CAD/CAM tools and eventually become an optimization parameter used during parts' and products' design by the mechanical engineers. It would be a further step towards an integration to more holistic approach of aeroengine lifecycle suchas Integrated Computational Materials Engineering (ICME).
6. Conclusion<br><br>On the methodological aspect, the case study established that<br>the statistical models usually encountered in the literature related to statistical cost <br>modelling—namely MLR and ANNs—are super- seded in terms of performance by more recent techniques <br>from the fields of Data Mining and Machine Learning, notably Support Vector Regression and Gradient <br>Boosted Trees. It also appears that the various statistical techniques yield complementary perspec- <br>tives on the cost data and should thus be used concurrently. Moreover, ensemble methods may be a <br>worthwhile solution to optimally average the cost estimates from several models. The<br><br>analysis also generated valuable engineering insights. First of all,<br>MLR showed that the cost of material accounts for a small portion of total manufacturing cost. <br>Moreover, the bigger the part, the higher the cost of material represents as a percentage of total <br>manufacturing cost. Semiparametric GAM and assessment of variable influence in GBTs demonstrated <br>that the manufacturing<br>cost seems to follow a logarithm function α − β log(Q + γ) of the accumulated production volume <br>with α being related to the set-up and equipment preparation operations, β as a scaling or damping <br>effect related to production volume and γ related to the initial<br>running cost of the equipment. Finally, Gradient Boosted Trees indicate that the accumulated <br>生產量的施加大端GEST在FL uence上製造成本,隨後跨度,machin- <br>能力,和弦和FI應受部件大類。<br>一些觀點和今後的工作可以在此初步調查後identi網絡版。首先,<br>它可能是值得的相同的統計分析應用到其他噴氣發動機部件,甚至<br>整個引擎,驗證結論是否在其他情況下仍然有效。這也可能<br>是特別有趣的統計模型的結果進行比較,以製造業<br>由手工造價工程師估算成本的基礎上,他們的EX- perience和不同的方法<br>(分析和類似的成本模型)。在中期來看,成本估計可能是一部分<br>使用生產工程師和親克管理者識別異常決策系統<br>的單位成本和發射成本削減舉措的長期成分; 成本預測系統<br>也可以通過供應鏈專家用來識別製造商之中最優惠的價格。在<br>長期來看,製造成本的統計估計可以在一概而論<br>設計階段,嵌入CAD / CAM工具,並最終成為期間使用的優化參數<br>部分“和產品的設計由機械工程師。這將是邁向進一步的步驟<br>整合到這樣的航空發動機生命週期的更全面的方法<br>為綜合計算材料工程(ICME)。
正在翻譯中..
