以機具存量為基礎的售後零組件需求預測

Abstract

隨著產品市場競爭日趨激烈，售後服務的重要性也日益提高，它不僅能累計對於顧客行為的知識，以建立廠商服務的差異性，更是產品銷售後另外持續創造利潤的來源。一般的售後服務提供顧客一段產品的保固期，在期間內免費提供維修服務，而對於耐久財或是產品單價較高的產品，例如汽車產業，其零組件服務期間比產品的生命週期還長，在產品停止生產後，由於零組件供應商必須在考量成本因素下將不會持續提供零件至服務期間結束，此時汽車代理商須進行零件的最後一次訂購，以滿足零件在衰退期間的剩餘需求。基於僅有一次的訂購機會，為了避免訂購過多導致存貨成本的增加，或是訂購太少而影響服務水準，因此零組件的最後訂購量議題在實務上往往對服務廠商造成管理存貨上的問題。 因此，本研究以汽車產業為個案研究對象，以此提出一個較準確且符合實際的零件需求預測模型。透過Installed base 模型，將零件的需求原因歸納為三種因素，分別是市面上汽車流通數量、顧客回廠維修機率以及零件的失效機率，利用前兩項已知資訊及零件於最後訂購點前的需求量，估計零件在各使用年齡下的失效機率。接著，由於失效機率為一時間序列值並具有趨勢，因此透過趨勢判斷將零件分類，各自採用不同的方程式進行估計，並再次利用Installed base模型逆推求出零件的各期需求量，加總後成為最後訂購數量。 最後，為了使本研究所提出之預測模型更具廣泛性與適用性，因此以不同車型零件與筆記型電腦零件進行模型驗證，並比較本研究所提出的方法與個案公司現行方法之優劣。模型驗證的結果顯示不同種類的零件其失效機率確實存在著不同趨勢，而本研究所提出之預測模型因結合不同趨勢的預測方法，在預測誤差上有較佳之表現。

While the competition in the market becomes severe, the importance of after-sales service has been much emphasized. It not only help companies to gain a deep understanding of customer’s behavior which provides a competitive advantage, but also generates a revenue stream after products are sold. Basically, after-sales services provide a warranty period to maintain the product by service parts. Since the periods of maintenance and replacements of spare parts are much longer than the product’s production periods, after the sale of a product is discontinued, there is an installed base to be serviced and there is only one final chance to stock up the part inventory. As a result, solving this end-of-life final-order inventory problem is crucial in practice. This paper presents an empirical study of an automobile firm on this problem by applying an installed-base forecast method. Installed base model divides parts’ demand into three factors: the population of products in use, replacement probability of the failed parts and the failure rate of parts. For each part type, the failure probabilities over the life time are first estimated and a trend test is applied to the failure probability. A hybrid method is proposed by fusing the trend and end-of-life customer behavior. At last, in order to strengthen the application of the proposed model, data from the automobile and notebook computer industries are used to validate the model which shows significant improvement over an existing method used in practice.