考量安裝基礎產品失效率之零件存貨的最後訂購預測模型

Abstract

由於市場競爭日趨激烈，消費者除了重視產品的品質與價格外，產品的售後服務也成為消費者選擇品牌的重要參考依據，尤其具備單價高、使用年限長的耐久財產品(如汽車)更為重要。當服務零件已介於生命週期的衰退期時，需求快速遞減，而零件供應商基於生產批量的考量，並不會持續生產零件至服務終止時間點，產品代理商通常會面臨最後一次訂購的問題，以滿足剩餘服務期間的零件需求。為了避免訂購太多，造成零件呆滯的風險，而訂購太少又可能造成缺貨，產生顧客抱怨，最後一次訂購的議題一直是業界急欲解決的問題。 針對服務零件在產品衰退期最後訂購量的需求預測，常受到零件需求資料不足、消費者購買行為不斷改變、傳統的需求預測模式不適用等因素所困擾。本研究以汽車代理商為個案研究的對象，提出Regression 模型與Installed base product模型。Regression模型直接針對服務零件不同的需求趨勢，以不同的回歸方程式進行預測，預測誤差遠高於個案公司現行的方法。 Installed base product模型首先透過市場上汽車的流通數量、顧客的回廠維修率與服務零件需求量，用以推估最後訂購時間點前服務零件的失效機率。其次，分析失效機率是否存在趨勢，將不同趨勢失效機率的零件進行分類，並以不同的回歸方程式進行最後訂購時間點後失效機率的預估，再乘以市面上汽車的流通數量，即可預估未來的服務零件需求與最後訂購量。第三，本研究也發現一個資料點數量與資料接近性的重要研究議題，就統計分析而言，當輸入較多資料點時，通常會產生較準確的結果，但本研究藉由採用Installed base product模型，發現資料趨勢、資料點數量、資料接近性等因素的交互影響。 為了使本研究所建構的Installed base product模型能廣泛的應用，最後以兩個不同產業服務零件的需求資料進行實證研究。第一個是汽車產業，透過第一種車型數千個服務零件進行模型建立，再利用本研究發展出的模型，以不同車型的服務零件進行驗證。第二個是筆記型電腦產業，以電腦主機板等服務零件做為驗證對象。模型驗證結果顯示本研究所提出的需求預測模型因結合不同趨勢的預測方法，與個案公司現行方法相比較，在需求預測誤差上有較佳的表現。

While the competition in the market becomes severe, in addition to the quality and price of products, after-sales services have become all the more important for consumers to choose brands especially for the durable goods with high price and long life time. When service parts are at the end-of-life phase, suppliers faced with the dwindling demand will subsequently set target dates to discontinue part production. The agent will face the final ordering problems to satisfy the demand in the remaining service period. In order to avoid the dead stock and back orders of service parts, the final ordering problems have been an urgent issue in the industry. Demand forecast of service parts at the end-of-life phase is plagued with limited demand data, changing purchasing behavior and inapplicable conventional forecast method. This paper presents an empirical study of an automobile firm by applying regression and installed base product forecast method. Several archetypes of demand trend are identified and direct regression on the historical demand of service parts is proved to be inadequate in predicting future demand. The installed base product method first considers the population of vehicles in use, return rate for customers to repair and historical demand data of spare parts to estimate the failure probability of service parts before the final ordering decision time. Second, this paper conducts the trend test on the failure probability in order to classify the service parts with different tend and use different regression function to predict the failure probability after the final ordering time. Third, in the case of statistical analysis, the results are more accurate when there is more data input. However, by applying installed base product method, the interrelated effects of data trend, data quantity and data recency are unraveled. At last, the proposed installed base product method is validated with two automobile series and notebook computer models. It is shown that the proposed method outperforms a current meth-od by large margins in forecast errors. In addition, the proposed method shows consistent results for three data sets in two industries.