After-Market Spare Parts Forecasting at Sandvik Stationary Crushing & Screening

Abstract

Title: After-Market Spare Parts Forecasting at Sandvik Stationary Crushing & Screening.

Authors: Arian Marofkhani and Artur Jusopov

Supervisors: Professor Gudrun Kiesmüller, Lund University, Faculty of Engineering, Division of Production Management.
Macarena Ribalta, Planning & Logistics Manager, Sandvik Stationary Crushing & Screening.

Examiner: Professor Johan Marklund, Lund University, Faculty of Engineering, Division of Production Management.

Background: Sandvik Stationary Crushing & Screening in Svedala is planning to fully roll out a forecasting system called Voyager and needs guidance in their forecasting process. Sandvik wants to explore how different forecasting methods could help to improve forecast accuracy on a SKU level as well as on a SKU/Stockroom/Customer Cluster level.

Purpose: The purpose of the master thesis is to identify and propose quantitative forecasting methods with the aim to improve forecasting accuracy on a SKU and SKU/Stockroom/Customer Cluster level.

Methodology: The research approach aims to fulfill the purpose of the study by performing a single case study research. The study adopts an exploratory, explanatory and a descriptive focus to gain deep insight into the research area as well as to understand the current situation of the case company - Sandvik SRP AB. The study incorporates an empirical, data driven approach to collect and filter historical data as well as apply quantitative forecasting methods.


Result: The best forecasting method was determined for all ABC-XYZ classes. The best method for each ABC-XYZ class performed better than Moving Average 12. Simple Exponential Smoothing yielded the best forecast accuracy for classes AX, AY, BX, BY, CX and CY on both level 1- and 3 with an exception of class CX on level 1. SBA and Croston’s method yielded the best forecast accuracy for classes AZ, BZ and CZ on both level 1- and 3. The reliability of point forecasts seems to increase with a lower coefficient of variation in time-series.

Recommendations: It is recommended that Sandvik classifies their products according to an ABC-XYZ classification, where Simple Exponential Smoothing is the recommended forecasting method for classes AX, AY, BX, BY, CX and CY on both level 1 and 3. It is also recommended that SBA and Croston’s method should be used for classes AZ, BZ and CZ on both level 1 and 3. Further, it is recommended that the forecasting accuracy is monitored with the help of a tracking signal to ensure tolerable forecasting accuracy.

Popular Abstract

After-Market Spare Parts Forecasting at Sandvik Stationary Crushing & Screening

By Artur Jusopov & Arian Marofkhani (April 2022)

Maintaining appropriate stock levels while meeting customer demand is a challenging feat most companies face on a regular basis. Therefore, it is of great importance for companies to have suitable forecasting methods to predict the future demand of their SKUs. In our thesis, we investigate alternative forecasting methods that are suitable for the assortment of SKUs Sandvik Stationary Crushing & Screening in Svedala offers. The main focus of the thesis was the after-market products Sandvik has. In the world of forecasting, the demand of these types of products are notorious for being difficult to accurately forecast. This is mainly due to the irregular demand that usually comes with them, making it hard to predict future demand based on historical demand. This is why Sandvik struggles with achieving satisfying forecasting accuracies on their SKUs, and the reason why we conducted the thesis.

In order to propose forecasting methods that are suitable for their products, we decided to divide all products based on historical sales value and demand irregularity. With this, we could classify products based on their monetary value and how easy it would be to forecast each product. The benefit is the ability to see which items should mostly be focused on and which items to focus less on. For example, if an item has high value and is relatively easy to forecast, focusing on this particular item would have a large impact on the company’s financial gains. Compare this to focusing on an item that has a low value and is very difficult to accurately forecast, the benefit would be minimal. Maybe these types of SKUs should not be forecasted at all. After grouping each SKU in such a way, we tried different forecasting methods for all the items in each class in order to determine the overall best method for each class. The methods that we investigated are: Croston, SBA, Simple Exponential Smoothing, Holt’s Linear, Holt’s Linear Damped, Naïve method and the Seasonal Naïve method. The method that Sandvik currently uses is called Moving Average 12, where the monthly forecasts are based on the average of the last 12 months.

The results of our study showed that the methods that we investigated made an improvement in forecasting accuracy compared to the current method, Moving Average 12. Out of the seven methods mentioned, Croston, SBA and Simple Exponential Smoothing were the ones that yielded the best results.
The findings in our study could be useful for other companies that have SKUs with similar demand patterns. By classifying their products as we did, it would be easier to appoint different forecasting methods for each class. Hence, possibly improve their forecasting procedures.

Lund University, Faculty of Engineering.

• author: Jusopov, Artur and Marofkhani, Arian
• supervisor: Gudrun Kiesmüller
• organization: Production Management
• course: MIOM05 20212
• year: 2022
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: Forecasting, Intermittent Demand, Spare Parts, After-Market, Crostons, Exponential Smoothing, Hierarchical Forecasting
• report number: 22/5687
• language: English
• id: 9093736
• date added to LUP: 2022-07-01 14:13:34
• date last changed: 2022-07-01 14:13:34