Developing a Forecasting Simulation Model for Efficient Warehouse Operations

Developing a Forecasting Simulation Model for Efficient Warehouse Operations

Overview

In Thailand, there is a logistics center that is supplied by two factories. This logistics center, also known as a warehouse, engages in automated storage and retrieval system (ASRS) logistics and operates manual service pallet racks.

This warehouse is made up of 2,000 m² of floor space, which is used for picking and staging operations. Ten loading docks are used to move approximately 1,500 pallets each day for inbound and outbound activities.

Problem

The operations in this warehouse are quite complicated because there is an unpredictable load during the course of a regular day. In addition, mostly partially filled pallets arrive from the factories, which means that they need to be joined together with other pallets inside the warehouse.

There are also lots of different product configurations that need to be considered, creating challenging processes that are hard to monitor.

Finally, due to the storage space being overutilized and the varying loads, it is difficult to predict in advance how many operators will be required every day. As a result, these operators are frequently under- and overutilized.

The two companies that work with this warehouse, Western Digital and DSV, did not have a tool that would allow them to input forecasts, take into consideration these complicated operations, and then retrieve expected values for certain KPIs.

Solution

EPIC InnoLabs, a small digitalization company, was contracted to develop a forecasting simulation model for Western Digital and DSV. It would need to include a variety of experiments, system optimization, what-if scenarios, forecasts, and estimates in order to create more efficient warehouse operations.

The model was built using AnyLogic because it supports multimethod modeling, which provides the ability to simulate business systems of any complexity, from small warehouses to huge logistics centers. AnyLogic is also flexible since it is Java-based and enables the visualization of warehouse activities. It is also possible for the model to be migrated to AnyLogic Cloud, which allows for integration with business analytics tools in the future.

The model is currently still under development. However, the base model is finished and has been validated. It has a data-driven solution that can predict system behavior based on input data.

Excel files are used as inputs in the model and include:

This is all historical data that is being used. Other parameters can be defined directly in the model.

Below, the process flows of the model architecture are displayed in detail.

The different process flows for the forecasting simulation model displayed with details about different rules in operation

Process flows of the forecasting simulation model for efficient warehouse operations

The model was created using the AnyLogic Process Modeling Library and a combination of agent-based modeling and discrete-event simulation. Statecharts were used to coordinate warehouse activities like picking and placing operations as well as resource monitoring.

The model has an easy-to-use interface where the user can input the data, modify some parameters, and then initialize the warehouse model. There are numerous views available that can provide information. In the 3D view, for example, the pallets can change color depending on whether they are full or not. During runtime, KPIs about the product statistics, lead times, and other charts can be followed or checked. The data can then be exported to Excel to be analyzed in more detail.

The model also provides the opportunity to run two parameter variation experiments. The first is used to understand the effects of changing the number of workers on lead times, productivity, and other KPIs. In the second, the parameters of the ABC logic (an inventory management technique in which goods are divided by their value to the business) can be varied. This ABC logic is based on the number of orders for every product, either for one week or two weeks, and so on. The effects of this will have an impact on lead times.

The user interface of the forecasting simulation model showing the warehouse in 2D and 3D views and different statistics (click to enlarge)

Results

The first round of experiments has already been run, and the model has been improved. It has now incorporated ABC logic into the ASRS logistics system and has more detailed resource monitoring to measure more specific KPIs in the warehouse operations model. As the developers are still in the middle of this project, exact information is not available.

In the future, the time horizon of the model will be extended to 7 months from the current 3 months. The second round of experiments, focusing on resource utilization and warehouse efficiency, will then be run. Forecast data will be used instead of historical data. Finally, they will migrate the model to AnyLogic Cloud.

The case study was presented by Andrea Mácz, of EPIC InnoLabs, at the AnyLogic Conference 2022.

The slides are available as a PDF.



Études de cas similaires

Plus d’études de cas

Obtenez une brochure avec des études de cas de l’industrie (en anglais)

Télécharger