Building a Digital Twin for a Retail Store
A blogpost by Master Student, Janis Schacht
in collaboration with Andreas Huppert
The concept behind the Digital Twin is fairly simple.
Generally speaking, a Digital Twin consists of a physical object or system and a digital counterpart. These two parts are connected, the physical object or system shares its data with the digital copy. This concept was first described in 2002. The idea behind that concept is much older, it can be traced back to the 1960s and 1970s when NASA used analog twins of their spaceships for training and simulation purposes, before and during the Apollo missions. In 2010, NASA also coined the term Digital Twin in one of their Technology Roadmaps. The digtial concept then gained traction in the context of Industry 4.0, which started in 2013 and heavily focuses on the digital transformation of manufacturing systems.
The Digital Twin has two general ways of usage. The first being to use Digital Twin to interrogate the real-world system based on the data contained in its digital counterpart. Independently from where the real-world system is located, the Digital Twin allows to monitor performance provides insights into the state of the system. The second usage of the Digital Twin is to predict possible future states or behavior of the real-world object through simulations based on its digital counterpart, aiming to improve performance and prevent malfunctions.
The complexity of the Digital Twin depends on the use case. For the sole purpose of interrogating the current status of a real-world system, the complexity lies in holding the real-world system and its digital counterpart in sync. Complexity rises when the use case involves prediction of future behavior and state, then simulation models and Machine Learning come into play. Peak complexity is reached when the Digital Twin is supposed to automatically act on simulation results or sensor date received from the real-world system.
To realize a Digital Twin, technologies from the areas data processing, data analytics, data acquisition and modelling need to be brought together. For simulation, a dedicated simulation model is needed, for visualization 3D Models need to be created and maintained. Data from the real-world system needs to be collected by means like IoT sensors, cameras etc. Large sets of data need to be analyzed and processed, for exampling using tools like Machine Learning and platforms like SAP HANA.
Currently, most Digital Twins are developed in the manufacturing context of Industry 4.0.
….So what would a Digital Twin look like in a retail environment?
To investigate this topic, we started a project to do just that: creating a prototype for a Digital Twin to a retail store.
As mentioned earlier, complexity of a Digital Twin depends on the use case. For our first prototype we therefore planned to merely consolidate and visualize already available data, while already having the broader picture in our mind. For our prototype architecture, we decided to work with two main components. A cloud service which serves as data backbone and integration point for external systems and services, and locally running application to visualize the consolidated data and in the future to run the simulations in.
All in all, we wanted to create a platform that serves as a fundament for a Digital Twin for a retail store, that can be further extended to examine innovative use cases. As a result, we choose the Business Technology Platform to host our cloud service, leveraging its integration capabilities to access systems like SAP S/4HANA or potentially SAP Customer Activity Repository (CAR). Using a cloud connector and on-premise system available to use, we made use of some APIs to access basic data from S/4 like article descriptions, stock, and price. This data would be consumed on demand by our local application, meaning we did not replicate any master data.
Currently, our cloud service itself holds data that describes how our store looks like, e.g. how shelves and articles are placed on the shop-floor, and what other objects are present, like cash registers. The 3D models (assets) for the shelves, articles and other objects are either stored locally, or can be downloaded at runtime from an external asset storage service. In our prototype, we used the XR-Cloud to store our 3D models. The XR-Cloud is an SAP Venture, which stores so called Smart Assets, that contain 3D models and scripting logic if necessary.
This allows us to fully replicate a store, what in turn enables us to get an impression of the store, without actually being in the store. The data from our S/4 system is integrated, so that we can inspect articles and see how much stock is left or what their price is. This shows that every bit of information in S/4 that is exposed via an API can be visualized in a 3D Interface. To summarize, we created a representation for a retail store in virtual reality.
The next step is to investigate possible use cases, and to further extend our platform to make this possible. A fairly simple use case that is realized by our current solution is checking if a store is in compliance with guidelines from its headquarters, by checking if the shelves in the real store look like the shelves in the real store. Also, store layout can be evaluated before rearranging shelves in the store. For example, by using an VR-Headset and going through the twin, store layouters can get a feeling if a layout works out, or to check visibility of articles, shelves and departments.
Another visualization use case requires tracking of customer movement through the store. This allows the creation of heatmaps to visualize areas that are highly frequented by customers, and in general to see how customers walk through a store to do their shopping. Store Layouters/Analysts can gain valuable insights from such visualizations, that aid them in their work and to optimize the store layout.
But also use cases like simulating customer flow through a store and how different layouts and article placements affect customer flow can be investigated. Of course, these simulations require realistic customer models to be representative.
Conclusion
To summarize, using various tools and products from SAP like the Business Technology Platform, the XR-Cloud and S/4HANA, we were able to build a platform for a Digital Twin for a retail store. This platform will serve as a continuously expanded framework to investigate various innovative use cases.