Register for free to
view the full innovation profile.
MCTS methods have contributed greatly to recent AIs, allowing to simulate promising scenarios to make decisions in real time, which is a perfect fit for fulfilment centres.
About
MCTS-micro-fulfilment is born as a concept after working in two complementary projects. The first (1) project was a collaboration with Siemens Digital Lab to develop a monitoring system for Just-in-Time manufacturing. In there we aimed to reduce inventory costs in production processes. Our methodology used historical and real-time data to optimise their operations. We developed a model for prediction of lead-times, helping to monitor the suppliers’ performance; and a model for demand forecasting of different materials to identify slow/fast-moving items. Determining critical levels where actions were needed. For this project we had a Master student (from the technical University of Munich) working under our supervision. The systems created were developed on top of Siemens IT infrastructure as an add-on using Amazon Web Service (AWS) Sagemaker. After finishing our project, they hired the student and further developed the initial concepts. As far as I know this is part of an ongoing patent application.
The second (2) project was not in collaboration with a company, but took inspiration from previous interactions with logistics companies in Germany. Particularly we explored the effects of uncertainty and miscommunication in a supply chain. We assumed certain distribution of demand and lead times, and we focus on the actions that needed to be taken to operate the correct levels of inventory. For that purpose we developed a Monte Carlo Tree Search (MCTS) set of algorithms to optimise operations and management in a warehouse that was part of a supply chain. We took the role of a wholesaler that depended on the distribution centres (upward in the supply chain) as we as our clients (downward in the supply chain). This was made through efficient implementations of several scenarios these models can run in parallel, estimating the 'best' decision (for example how much to order/send) at every single point in time. Creating a digital twin - type of software (that can be built on top of existing IT infrastructure) that represented the constraints and expected outcomes (space, contracts, personnel,..).
The current innovation (this proposal) aims to combine both developments into a single digital-twin that is built exclusively for retailers, taking into consideration their big variety of products (for instance fresh products as well as clothing) and their respective cycles.
Key Benefits
One of the key features of MCTS methods is their suitability for parallel implementations, in this sense we can individually analyse/control the necessary stock for each set of products (fresh products vs clothing, for example).
The digital twin implementation would be based on the in-house data (for instance, daily, weekly sells) that exists on their current systems (for example SAP, Database management systems, or cloud services such as Azure or AWS) to train the respective forecasting models: a lead-time model to monitor suppliers; a demand model to determine the correct stock level. Those models will feed the MCTS-micro-fulfilment model to take actions (for example to place the orders to the suppliers).
Complementary data in the form of textual data extracted from social media, could eventually also be used in the MCTS-micro-fulfilment model. For instance, to perform text-analytics on trends to determine demand levels of certain products. The key idea of this innovation is to use a wide range of data available for the company, and make optimal decisions for their operations.
You can listen to a related podcast session (https://www.hivery.com/resources/category/podcast/dr-felipe-maldonado/) I was invited with the Australian tech company Hivery (https://www.hivery.com/).
In terms of concrete benefits, at least in the case of my experience with Siemens Digital Lab/Siemens Energy was that even earlier implementations of the software led to a more efficient inventory optimisation (that had as consequence happier costumers). That was the whole idea of the Just-in-time manufacturing project, having precisely (no more, no less) the amount of material to build their products when an order was received.
Applications
MCTS-micro-fulfilment would perfectly suit a retail with a rich amount of data (for instance, sales and social media interactions). And it would be a part of a much larger environment.
A systems such as the proposal could be built on top of the current IT infrastructure/Software (for example SAP or AWS). Storing and sorting the sells' information, and running the different algorithms that would keep track of the current incoming orders (from suppliers), stock levels, and giving recommendations on how to optimise the centre (for example suggesting to order certain products).
The ultimate goal would be to create a fully integrated system that would take care of the whole process. Having fully autonomous micro-fulfilment centres, where for instance stock levels could be check automatically and purchasing orders (to suppliers) could be sent by the system (based on demand forecasting and trend-analytics). Robots/mechanical arms could pick up the products needed for restocking the shelves in the physical shops, or alternatively, they will pick basket orders for on-demand delivery.