Greedy Algorithms Part II: The Data and the Models
Read Part I here.
Warehouses are very busy places. They are fast-paced environments that require experienced workers to efficiently scan and process thousands of bin locations to avoid any delay in operations and activities of the warehouse.
Therefore, execution time is a critical factor when we design and develop data pipelines and algorithms to provide a seamless experience for the end user.
We have used three data pipelines which consist of static and live data:
- Transactional data (data which has a time dimension, a numerical value and refers to one or more objects)
- Map of the warehouse — Every warehouse has a different architecture, so we captured specific maps for each warehouse and transformed it so our algorithm can actually use it.
- Current on-hand inventory of the warehouses — Since thousands of tires move in and out of warehouses, inventory data needs to be live and accurate.
Since the warehouses hold a vital role in the distribution of tires to other distribution centers (DCs), we cannot just “shut down” their operation or spend all of our resources to redistribute the tires. To tackle this tricky situation, we developed two algorithms/solutions: “cleaning” and “put-away.”
Cleaning algorithm: It helps redistribute products (tires) in right locations e.g. Move a fast mover to the front of the warehouse and a slow mover to the back of the warehouse in order to reduce pick up time.
It basically scans the entire warehouse and runs a greedy algorithm to find out the appropriate locations for the products.
The final output is a set of moves that helps the operator to move products in right locations (bins). Another important aspect of this algorithm is to tie each move with value savings to the business. Therefore, the manager can prioritize moves and have a stronger impact in a short period of time.
Put-away algorithm: It helps slot the product in right locations (bins) when they arrive at warehouse.
The result of the put-away algorithm had a significant impact on our warehouse as we regularly receive thousands of tires and our goal is to slot them in the right manner to avoid the need of cleaning.
This algorithm runs through different rules and scenarios as well as executes space and distance optimization to suggest the best locations for the tires as they arrive. The final output guides the operators and saves significant time ensuring we use the best possible location of the warehouse for each and every tire.
The cleaning and put-away algorithms are dependent on multiple constraints such as capacity of the bin (location), relative distance to the docking area, bin dimensions, etc.
In order to develop these algorithms/solutions, it was important to capture the knowledge from subject matter experts and standard operating procedures, as well as bin and aisle locations in order to prioritize the pick-up and drop off locations. The algorithms needed to have accurate data (real time) in order to deliver a solution that fits within the day-to-day activities and processes of warehouse. This solution was designed in order to redistribute and slot products during normal business operations, rather than shutting down operations to do so.
The main challenge of the data pipelines and algorithms is the design, reliability and automation in order to deliver fast and seamless real-time experiences.
Check back on Tuesday to see how we made these algorithms usable for our associates in Part 3!
