Introduction & challenge
ZAS needed a more efficient way to organize its stockroom. The current system resulted in under-utilised storage space and inefficient stockroom management, leading to wasted time in retrieving products. The challenge was to minimize the number of module units used, mainly based on product dimensions, without compromising on the required separation between sterile and non-sterile items and ensuring proper product placement based on weight, the ability to stack, and other characteristics (the 3D bin packing problem).
Instead of finding the most mathematically optimal solution - which becomes increasingly complex the more products you add - the goal was to use heuristics to create a solution that would improve stockroom efficiency while keeping the model's runtime manageable.
Project approach
Model requirements
The model was required to take into account various product and module characteristics. Product-specific characteristics were product dimensions, product group, sterile and non-sterile products, weight of the items, stackable items, and nestable products. Also, different module dimensions had to be considered when determining storage capacity.
Model execution
The model was designed to be run directly from a web application, which interfaced with a backend algorithm to handle the heavy computation. Users could upload a template file containing product data, configure parameters, and initiate the model run from the interface. The algorithm would then generate an optimised stocking arrangement, taking into account the user-defined parameters and the heuristic algorithm.
Output and evaluation
Once the model was run, users could access an overview of the results in a web application, where a summary of the model’s output could be consulted. Key metrics included:
- The number of storage modules and racks utilised.
- The percentage of empty space in both modules and racks, indicating the efficiency of space utilisation.
- A graphical representation of how products were organised within the modules.
The model also generated a list of products that did not fit within any available bins and required special attention. This output could be consulted in the web application itself and a report was also available for download in PDF.
Results
The use of heuristics allowed us to strike a balance between space efficiency and computational speed. The model achieved a high degree of space utilisation while respecting the constraints around product placement and separation. Additionally, the model can be used to simulate the number of modules and storage units needed for new stockrooms or whenever there is a change in existing stockroom product portfolio.
ZNA conducted parallel testing and validation to ensure the model’s effectiveness and alignment with the hospital’s operational requirements. This involved running the model concurrently with the hospital’s existing stockroom management practices, allowing us to compare the outcomes in real-time.
This testing approach provided valuable insights and ensured that the model performed reliably across various departments and storage configurations. It also gave the hospital confidence in transitioning to the new system without disrupting existing processes.
Conclusion
This project showcased the effective use of heuristics in solving a complex stockroom optimisation problem within a hospital environment. By prioritising runtime efficiency over mathematical optimality, we were able to deliver a practical solution that improved the hospital’s operations. Our model is adaptable to future changes in inventory and storage configurations, ensuring its long-term utility.
For more on how we’re using Machine Learning to optimize workflows, visit our Machine Learning page.
