Optimizing Product Production Schedules

The Challenge

Manufacturing plants for Consumer Packaged Goods (CPG) must determine the order products will be produced in each week. An efficient production order is critical since carrier schedules and order delivery dates are dependent on “just in time” manufacturing operations. One of our CPG client’s plants developed a preferred schedule to minimize total line changeover time (the time required for plant operations teams to change equipment and specifications of a manufacturing line to produce the next product) each week. Despite this development, the efficiency (using the metric OEE) of this plant has generally been lower than their other plants. As a result, the client asked Elder Research to find ways to further improve this plant’s production schedule.

The Solution

Our team used Google OR-Tools to develop an algorithm to determine the optimal production schedule for each line at the plant. This type of problem is known as a Single-Machine Scheduling Problem (SMSP), which is a special case of the well-known Traveling Salesman Problem (TSP). The primary input data to the model is a matrix of all the possible product transitions on each line, which we developed by cleaning and processing historical transactional data. The transition options and optimal ordering is shown in the following figure.


The algorithm that our team developed showed an average changeover time reduction of 19.9% compared to the client’s current process when tested on data from a set of historical weeks. This time reduction is realized each week when the algorithm is run to determine the optimal production schedule. However, due to implementation considerations, our client preferred using a single optimal production schedule. Therefore, we adapted the algorithm’s results to produce one optimal schedule that could be used on any week for any subset of products. This adapted model still showed an average reduction of 5.8% in changeover time compared to the client’s current process when tested on data from a set of historical weeks and will help increase user adoption since it was delivered to the client in a format that is similar to their current process. It is expected that this amount of time reduction could save the client approximately $425,000 at this plant over 6 months of implementation across all lines.