The Impact of Predictive Analytics on Spare Parts Inventory Planning

Editor’s Note: Every year, 40 or so students in the MIT Center for Transportation & Logistics' (MIT CTL) Master of Supply Chain Management (SCM) program complete one-year thesis research projects. The students are early-career business professionals from multiple countries with 2 to 10 years of experience in the industry. The research projects are sponsored by and carried out in collaboration with multinational corporations. Joint teams of company people, MIT SCM students, and MIT CTL faculty work on real-world problems chosen by the sponsoring companies. In this series we summarize a selection of the latest SCM research. The researchers for the project described below, Mike Brocks and Renzo Trujillo, analyzed how to take into account predictive analytics in spare parts inventory planning in their SCM thesis The Impact of Installed Base and Machine Failure Prediction on Spare Parts Forecasting and Inventory Planning. The work was carried out for a leading service supply chain provider, and the project was supervised by MIT CTL's Dr. Chris Caplice, Dr. Daniel Steeneck, and Dr. Francisco Jauffred.

The Internet of Things gives companies access to large amounts of data streams from machines in their installed base. While many companies are using this new capability to perform preventative maintenance and respond to machine breakdowns proactively, they have not figured out how to incorporate this new data into their spare parts inventory planning process, and what the impact of doing so would be.
The MIT CTL researchers developed and tested a new predictive failure spare parts planning method, and compared it to a traditional time-series forecasting method via an inventory planning simulation model.

The Value of Preventing Downtime

The sponsor company handles service supply chain operations for a wide variety of companies in a number of different industries. In a typical process flow, they are responsible for the coordination and execution of all activities from the initial customer service call to the dispatching of spare parts and service technicians to the point-of-failure in the field.

In many cases, the sponsor company's customers have large, real-time datasets pertaining to the status of each machine's operating condition. Additionally, these companies pay penalties to their customers and/or suffer large revenue losses for every interval of time a machine has failed in the field. Accordingly, the customers are constantly balancing the amount of working capital they have tied up in spare parts inventory with maintaining a high level of uptime for their products in the field.

One way to address these alternative goals would be to incorporate the real-time data streams into a predictive failure spare parts planning method. This would potentially allow companies to hold minimal spare parts inventory, and proactively order inventory only ahead of a known failure occurring.

The researchers developed a method to incorporate a predictive failure approach into a periodic review spare parts planning system by using a binary classification matrix. They then analyzed the results of various levels of predictive failure on the ability to reduce spare parts inventory levels, while maintaining similar service levels in comparison to a traditional time-series forecasting method.

The two key inputs the researchers used in their model were the True Positive Rate (TPR) and the False Positive Rate (FPR), or the number of times ‘signals' in the machine data accurately and inaccurately predicted product failure, respectively. The three key measurements they used to compare the performance of each model were the Cycle Service Level, Item Fill Rate and Average Inventory Level throughout the demand horizon.

Predictive Forecast Potential

The researchers found two key relationships in the results of their analysis. First, as the difference between the TPR and FPR grows, the average inventory level decreases while the service level increases. At either a TPR of 100% and an FPR at 0%, or a TPR of 0% and an FPR of 100%, the researchers found that the average inventory level reaches its theoretical minimum, with zero safety stock needing to be held, and a service level of 100%.

Second, they find that even at equivalent TPR and FPR values, the predictive forecasting model performs the same as the time-series forecasting model. This suggests that for companies who have access to large streams of machine-data, they can take a standardized spare parts inventory forecasting approach across all products and components, regardless of the predictive power of each individual component analysis.

For more information on MIT CTL's Supply Chain Master's program and the student theses visit http://scm.mit.edu/program or contact SCM program Executive Director Dr. Bruce Arntzen at: [email protected].

SC
MR