Umeå University's logo

As the Fourth Industrial Revolution blurs the boundaries between the physical, biological, and digital worlds, new opportunities arise. This thesis, conducted in collaboration with Smurfit Kappa, investigates the possibility of predicting maintenance in the manufacturing industry using machine learning methods such as Support Vector Machines, Random Forests, and Deep Neural Networks. To reduce operational and financial risk, the company aims to shift from tradi- tional spare parts strategies to modern data-driven solutions. Historical stock withdrawals, total inventory quantity, and lead times have been central compo- nents of the methodology, with historical data from 2012 to 2023 being analyzed. This resulted in Support Vector Machines with a linear kernel function and a fine-tuning parameter of C = 0.0001 tending to perform best according to sta- tistical validation measures. However, this result was contradicted when a stress test across five different realistic scenarios was conducted, where Random Fo- rests performed best. The study’s findings indicate that irregular patterns in the historical data affect the overall performance of the models. The overall econo- mic analysis presented shows that a few items contribute to a very large part of the total inventory value. To address this, an alternative inventory strategy has been defined to reduce the total inventory value by counteracting overstocking. The strategy was based on the spare part’s historical criticality, lead time, and predicted withdrawal from the model. The developed inventory strategy could potentially result in cost savings of up to SEK 13,000,000, increasing financial flexibility as the inventory ties up less capital, thus reducing financial risk. It would have been of great interest to relate the calculated inventory strategy to the cost if the factory were to shut down due to the lack of a specific spare part, but this data was not available for this project.