Predictive maintenance of chromatographs
Titova, Diana (2022)
Pro gradu -tutkielma
2022
School of Business and Management, Kauppatieteet
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2022040426856
https://urn.fi/URN:NBN:fi-fe2022040426856
Tiivistelmä
New methods of prognostic analytics for industries are invented constantly, and research of predictive maintenance has become widely popular in recent years, so it becomes a requirement for the company to implement data-driven decisions into the working process. There are many works considering predictive maintenance implementation. Nevertheless, input data highly influences the process of models implementation, and business cases devoted to this phenomenon are limited. This Master’s Thesis provides an analysis of predictive maintenance models implementation in the biotechnological industry.
In the literature review, the main directions of predictive maintenance research and remaining useful life estimation topic were discussed. In the empirical part of the thesis methods of Machine Learning and Deep learning were applied and compared. Decision Tree Classifier and Random Forest Classifier outperformed other models in terms of four different metrics - accuracy, precision, recall, and f-score values – and training time. Since the company is interested in the interpretation of models predictions, the two models were chosen as the best algorithms for predictive maintenance with four metrics equal to 1 on both validation and test data. Based on the findings, managerial implications and theoretical contribution of the work are given.
In the literature review, the main directions of predictive maintenance research and remaining useful life estimation topic were discussed. In the empirical part of the thesis methods of Machine Learning and Deep learning were applied and compared. Decision Tree Classifier and Random Forest Classifier outperformed other models in terms of four different metrics - accuracy, precision, recall, and f-score values – and training time. Since the company is interested in the interpretation of models predictions, the two models were chosen as the best algorithms for predictive maintenance with four metrics equal to 1 on both validation and test data. Based on the findings, managerial implications and theoretical contribution of the work are given.