Data-based prediction of equipment failure in industrial processes : Case: trace heating systems
Samad, Abdul (2025)
Pro gradu -tutkielma
2025
School of Business and Management, Kauppatieteet
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2025061971950
https://urn.fi/URN:NBN:fi-fe2025061971950
Tiivistelmä
Heat trace systems are used in industrial settings to control the temperature of pipelines and equipment. Heating cables are the core elements of these systems. This thesis studies the use of predictive maintenance algorithms on heating cables to anticipate performance degradation and reduce the risk of failure. A circuit level data for Self-Regulating (SR) cables and Mineral-Insulated (MI) cables provided by Finnish company in real industrial settings is used for modeling.
This study applies clustering, anomaly detection and classification to assess performance and predict faults in both cables. In SR cables, K-Means clustering groups circuits based on temperature, current and leakage patterns. Isolation Forest identifies abnormal current or temperature behavior that may sign some failure. A Random Forest model is trained to detect current leakage.
In MI cables, clustering separates circuits by performance patterns derived from averaged weekly features. These groups reflect differences in efficiency and temperature response. An XGBoost classifier is trained to predict degradation risk based on these patterns. The models support early identification of cables likely to fail. These results support the predictive modeling in identifying performance issues in heat trace systems.
This study applies clustering, anomaly detection and classification to assess performance and predict faults in both cables. In SR cables, K-Means clustering groups circuits based on temperature, current and leakage patterns. Isolation Forest identifies abnormal current or temperature behavior that may sign some failure. A Random Forest model is trained to detect current leakage.
In MI cables, clustering separates circuits by performance patterns derived from averaged weekly features. These groups reflect differences in efficiency and temperature response. An XGBoost classifier is trained to predict degradation risk based on these patterns. The models support early identification of cables likely to fail. These results support the predictive modeling in identifying performance issues in heat trace systems.