Publicação
Predictive Maintenance use case employing Survival Analysis in a telecommunication company
| Resumo: | Driven by the digital revolution, telecommunications companies need to adopt innovative technologies and services to be competitive. In this context, the company invests in its first Predictive Maintenance solution, intelligent anticipation of device failure through sensorial data. This solution has the power to anticipate and plan Reactive Maintenance measures that extend equipment’s life, reduce downtime, aim for cost savings, and avoid negative feedback, consequently improving the service quality. This project explores a Fault Prediction tool such as Survival Analysis. It undergoes the six phases of a Data Science Project following the CRISP-DM methodology. For applying the Survival Analysis technique (e.g. Kaplan-Meier), it is crucial to identify two key events using the equipment’s historical data (e.g. STB): The beginning of the anomalous event and the exact moment of the fault event. Several techniques, such as Statistical Smoothing models and Anomaly Detection models, were analysed and compared in detail to detect the beginning of the device malfunction. The best results to detect the devices’ anomalous event were employed by a Statistical technique, the SMA where the anomalous event is reaching 50 degrees for the one-day smooth average. Therefore, it is possible to obtain an acceptable anticipation period of 38 days for future equipment maintenance intervention. In this sense, employing a Predictive Maintenance solution guarantees the reduction of 71% of the actual emergency interventions. Consequently, the company saves more money rather than not making any prediction at all. Moreover, it was also developed a visualisation tool to demonstrate the solution and explore it, where it employs the different models to detect the beginning of the anomalous event’s. Consequently, all the proposed goals of the company were accomplished. |
|---|---|
| Autores principais: | Costa, Marta Carocho de Sousa |
| Assunto: | Predictive Maintenance Survival Analysis Anomaly Detection Fault Prediction |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Driven by the digital revolution, telecommunications companies need to adopt innovative technologies and services to be competitive. In this context, the company invests in its first Predictive Maintenance solution, intelligent anticipation of device failure through sensorial data. This solution has the power to anticipate and plan Reactive Maintenance measures that extend equipment’s life, reduce downtime, aim for cost savings, and avoid negative feedback, consequently improving the service quality. This project explores a Fault Prediction tool such as Survival Analysis. It undergoes the six phases of a Data Science Project following the CRISP-DM methodology. For applying the Survival Analysis technique (e.g. Kaplan-Meier), it is crucial to identify two key events using the equipment’s historical data (e.g. STB): The beginning of the anomalous event and the exact moment of the fault event. Several techniques, such as Statistical Smoothing models and Anomaly Detection models, were analysed and compared in detail to detect the beginning of the device malfunction. The best results to detect the devices’ anomalous event were employed by a Statistical technique, the SMA where the anomalous event is reaching 50 degrees for the one-day smooth average. Therefore, it is possible to obtain an acceptable anticipation period of 38 days for future equipment maintenance intervention. In this sense, employing a Predictive Maintenance solution guarantees the reduction of 71% of the actual emergency interventions. Consequently, the company saves more money rather than not making any prediction at all. Moreover, it was also developed a visualisation tool to demonstrate the solution and explore it, where it employs the different models to detect the beginning of the anomalous event’s. Consequently, all the proposed goals of the company were accomplished. |
|---|