On Feasibility of Use of Adaptive API Caching Mechanism for Predictive Maintenance in Energy Systems : Adaptive API Caching for Predictive Maintenance: A Proof of Concept

Sammanfattning

Predictive maintenance plays a critical role in ensuring the reliability of modern infrastructure, particularly as society becomes increasingly connected and dependent on uninterrupted access to digital services. Telecommunication systems form the backbone of this connectivity, supporting essential sectors such as healthcare, aviation, and emergency response. In such contexts, even minor faults can lead to severe disruptions that are both costly and time consuming. Predictive maintenance relies on continuous data monitoring, which is repetitive and increasingly demanding due to the transient and high-volume nature of the data. In collaboration with Ericsson, this thesis proposes the use of adaptive RESTful API caching to alleviate that strain on the system; specifically within the energy systems powering radio sites. The work aimed to assess its feasibility by measuring improvements in response time while ensuring the consistency of data freshness. It also provides a proof-of-concept model for how such an adaptive mechanism could be designed, a dynamic technique that adjusts caching behavior based on data volatility and context. The work involved the development of the entire system pipeline; including the database, data stream and API, and emphasized how it, in particular the data stream, influenced the caching design. Results showed that adaptive caching led to a 15.7% improvement in response time while maintaining 100% data freshness. The findings also highlighted the importance of designing the caching around the capabilities and limitations of the system. It also demonstrated the challenges in developing and assessing predictive maintenance features in early-stage environments. The natural next course for this work is to extend the mechanism with greater granularity and be more comprehensive, and to implement in on-site to accurately reflect a real world data conditions.