A data-driven approach to maintenance-preventable causes of failure analysis in power distribution systems

Goal: This study proposes to investigate the main causes of failures in a power distribution system (PDS) that can be mitigated through the implementation of best practices in maintenance management. Design/Methodology/Approach: This research proposes a data-driven decision-making approach to aid the preventive maintenance management through the analysis of Maintenance-Preventable Causes (MPC) of failure in a real-life PDS. The proposed methodology is structured into three key steps: (1) Data collection and processing of 7,721 power distribution failure records over 12 months (all resolved within 6 hours); (2) Pattern detection using machine learning (ML) algorithms, specifically Association Rule Learning (ARL); and (3) Critical event identification via Social Network Analysis (SNA) with graph-based visualization. Results: The results show that there was a reduction in the continuity indicators Equivalent Interruption Duration (EID) and Equivalent Interruption Frequency (EIF) by 8.54% and 6.26% respectively, taking as a basis only one MPC (vegetation). The model enables more assertive guidance for both resource planning and the execution of preventive maintenance actions in distribution networks. Limitations of the investigation: The data used were limited to the southern region of Ceará, Brazil. Therefore, by applying the same methodological approach, other power distribution systems can also be analyzed. Practical implications: A case study in the southern region of Ceará, Brazil, was conducted to demonstrate the practical applicability of the proposal. This study contributes to identifying variable dependency between failures associated with each MPC and the critical points with the highest impact on the distribution system. Originality/Value: This study contributes to the academic literature by applying a model that aids in identifying and mitigating the primary failures occurring in power distribution systems, which result in financial losses associated with power supply interruptions, through the use of text mining techniques.