Data-driven asset health monitoring and electrical disturbance analysis for industrial energy applications

Industrial monitoring platforms collect operational and electrical data that can potentially support condition assessment and disturbance analysis, although their usefulness depends strongly on measurement quality and resolution. This thesis aims at developing and evaluating data-driven monitoring methods that use the low-frequency measurements already available through EBOOST, an industrial energy and electrical monitoring platform, to assess the health and operating condition of industrial assets, processes, and electrical systems. Two industrial case studies are considered. The first concerns SACI, a paper-production plant, where auxiliary-specific normal-behaviour models were developed for 18 machines. Five-minute rolling windows, evaluated every minute, were compared with healthy operating conditions using torque current and velocity. Deviations in mean power and power variability were combined into a 0–100 Health Index (HI) with persistence-based alarm logic. Historical validation showed that HI deterioration could precede shutdown; in the clearest case, a sustained decline became visible approximately 30–40 min before shutdown. Plant-wide validation also identified responses associated with all six analysed paper-break periods, with up to 11 auxiliaries reacting simultaneously. The second case concerns BBM, a precision-machining plant affected by short electrical interruptions. Among 120power-qualityevents, 29 interruption records were identified, of which 22 (75.9%) were associated with a voltage dip within ±5 s and 22 showed near-zero three-phase voltage collapse. No repeatable pre-event precursor was identified, and 97.04% of high-risk full-day windows occurred outside the target pre-event interval. The results show that existing measurements can support practical condition monitoring when sufficient information is available, while reliable interruption prediction requires higher-resolution and spatially distributed electrical measurements.