A Study on Degradation-Topology Graphs for Asset Failure Probability Prediction in MVDC Distribution Systems

Asset management for MVDC distribution systems faces two obstacles: the modular multilevel converter couples hundreds of components so that one asset's aging accelerates its neighbors', and no operating MVDC system has failure records to learn this coupling from. This paper answers both with a data representation. First, an MVDC station is modeled as an attributed graph whose nine edge types are split by a FMEA into an operational and a degradation topology; in the KEPCO Naju ±35 kV, 30 MW testbed, 358 of 417 edges are degradation edges. Second, a synthetic dataset is built from published failure rates by assigning per-type Weibull parameters and modulating each asset's characteristic life along the adjudicated paths; generating parameters are withheld from the inputs. Using a graph model as a validation tool over 12 seeds, the 358 degradation edges alone match the full graph, whereas the 59 operational edges alone degrade error by 38.7%, no better than no edges.