Early Fault Warning and Identification of Wind Turbines Based on Three-Dimensional Visualization of Dynamic Network Marker

Due to differences in design, manufacturing, installation, commissioning, operating conditions, and working environments, degradation of wind turbines (WTs) exhibits significant individual heterogeneity. This limits the generalization capability of machine-learning-based fault early warning models. To address this problem, this paper proposes an early fault warning and identification method for WTs based on Three-Dimensional Visualization of Dynamical Network Marker (3DV-DNM). Firstly, monitored variables from the SCADA system of the target turbine are mapped to network nodes to construct an observable network representing the turbine state. Then, Pearson correlation features between nodes and fluctuation features across adjacent observation periods are computed to build a differential network that reflects the dynamic characteristics of the turbine state. Furthermore, 3DV-DNM is employed to construct a comprehensive state index, which quantifies the evolution of the turbine state and identifies the critical state and key factors before fault occurrence. The proposed method is validated using SCADA data from actual WT operation. The results show that the method effectively provides early fault warning by detecting abrupt changes in the comprehensive state index and accurately identifies fault types based on key nodes in abnormal time windows.