Digital diagnostics of internal damage in electrical machines based on adaptive identification and decomposition of magnetic field signals

Introduction. Existing industrial electrical machines, being key components of energy systems, are susceptible to internal damages such as turn short circuits, insulation degradation, magnetic field imbalance, and mechanical wear. Traditional diagnostic methods based on vibration analysis, thermal characteristics, or static electrical parameters are often insufficiently sensitive for early detection of hidden defects, particularly under non-stationary operating conditions and external interference. In recent years, the focus has shifted toward intelligent monitoring systems that utilize digital signal processing and machine learning methods. Given the limitations associated with non-parametric prior uncertainty, adaptive identification algorithms are of particular interest, as they can dynamically adjust to changing machine operating conditions and integrate additional prior information such as raw data, expert assessments, and physical models of processes. Goal. The article aims to develop methods of adaptive identification and signal decomposition for early detection of turn-to-turn short circuits in the rotor winding of synchronous generators, ensuring enhanced diagnostic accuracy and prevention of emergency operating conditions. Materials and methods. The study is based on the analysis of signals from leakage magnetic field sensors, acquired during various operating modes of the synchronous generator (no-load, 25–100 % load) and under simulated turn-to-turn short circuits (1.2–17.2 % of winding turns). Adaptive identification algorithms, signal decomposition, and data processing via a 12-bit ADC (sampling frequency: 10 kHz) were employed for the analysis. Results and discussion. The use of data from magnetometric sensors and signal decomposition has enabled achieving a damage detection threshold of 2.7 % of the total winding turns, confirming the practical relevance of the approach. The decomposition coefficients for positive and negative half-waves are compared using an integral asymmetry criterion. Experiments demonstrated that when short-circuiting 4 % of the turns, the criterion value increases from 1.3 % to 3.7 %, and for 17.2 % of the turns, it rises to 13.5 %. Conclusion. The combination of adaptive algorithms and a priori information creates a powerful tool for solving diverse tasks that demand flexibility and precision. A priori information acts as a conceptual core, integrating expert knowledge and data, which enables adaptive algorithms to become more intelligent and efficient.