Purpose-The paper presents a new approach for safe operation, and maintenance cost reduction, regarding electrical and electromechanical systems of power production, power conversion and power transmission, primarily in industrial units. Design/methodology/approach-The paper adapts a theoretical approach to real-time monitoring of pulse energy conversion systems (PECSs), and prediction of abnormal dynamics incipient and developing failure. The approach utilizes the preliminary bifurcation analysis results and the geometrical interpretation of the fractal regularities in PECS dynamics, to reveal degradation development. Findings-It turns out that this new approach enables to fill the joint requirements of real-time failure prediction of the high frequency power control devices, and of the relating failure symptoms to cause parameters. Discussions are led on the fundamental outcomes of numerical and experimental investigations of a DC-DC buck voltage converter with pulse-width-modulation (PWM) control. Originality/value-About «failure prediction», we propose to pay attention, not to identification of the dynamics evolution specific reason, but real-time monitoring of this reason consequenceincipient abnormal dynamics in the electrical and electromechanical systems-, that can lead to failure. The advantage of this approach consists in the possibility of taking into account PECS operating conditions of models ambiguity of both disturbing parameter changes and PECS behavior. Practical implication-The real-time monitoring of incipient abnormal dynamics in key nonlinear devices of electrical and electromechanical systems constitutes a mean to predict and prevent failures. It provides invaluable information for deciding and planning predictive maintenance actions, from the insurance of optimal operating conditions until abnormal operating prevention, either by means of modification of controlled parameters and control laws, or in the worst case, by change of the power components' structure.