Scaling of the Method of Joint Physical and Intellectual Protection to on-Board Systems of Locomotives

This paper substantiates scaling the combined physical-intelligent protection method, previously developed for stationary microprocessor interlocking cabinets, to onboard locomotive systems operating under fully autonomous (GoA3/GoA4) control. Vulnerability of critical locomotive subsystems to destructive electromagnetic influences is analyzed across three physical paths: the galvanic track circuit of ALSN receiver coils and Eurobalise readers, the airborne radio-channel path of FRMCS/5G-Rail and satellite navigation antennas, and internal locomotive CAN-bus/MVB/Ethernet data buses exposed to ultra-wideband transient pulses. A hybrid onboard module architecture is proposed, combining a passive multilayer shielded enclosure with vibration damping and an active Edge AI computational core acting as a transparent gateway between antenna inputs and the onboard safety computer. A dedicated mathematical apparatus and AI logic are developed for onboard operation under motion dynamics: dataset engineering accounting for the Doppler frequency shift and pantograph arcing transients; an adaptive Short-Time Fourier Transform with a Gaussian window depending on locomotive speed; INT8 quantization of a lightweight depthwise-separable convolutional network for fixed-point inference; and a speed-dependent Softmax threshold that tightens automatically at higher speeds to minimize reaction delay. A Hardware-in-the-Loop test stand combining an anechoic chamber, UWB/HPM generators, and a vibration bench determined reversible bit-flip and irreversible burnout thresholds, confirming the effectiveness of nanosecond active protective switches triggered by Edge AI, and a two-tier architecture integrating onboard modules with centralized railway cyber-monitoring for EW threat mapping is outlined.