Real-time output capacitor health monitoring in a DC-DC buck converter: automatic tuning and mode-change discrimination

DC–DC buck converters are widely used in automotive systems to derive low-voltage rails, communication interfaces and sensor networks from the vehicle battery. As their number per vehicle increases, systematic reliability monitoring becomes increasingly important. In the output LC filter, the capacitor is especially susceptible to degradation, which alters the spectral content of the output-voltage ripple. Building on prior Infineon research, this thesis develops a non-invasive, real-time algorithm for online monitoring and assessment of the output-capacitor health in an automotive buck converter. The work proceeded in three phases. First, the detection scheme inherited from earlier theses was redesigned into a single-FFT architecture and then synthesized and implemented on a Zynq FPGA using VHDL and SystemVerilog. Second, a dedicated PCB demonstrator—integrating the buck converter, the analog signal-conditioning chain and ADC interface—was tested and debugged. This revealed limitations that made the demonstrator unsuitable as the only quantitative validation source. Third, the detection algorithm was reformulated as an independent two-state Kalman filter that tracks both the dominant spectral component of the output voltage ripple and the switching frequency, derived from a sliding-window FFT of the output voltage. Anomalies are detected via the Normalized Innovation Squared (NIS) metric and confirmed by a load-aware step-detection mechanism that discriminates between load transients and actual capacitance degradation. The algorithm supports PWM-CCM, PWM-DCM and PFM-DCM modes, reducing false alarms during mode transitions. Process and measurement noise covariance matrices are automatically identified and calibrated during an initial fault-free burn-in period and then kept constant, eliminating manual parameter tuning. The method was validated using SIMetrix/SIMPLIS simulations under varying load and input-voltage conditions and cross-verified with oscilloscope measurements from the PCB demonstrator. The algorithm correctly classified all SIMPLIS test cases, covering nominal operation, progressive capacitance degradation, input-voltage variation and combined load and fault conditions. Correct healthy-state operation was additionally confirmed on the real PCB oscilloscope acquisitions at different steady-state load levels.