Signal Processing Techniques in Condition Monitoring and Control of Power Electronics Converters and Battery System

There are several fundamental challenges in condition monitoring, fault detection, and control of power electronics converters and Li-Ion batteries. These are linked to the computation complexity of the signal processing algorithms, suitability for on-line and real-time implementation, cost of implementation, ability to deal with rapid real-time changes, etc.

Advanced signal processing algorithms have been found to provide superior condition monitoring, system identification, and state estimation solutions to tackle the aforementioned issues with fast dynamic performance, cost-effective solution, high accuracy prediction, good tracking ability to system changes.     

This talk offers basic theory and some recent advances of innovative signal processing algorithms for condition monitoring and control of power converters and batteries. The talk covers different system identification and parameter estimation of power converters, adaptive signal processing algorithms for on-line modelling and estimation of power converters, parametric/ non-parametric system identification, remaining useful prediction, and sate of health estimation of Lithium-Ion batteries. This talk is intended for researchers and engineers in the area of signal processing and artificial intelligence to explore new applications of signal processing algorithms in power electronics applications and battery management systems, and for postgraduate students in these fields.

What this presentation is about and why it matters

This talk surveys practical signal‑processing techniques used for condition monitoring, fault detection, and control of switched‑mode power converters and lithium‑ion battery systems. The presenter shows how modern algorithms — adaptive filters, parametric and non‑parametric identification, time–frequency analysis, and particle filters — can be used to estimate circuit parameters, detect wiring or component faults, track battery state‑of‑health (SoH) and state‑of‑charge (SoC), and predict remaining useful life (RUL).

Why this matters: power converters and batteries are everywhere (from mobile chargers to grid converters). Failures, aging, and load changes affect performance and reliability. Signal‑processing methods make it possible to monitor systems online, adapt controllers automatically, reduce downtime through predictive maintenance, and implement embedded real‑time solutions on DSPs, microcontrollers and FPGAs. For engineers, understanding how these techniques map to practical hardware and real data is essential for designing robust systems that operate safely and efficiently.

Who will benefit the most from this presentation

• Signal‑processing engineers interested in power electronics applications.
• Control and embedded systems engineers who implement real‑time monitoring and adaptive control on DSPs or microcontrollers.
• Researchers and postgraduate students working on system identification, battery management systems, or predictive maintenance.
• Practicing engineers who want practical, lower‑complexity algorithms suitable for high switching frequencies and constrained hardware.

What you need to know

To get the most from the talk, refresh the following concepts and tools. The presenter keeps the focus practical, but knowing these basics will help you follow the examples and appreciate tradeoffs.

• Linear system models and the z‑domain: many converter models are linearized and represented by a transfer function in discrete time, e.g. $H(z)=\frac{Y(z)}{U(z)}$. A second‑order digital model often has four coefficients (two numerator, two denominator) to be estimated.
• Parametric vs non‑parametric identification: parametric methods estimate coefficients of a chosen model structure; non‑parametric methods estimate input–output behaviour (e.g. frequency response) without committing to a model form.
• Adaptive filters and RLS: Recursive Least Squares (RLS) is a common adaptive algorithm for fast convergence in parameter estimation. The talk demonstrates a one‑tap prediction‑error filter tied to control action for fast transient tracking.
• Prediction‑error filters and PID linkage: a prediction‑error filter produces an error signal that can be used to adapt control gains. The presenter relates one‑step prediction filters to proportional/derivative control behaviour.
• Frequency‑domain tools and FFT: non‑parametric monitoring can use swept‑sine excitation and FFT to estimate frequency responses; comparisons between healthy/faulty responses support classification of faults like lifted wires.
• Time–frequency analysis and wavelets (CWT): continuous wavelet transforms create time–frequency images of battery signals (voltage, current, temperature) that reveal degradation patterns used for SoH/SoC classification.
• Particle filters for RUL: sequential Monte Carlo methods (particle filters) propagate probabilistic estimates of capacity and RUL. A smooth particle filter variant is used in the talk to improve RUL prediction.
• Practical issues: sampling, quantization, excitation design (adding controlled noise for identification), computational cost, and on‑board implementation constraints.

Glossary

• Adaptive filter — algorithm that updates its coefficients online to track changing system dynamics.
• Parametric system identification — estimating numerical parameters of a predefined model structure (e.g., ARX, transfer function).
• Non‑parametric identification — estimating input–output behaviour (e.g., frequency response) without assuming a specific model form.
• RLS (Recursive Least Squares) — an adaptive estimation method with fast convergence used to update model coefficients.
• Prediction‑error filter — a filter that predicts the next sample and produces an error used for adaptation or control.
• FFT (Fast Fourier Transform) — algorithm to compute the discrete Fourier transform, used for frequency‑response estimation.
• CWT (Continuous Wavelet Transform) — time‑frequency transform that produces a localized spectrum useful for transient and degradation features.
• SoC / SoH — state of charge (available energy) and state of health (degree of degradation) for a battery.
• RUL (Remaining Useful Life) — predicted remaining number of cycles or time until the battery falls below a usable capacity threshold.
• Particle filter — sequential Monte Carlo method for Bayesian state estimation useful in non‑linear, non‑Gaussian tracking (used here for RUL).

Final notes — why watch this talk

This presentation connects solid signal‑processing theory with hands‑on examples: DSP implementation of system identification for a buck converter, a low‑complexity RLS‑based regulator tuned for high switching frequencies, FFT‑based fault detection on wiring, and wavelet + particle‑filter methods for battery RUL. The speaker balances practical implementation concerns (computation, sampling, excitation) with algorithmic choices, which is especially valuable for engineers who must deploy real‑time monitoring and control on constrained hardware. If you work on converters, BMS, or tidy, practical adaptive algorithms, you will find the examples and the presenter's pragmatic perspective both useful and motivating.

This overview is AI-generated from the session transcript. Spot an issue? Let us know.