REPOSITORY > RESULTS

Abstract

According to the EU’s Green Deal plan, one of the main objectives for the EU is to
become a net-zero greenhouse emission society by 2050. Achieving this goal requires advancements
across many areas ranging from energy production, conversion, distribution
and energy storage where the key role is taken by the electrochemical energy conversion
devices (EECD). EECD is the umbrella notion for the family of the devices such as fuel
cells, batteries, electrolysers, supercapacitors and hydrogen pumps. They share common
fundamentals of electrochemical thermodynamics and kinetics but differ in the design and
operational requirements since neither batteries, fuel cells nor other EECD can serve all
applications alone. In the very nature of the EECD are complex electrochemical processes
that are fundamental for their operation. On the other hand, they are the main source of
the degradation that significantly affects their reliability and durability. Therefore, online
health monitoring is being percepted as an indispensable integral part of the systems with
EECD, which is expected to elevate their market deployment and economic exploitation.
Current state-of-the-art health monitoring approaches strongly build on electrochemical
impedance spectroscopy (EIS). EIS contains the fingerprint of the dynamics of the EECD,
which depends on physical and chemical processes inherent to the device. By definition, EIS
is performed by perturbing EECD with mono-component, low-amplitude sinusoidal signals
with the predefined frequencies. To ensure confident characterisation, the system under
test has to comply with the conditions of stationarity, linearity and causality. However,
as the perturbation times increase, load and environmental disturbances are more likely
to cause distorting effects on the evaluation results, especially in the low-frequency region
below 100 mHz. A way to alleviate the limitations of the conventional sine-based EIS is to
employ broadband switching perturbation like discrete random binary sequence (DRBS).
In that case all the frequencies from a wide band are excited at the same time. The
approach is referred to as fast EIS method.
The aim of this dissertation is to revisit fast EIS and propose upgrade that will render
the approach applicable for cost-effective in-field monitoring. The main contributions of
the thesis are threefold.
First, the accuracy of the evaluated impedance spectrum is improved by applying continuous
wavelet transform (CWT) with Morse mother wavelet. It is shown that at ultra-low
frequencies (< 1 mHz) Morse wavelet remains analytical, i.e. completely concentrated at
positive frequencies despite its dispersion. That is essential for correct spectral decomposition
and, consequently, correct impedance estimation. The said properties turned essential
for accurate estimation of the state-of-health (SoH) and state-of-charge (SOC) of Li-ion
batteries.
Second, a numerically efficient approach for the spectral deconvolution by means of the
equivalent circuit model (ECM) and distribution of relaxation times (DRT) is proposed.
By detecting changes in the resulting model parameters, one can track changes in the
impedance spectrum and possibly identify the degradation mode. The demonstration on
a case study with occasional fuel supply deficiency during solid oxide fuel cell (SOFC)
operation is performed.
Third, a concept of modular HW setup for performing EIS not only in the laboratory
but also in the in-field applications is proposed. It relies on using commercially available
low-cost components. The setup is rather general and applicable to a broad range of
EECD. It enables the use of arbitrary excitation signals and provides great flexibility both
in terms of power ranges and modes of operation.
Finally, the effectiveness of HW and algorithmic solutions is demonstrated through
numerous accelerated and long-term experiments performed on various electrochemical
energy devices, such as single cell SOFC, SOFC stack, single cell SOE, SOE stack, electrochemical
hydrogen compressor, Li-ion battery, etc. During these experiments more than 5
TB of data was generated. Algorithms are implemented in a free open source format and
made publicly available.