David A. Howey

Masaki Adachi, Brady Planden, David A. Howey et al. · oxford

Like many optimizers, Bayesian optimization often falls short of gaining user trust due to opacity. While attempts have been made to develop human-centric optimizers, they typically assume user knowledge is well-specified and error-free, employing users mainly as supervisors of the optimization process. We relax these assumptions and propose a more balanced human-AI partnership with our Collaborative and Explainable Bayesian Optimization (CoExBO) framework. Instead of explicitly requiring a user to provide a knowledge model, CoExBO employs preference learning to seamlessly integrate human insights into the optimization, resulting in algorithmic suggestions that resonate with user preference. CoExBO explains its candidate selection every iteration to foster trust, empowering users with a clearer grasp of the optimization. Furthermore, CoExBO offers a no-harm guarantee, allowing users to make mistakes; even with extreme adversarial interventions, the algorithm converges asymptotically to a vanilla Bayesian optimization. We validate CoExBO's efficacy through human-AI teaming experiments in lithium-ion battery design, highlighting substantial improvements over conventional methods. Code is available https://github.com/ma921/CoExBO.

Adrien M. Bizeray, Jin-Ho Kim, Stephen R. Duncan et al.

This paper investigates the identifiability and estimation of the parameters of the single particle model (SPM) for lithium-ion battery simulation. Identifiability is addressed both in principle and in practice. The approach begins by grouping parameters and partially non-dimensionalising the SPM to determine the maximum expected degrees of freedom in the problem. We discover that, excluding open circuit voltage, there are only six independent parameters. We then examine the structural identifiability by considering whether the transfer function of the linearised SPM is unique. It is found that the model is unique provided that the electrode open circuit voltage functions have a known non-zero gradient, the parameters are ordered, and the electrode kinetics are lumped into a single charge transfer resistance parameter. We then demonstrate the practical estimation of model parameters from measured frequency-domain experimental electrochemical impedance spectroscopy (EIS) data, and show additionally that the parametrised model provides good predictive capabilities in the time domain, exhibiting a maximum voltage error of 20 mV between model and experiment over a 10 minute dynamic discharge.

Robert R. Richardson, David A. Howey

This study presents a method of estimating battery cell core and surface temperature using a thermal model coupled with electrical impedance measurement, rather than using direct surface temperature measurements. This is advantageous over previous methods of estimating temperature from impedance, which only estimate the average internal temperature. The performance of the method is demonstrated experimentally on a 2.3 Ah lithium-ion iron phosphate cell fitted with surface and core thermocouples for validation. An extended Kalman filter, consisting of a reduced order thermal model coupled with current, voltage and impedance measurements, is shown to accurately predict core and surface temperatures for a current excitation profile based on a vehicle drive cycle. A dual extended Kalman filter (DEKF) based on the same thermal model and impedance measurement input is capable of estimating the convection coefficient at the cell surface when the latter is unknown. The performance of the DEKF using impedance as the measurement input is comparable to an equivalent dual Kalman filter using a conventional surface temperature sensor as measurement input.

Jorn M. Reniers, Grietus Mulder, Sina Ober-Blobaum et al.

The increased deployment of intermittent renewable energy generators opens up opportunities for grid-connected energy storage. Batteries offer significant flexibility but are relatively expensive at present. Battery lifetime is a key factor in the business case, and it depends on usage, but most techno-economic analyses do not account for this. For the first time, this paper quantifies the annual benefits of grid-connected batteries including realistic physical dynamics and nonlinear electrochemical degradation. Three lithium-ion battery models of increasing realism are formulated, and the predicted degradation of each is compared with a large-scale experimental degradation data set (Mat4Bat). A respective improvement in RMS capacity prediction error from 11\% to 5\% is found by increasing the model accuracy. The three models are then used within an optimal control algorithm to perform price arbitrage over one year, including degradation. Results show that the revenue can be increased substantially while degradation can be reduced by using more realistic models. The estimated best case profit using a sophisticated model is a 175% improvement compared with the simplest model. This illustrates that using a simplistic battery model in a techno-economic assessment of grid-connected batteries might substantially underestimate the business case and lead to erroneous conclusions.

Shi Zhao, Stephen R. Duncan, David A. Howey

This paper investigates the observability of one of the most commonly used equivalent circuit models (ECMs) for lithium-ion batteries and presents a method to estimate the state of charge (SOC) in the presence of sensor biases, highlighting the importance of observability analysis for choosing appropriate state estimation algorithms. Using a differential geometric approach, necessary and sufficient conditions for the nonlinear ECM to be observable are derived and are shown to be different from the conditions for the observability of the linearised model. It is then demonstrated that biases in the measurements, due to sensor ageing or calibration errors, can be estimated by applying a nonlinear Kalman filter to an augmented model where the biases are incorporated into the state vector. Experiments are carried out on a lithium-ion pouch cell and three types of nonlinear filters, the first-order extended Kalman filter (EKF), the second-order EKF and the unscented Kalman filter (UKF) are applied using experimental data. The different performances of the filters are explained from the point of view of observability.

Robert R. Richardson, Shi Zhao, David A. Howey

Impedance-based temperature detection (ITD) is a promising approach for rapid estimation of internal cell temperature based on the correlation between temperature and electrochemical impedance. Previously, ITD was used as part of an Extended Kalman Filter (EKF) state-estimator in conjunction with a thermal model to enable estimation of the 1-D temperature distribution of a cylindrical lithium-ion battery. Here, we extend this method to enable estimation of the 2-D temperature field of a battery with temperature gradients in both the radial and axial directions. An EKF using a parameterised 2-D spectral-Galerkin model with ITD measurement input (the imaginary part of the impedance at 215 Hz) is shown to accurately predict the core temperature and multiple surface temperatures of a 32113 LiFePO$_4$ cell, using current excitation profiles based on an Artemis HEV drive cycle. The method is validated experimentally on a cell fitted with a heat sink and asymmetrically cooled via forced air convection. A novel approach to impedance-temperature calibration is also presented, which uses data from a single drive cycle, rather than measurements at multiple uniform cell temperatures as in previous studies. This greatly reduces the time required for calibration, since it overcomes the need for repeated cell thermal equalization.

Robert R. Richardson, Shi Zhao, David A. Howey

Estimating the temperature distribution within Li-ion batteries during operation is critical for safety and control purposes. Although existing control-oriented thermal models - such as thermal equivalent circuits (TEC) - are computationally efficient, they only predict average temperatures, and are unable to predict the spatially resolved temperature distribution throughout the cell. We present a low-order 2D thermal model of a cylindrical battery based on a Chebyshev spectral-Galerkin (SG) method, capable of predicting the full temperature distribution with a similar efficiency to a TEC. The model accounts for transient heat generation, anisotropic heat conduction, and non-homogeneous convection boundary conditions. The accuracy of the model is validated through comparison with finite element simulations, which show that the 2-D temperature field (r, z) of a large format (64 mm diameter) cell can be accurately modelled with as few as 4 states. Furthermore, the performance of the model for a range of Biot numbers is investigated via frequency analysis. For larger cells or highly transient thermal dynamics, the model order can be increased for improved accuracy. The incorporation of this model in a state estimation scheme with experimental validation against thermocouple measurements is presented in the companion contribution (Part II).

Antti Aitio, Dominik Jöst, Dirk Uwe Sauer et al.

Estimating state of health is a critical function of a battery management system but remains challenging due to the variability of operating conditions and usage requirements of real applications. As a result, techniques based on fitting equivalent circuit models may exhibit inaccuracy at extremes of performance and over long-term ageing, or instability of parameter estimates. Pure data-driven techniques, on the other hand, suffer from lack of generality beyond their training dataset. In this paper, we propose a hybrid approach combining data- and model-driven techniques for battery health estimation. Specifically, we demonstrate a Bayesian data-driven method, Gaussian process regression, to estimate model parameters as functions of states, operating conditions, and lifetime. Computational efficiency is ensured through a recursive approach yielding a unified joint state-parameter estimator that learns parameter dynamics from data and is robust to gaps and varying operating conditions. Results show the efficacy of the method, on both simulated and measured data, including accurate estimates and forecasts of battery capacity and internal resistance. This opens up new opportunities to understand battery ageing in real applications.

Ross Drummond, Shi Zhao, David A. Howey et al.

This paper is concerned with the synthesis of RC electrical circuits from physics-based supercapacitor models describing conservation and diffusion relationships. The proposed synthesis procedure uses model discretisation, linearisation, balanced model order reduction and passive network synthesis to form the circuits. Circuits with different topologies are synthesized from several physical models. This work will give greater understanding to the physical interpretation of electrical circuits and will enable the development of more generalised circuits, since the synthesized impedance functions are generated by considering the physics, not from experimental fitting which may ignore certain dynamics.

Shi Zhao, David A. Howey

This paper considers one of the most commonly used equivalent circuit models (ECMs) for lithium-ion batteries and investigates the sensitivity of the model output to changes of model parameters using the Morris method. Experiments are carried out on a lithium-ion cell with nickel manganese cobalt oxide (NMC) electrode and parameters of the model are identified in the state of charge (SOC) range [100%,10%]. Although all the model parameters do vary with SOC, global sensitivity analysis reveals that the uncertainties of some of the parameters generate very little uncertainty in the voltage output, implying that those parameters can be taken as constants without compromising the accuracy of the model. This is further confirmed by experimental validation.

Masaki Adachi, Yannick Kuhn, Birger Horstmann et al.

A wide variety of battery models are available, and it is not always obvious which model `best' describes a dataset. This paper presents a Bayesian model selection approach using Bayesian quadrature. The model evidence is adopted as the selection metric, choosing the simplest model that describes the data, in the spirit of Occam's razor. However, estimating this requires integral computations over parameter space, which is usually prohibitively expensive. Bayesian quadrature offers sample-efficient integration via model-based inference that minimises the number of battery model evaluations. The posterior distribution of model parameters can also be inferred as a byproduct without further computation. Here, the simplest lithium-ion battery models, equivalent circuit models, were used to analyse the sensitivity of the selection criterion to given different datasets and model configurations. We show that popular model selection criteria, such as root-mean-square error and Bayesian information criterion, can fail to select a parsimonious model in the case of a multimodal posterior. The model evidence can spot the optimal model in such cases, simultaneously providing the variance of the evidence inference itself as an indication of confidence. We also show that Bayesian quadrature can compute the evidence faster than popular Monte Carlo based solvers.

Zihao Zhou, David A. Howey

Accurate prediction of battery health is essential for real-world system management and lab-based experiment design. However, building a life-prediction model from different cycling conditions is still a challenge. Large lifetime variability results from both cycling conditions and initial manufacturing variability, and this -- along with the limited experimental resources usually available for each cycling condition -- makes data-driven lifetime prediction challenging. Here, a hierarchical Bayesian linear model is proposed for battery life prediction, combining both individual cell features (reflecting manufacturing variability) with population-wide features (reflecting the impact of cycling conditions on the population average). The individual features were collected from the first 100 cycles of data, which is around 5-10% of lifetime. The model is able to predict end of life with a root mean square error of 3.2 days and mean absolute percentage error of 8.6%, measured through 5-fold cross-validation, overperforming the baseline (non-hierarchical) model by around 12-13%.

Volkan Kumtepeli, Rebecca Perriment, David A. Howey

We present an approach for computationally efficient dynamic time warping (DTW) and clustering of time-series data. The method frames the dynamic warping of time series datasets as an optimisation problem solved using dynamic programming, and then clusters time series data by solving a second optimisation problem using mixed-integer programming (MIP). There is also an option to use k-medoids clustering for increased speed, when a certificate for global optimality is not essential. The improved efficiency of our approach is due to task-level parallelisation of the clustering alongside DTW. Our approach was tested using the UCR Time Series Archive, and was found to be, on average, 33% faster than the next fastest option when using the same clustering method. This increases to 64% faster when considering only larger datasets (with more than 1000 time series). The MIP clustering is most effective on small numbers of longer time series, because the DTW computation is faster than other approaches, but the clustering problem becomes increasingly computationally expensive as the number of time series to be clustered increases.

Robert R. Richardson, Michael A. Osborne, David A. Howey

Accurately predicting the future health of batteries is necessary to ensure reliable operation, minimise maintenance costs, and calculate the value of energy storage investments. The complex nature of degradation renders data-driven approaches a promising alternative to mechanistic modelling. This study predicts the changes in battery capacity over time using a Bayesian non-parametric approach based on Gaussian process regression. These changes can be integrated against an arbitrary input sequence to predict capacity fade in a variety of usage scenarios, forming a generalised health model. The approach naturally incorporates varying current, voltage and temperature inputs, crucial for enabling real world application. A key innovation is the feature selection step, where arbitrary length current, voltage and temperature measurement vectors are mapped to fixed size feature vectors, enabling them to be efficiently used as exogenous variables. The approach is demonstrated on the open-source NASA Randomised Battery Usage Dataset, with data of 26 cells aged under randomized operational conditions. Using half of the cells for training, and half for validation, the method is shown to accurately predict non-linear capacity fade, with a best case normalised root mean square error of 4.3%, including accurate estimation of prediction uncertainty.

Antti Aitio, David A. Howey

Hundreds of millions of people lack access to electricity. Decentralised solar-battery systems are key for addressing this whilst avoiding carbon emissions and air pollution, but are hindered by relatively high costs and rural locations that inhibit timely preventative maintenance. Accurate diagnosis of battery health and prediction of end of life from operational data improves user experience and reduces costs. But lack of controlled validation tests and variable data quality mean existing lab-based techniques fail to work. We apply a scaleable probabilistic machine learning approach to diagnose health in 1027 solar-connected lead-acid batteries, each running for 400-760 days, totalling 620 million data rows. We demonstrate 73% accurate prediction of end of life, eight weeks in advance, rising to 82% at the point of failure. This work highlights the opportunity to estimate health from existing measurements using `big data' techniques, without additional equipment, extending lifetime and improving performance in real-world applications.

Samuel Greenbank, David A. Howey

The complex nature of lithium-ion battery degradation has led to many machine learning based approaches to health forecasting being proposed in literature. However, machine learning can be computationally intensive. Linear approaches are faster but have previously been too inflexible for successful prognosis. For both techniques, the choice and quality of the inputs is a limiting factor of performance. Piecewise-linear models, combined with automated feature selection, offer a fast and flexible alternative without being as computationally intensive as machine learning. Here, a piecewise-linear approach to battery health forecasting was compared to a Gaussian process regression tool and found to perform equally well. The input feature selection process demonstrated the benefit of limiting the correlation between inputs. Further trials found that the piecewise-linear approach was robust to changing input size and availability of training data.

Robert R. Richardson, Michael A. Osborne, David A. Howey

Accurately predicting the future capacity and remaining useful life of batteries is necessary to ensure reliable system operation and to minimise maintenance costs. The complex nature of battery degradation has meant that mechanistic modelling of capacity fade has thus far remained intractable; however, with the advent of cloud-connected devices, data from cells in various applications is becoming increasingly available, and the feasibility of data-driven methods for battery prognostics is increasing. Here we propose Gaussian process (GP) regression for forecasting battery state of health, and highlight various advantages of GPs over other data-driven and mechanistic approaches. GPs are a type of Bayesian non-parametric method, and hence can model complex systems whilst handling uncertainty in a principled manner. Prior information can be exploited by GPs in a variety of ways: explicit mean functions can be used if the functional form of the underlying degradation model is available, and multiple-output GPs can effectively exploit correlations between data from different cells. We demonstrate the predictive capability of GPs for short-term and long-term (remaining useful life) forecasting on a selection of capacity vs. cycle datasets from lithium-ion cells.

Ross Drummond, David A. Howey, Stephen R. Duncan

This work investigates two physics-based models that simulate the non-linear partial differential algebraic equations describing an electric double layer supercapacitor. In one model the linear dependence between electrolyte concentration and conductivity is accounted for, while in the other model it is not. A spectral element method is used to discretise the model equations and it is found that the error convergence rate with respect to the number of elements is faster compared to a finite difference method. The increased accuracy of the spectral element approach means that, for a similar level of solution accuracy, the model simulation computing time is approximately 50% of that of the finite difference method. This suggests that the spectral element model could be used for control and state estimation purposes. For a typical supercapacitor charging profile, the numerical solutions from both models closely match experimental voltage and current data. However, when the electrolyte is dilute or where there is a long charging time, a noticeable difference between the numerical solutions of the two models is observed. Electrical impedance spectroscopy simulations show that the capacitance of the two models rapidly decreases when the frequency of the perturbation current exceeds an upper threshold.

Christoph R. Birkl, Damien F. Frost, Adrien M. Bizeray et al.

Lithium ion batteries are promising for small off- grid energy storage applications in developing countries because of their high energy density and long life. However, costs are prohibitive. Instead, we consider 'used' Li-ion batteries for this application, finding experimentally that many discarded laptop cells, for example, still have good capacity and cycle life. In order to make safe and optimal use of such cells, we present a modular power management system using a separate power converter for every cell. This novel approach allows individual batteries to be used to their full capacity. The power converters operate in voltage droop control mode to provide easy charge balancing and implement a battery management system to estimate the capacity of each cell, as we demonstrate experimentally.