1,720,967 research outputs found
Recommended from our members
Quantifying and Mitigating the Influence of System Uncertainties in Lithium-Ion Battery Electrochemical Modeling and Estimation through Data Structure Analysis and Machine Learning
Get PDFThe widespread adoption of lithium-ion batteries is driving the concurrent development of advanced battery management systems, which seek to maximize safety and performance through state-of-the-art diagnostic, prognostic, and control techniques. These capabilities rely upon physics-based electrochemical battery models, which must provide accurate predictions of output voltage and internal states through a physically-meaningful parameterization. However, electrochemical battery models are highly susceptible to system uncertainties (e.g., unmodeled dynamics, sensor bias/noise, parameter errors), which skew the physical significance (and thus the utility) of the outputs. This work takes a four-pronged approach to systematically quantify and mitigate the influence of system uncertainties in battery electrochemical modeling and estimation.
The first approach addresses the fundamental question—why is a parameter estimate accurate or inaccurate?—through the derivation and validation of a generalized multivariate estimation error formula for the ubiquitous least squares objective. This formula reveals the specific connections between the conventional error analysis criteria (i.e., parameter sensitivity, Fisher information, and the Cramér-Rao bound) and estimation accuracy, and highlights the limitations of each. The most important insight is that specific sensitivity-based data structures propagate system uncertainties to the estimation result, which may attenuate, amplify, or even (partially) cancel their effects.
These uncertainty-propagating data structures are then leveraged in the second approach of this work: the development of a data selection framework to enhance parameter estimation robustness to uncertainties under operational data. While random operational data are often used to estimate important electrochemical parameters, the presence of low-quality data segments (i.e., with low sensitivity and high uncertainty) can significantly degrade estimation accuracy. The data selection framework evaluates the potential for various data segments to mitigate/amplify the influence of system uncertainties, and implements the most favorable segments in the estimation. In a validation study, the framework reduced experimental estimation errors by one order of magnitude when compared with the conventional approach of using the full data set.
The third approach directly addresses the influence of system uncertainties on model output prediction accuracy through the development and validation of a hybrid physics-based and machine learning model. Specifically, a reduced-order electrochemical battery model is augmented with a Gaussian process regression residual model, which compensates for the output prediction errors caused by uncertainties. A key feature of the hybrid framework is the proposed data sampling procedure, which significantly reduces the computational expense to an amply sufficient rate for online BMS applications.
The final approach is the development of a parameter estimation framework for the hybrid model formulated in the third approach, in response to the parameter identification challenges associated with the interdependencies between the physical electrochemical parameters, machine learning (hyper)parameters, and system uncertainties. While traditional identification procedures estimate the physical parameters in the presence of system uncertainties (despite the high risk of error) and then tune the machine learning (hyper)parameters to predict the output residual, the proposed framework jointly estimates the physical and machine learning (hyper)parameters through direct consideration of the system uncertainties that connect them. Specifically, the system uncertainties are incorporated into the estimation procedure through the Gaussian process regression residual model, which is characterized by input signals that depend on the physical parameters. During validation, the framework yielded estimation errors that are one order of magnitude smaller than those of the conventional least squares approach. In addition, the parameterization under the new approach consistently delivered smaller output prediction errors than that of the least squares approach, attesting to its effectiveness. Finally, while developed for hybrid models, the framework is presented in a generalized form that is applicable to all models, with special relevance to those that are susceptible to large uncertainties
Recommended from our members
Data Optimization for Identification of Lithium Ion Battery Electrochemical Parameters
Get PDFEstimation of model parameters is a critical topic in battery modeling research, as the accuracy of parameters determines the efficacy of the widely used model-based battery engineering. The measurement-based approaches aim at measuring the battery physical parameters using advanced experiment and instrumentation techniques, but the associated complexity, time, and cost make it undesirable in many cases. Therefore, the data-based identification approach, which only uses easily available input and output measurement data, has been widely adopted due to its convenience and noninvasiveness. As the quality of data has significant impact on the estimation accuracy, data optimization, or optimal experiment design, is often utilized to improve and guarantee the accuracy of estimation. The common practice of data optimization aims at designing input excitation by maximizing a certain conventional criterion, e.g. Fisher information which measures the information content of the data and relates to the variance of the estimation error. However, such approach suffers from fundamental limitations, including inability to explicitly address estimation bias and system uncertainties in measurement, model, and parameter, which severely restrict the applicability and effectiveness of the method in practice. To overcome the existing limitations, new criteria and a novel framework are proposed in this research for estimation error quantification and data optimization. A generic formula is first derived for quantifying the estimation error subject to sensor, model, and parameter uncertainties for the commonly used least-squares algorithm. Based on the formula, data structures, represented in terms of parameter sensitivity, which could minimize the estimation errors caused by each type of uncertainty are identified. These data structures are then employed as new criteria to supplement the Fisher information and formulate a novel data optimization framework. In order to facilitate the solution of the formulated data optimization problem, this research also explores new methods for efficient computation of parameter sensitivity, which is a key for representing the data structures and enabling data optimization. Efforts have been made to derive the analytic expressions of the sensitivity of battery electrochemical parameters by leveraging reasonable assumptions and model reformulation and simplification techniques. The proposed methodology is applied to estimating the electrochemical parameters of a single particle lithium-ion battery model in simulation and experiments, showing excellent estimation and voltage prediction accuracy compared with the traditional approach
Recommended from our members
Energy-Optimal Motion Control and Mission Planning for Multirotor Unmanned Aerial Vehicles Based on Modeling of Integrated System Dynamics
Get PDFElectric multirotor aerial vehicles are an emerging technology with extensive potential applications across a wide range of fields, but flight time and range limitations currently impose significant constraints on the use of such vehicles. Improving the vehicle energy performance is therefore a critical research topic, and one promising strategy is to optimize operational energy efficiency through model-based motion planning and control. While there has been extensive research on the topic and important progress has been made, existing works generally oversimplify or disregard key vehicle subsystem behaviors, and therefore fail to capture the complete energy dynamics and exploit the full energy saving potential. To address this gap in the state of the art, a complete system-level vehicle model is developed and applied to planning and control, aiming at achieving significant energy performance improvements in this dissertation.The model captures all relevant subsystem dynamics related to the vehicle energy performance, including propeller aerodynamics, motor assembly electro-mechanical dynamics, battery electrical dynamics, and airframe rigid-body dynamics. Through experimental validation, the model demonstrates a high degree of fidelity over a wide range of operating conditions. The model is then used to demonstrate the importance and necessity of incorporating individual dynamics into model-based planning and control, highlighting the impact of battery dynamics on the propulsion limits, the influence of propeller (inflow) aerodynamics on the energy performance, and the breakdown of vehicle energy efficiency to each subsystem dynamics. An energy-optimal trajectory generation and feedback control framework is then developed based on this model, and is shown to reduce energy usage significantly relative to a baseline controller in both simulations and experimental validation over a range of waypoint-to-waypoint flight operations. Polynomial approximations of the optimized trajectories are then developed to enable rapid and computationally efficient trajectory generation. Relative to the true energy-optimal trajectories, these approximations significantly reduce computational complexity with only a slight increase in energy consumption. Finally, the framework is extended to mission planning, in which the minimum-energy order for traversing a series of waypoints in 3D space is identified. Of particular interest is to compare with the minimum-distance order, which is often assumed to be energy optimal according to conventional wisdom and frequently adopted in practice. Over a large number of missions with randomized waypoint locations, it is found that the minimum energy order differs from the minimum-distance order in a majority of the cases, and the difference in energy consumption between the two orders can be substantial among missions of varying ranges and number of waypoints
Recommended from our members
Active Learning for Excitation Optimization and Parameter Estimation of Li-ion Battery
No full textAccurate estimation of electrochemical model parameters is critical for advancing lithium-ion battery management, health diagnostics and control. Parameter estimation is typically performed by fitting model input/output to measurement data. The results, however, are significantly affected by two key factors. Firstly, the accuracy of estimation is highly sensitive to the information content of the data about the target parameters to be estimated, which depends on the quality of input excitation. While input excitation optimization has been studied in literature and shown to significantly improve the accuracy, the existing direct optimization method suffers from intrinsic parameter uncertainty due to their reliance on the unknown parameters to be estimated. Secondly, the inherent mismatch between physics-based models and actual battery dynamics, along with measurement noise, introduces uncertainty that degrades the estimation accuracy. Overcoming these challenges is essential for achieving robust and accurate parameter estimation.This dissertation addresses these challenges by developing advanced active learning strategies for excitation optimization and robust parameter estimation specifically tailored for Li-ion battery electrochemical models. Three key contributions form the core of this work.First, a novel reinforcement learning (RL) framework is proposed for input excitation design. By formulating input generation as a Markov decision process, an RL policy is trained to guide the closed-loop generation of optimal excitation sequences robust to uncertainty. The proposed RL-based approach outperforms conventional direct optimization by achieving significantly higher data information content and lower estimation errors when subject to uncertainty.Second, a nondimensionalization-based optimization framework is formulated to tackle the fundamental issue of excitation design for identifying battery electrochemical parameters, namely the dependence on unknown parameters. By reformulating the battery diffusion dynamics and the associated input optimization problem to a nondimensional parameter-free form, the framework requires only a one-time optimization, solved by RL, to generate a nondimensional excitation sequence. The obtained sequence only needs to be re-dimensionalized by scaling in time and magnitude based on the best knowledge of parameters to retrieve the applicable dimensional input sequence. This enables an iterative input optimization-parameter estimation procedure, which significantly enhances the estimation accuracy without reliance on accurate prior knowledge of parameters, which is unrealistic to have beforehand.Third, to mitigate the impact of system uncertainties manifesting as voltage residuals, a novel parameter estimation framework is formulated by combining battery physics with machine learning. First, a hybrid model is developed by using a Gaussian process regression (GPR) model to compensate the residuals of the physics-based battery model. The model is trained once at the battery beginning-of-life using a diagnostics current excitation. Then, over battery lifetime when diagnostics of battery health is needed, estimation of battery physics parameters is performed under the same diagnostic excitation, aided by the GPR model to compensate for the model uncertainty. This integration of data-driven correction with physics-based modeling combined with the use of a diagnostic excitation enables accurate estimation of battery electrode-level health-related parameters with minimal amount of data needed.Collectively, the methodologies developed in this thesis advance the state-of-the-art in Liion battery parameter estimation and diagnostics aided by excitation optimization. By integrating reinforcement learning, nondimensionalization and physics-informed machine learning, this work delivers robust, accurate and efficient estimation frameworks, validated by both simulation and experiments. The demonstrated improvement, particularly the ability to perform rapid and accurate health assessments, shows significant promise for enabling reliable battery diagnostics in both first and second life applications. 
Going Beyond Counting First Authors in Author Co-citation Analysis
Get PDFThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
Variations on the Author
Get PDF“Variations on the Author” discusses two of Eduardo Coutinho’s recent films (Um Dia na Vida, from 2010, and Últimas Conversas, posthumously released in 2015) and their contribution to the general question of documentary authorship. The director’s filmography is characterized by a consistent yet self-effacing form of authorial self-inscription: Coutinho often features as an interviewer that rather than express opinions propels discourses; an interviewer that is good at listening. This mode of self-inscription characterizes him as an author who is not expressive but who is nonetheless markedly present on the screen. In Um Dia na Vida, however, Coutinho is completely absent form the image, while Últimas Conversas, on the contrary, includes a confessional prologue that moves the director from the margins to the center of his films. This article examines the ways in which these works stand out in the filmography of a director who offers new insights into the notion of cinematic authorship
Appropriate Similarity Measures for Author Cocitation Analysis
Get PDFWe provide a number of new insights into the methodological discussion about author cocitation analysis. We first argue that the use of the Pearson correlation for measuring the similarity between authors’ cocitation profiles is not very satisfactory. We then discuss what kind of similarity measures may be used as an alternative to the Pearson correlation. We consider three similarity measures in particular. One is the well-known cosine. The other two similarity measures have not been used before in the bibliometric literature. Finally, we show by means of an example that our findings have a high practical relevance.information science;Pearson correlation;cosine;similarity measure;author cocitation analysis
Dispelling the Myths Behind First-author Citation Counts
Get PDFWe conducted a full-scale evaluative citation analysis study of scholars in the XML research field to explore just how different from each other author rankings resulting from different citation counting methods actually are, and to demonstrate the capability of emerging data and tools on the Web in supporting more realistic citation counting methods. Our results contest some common arguments for the continued
use of first-author citation counts in the evaluation of scholars, such as high correlations between author rankings by first-author citation counts and other citation
counting methods, and high costs of using more realistic citation counting methods that are not well-supported by the ISI databases. It is argued that increasingly available digital full text research papers make it possible for citation analysis studies to go beyond what the ISI databases have directly supported and to employ more
sophisticated methods
- …
