Development Of Data-driven Methods Research Articles

State-of-Health Estimation for Lithium-Ion Battery Based on an Attention-Based CNN-GRU Model with Reconstructed Feature Series

Health monitoring is an essential task for lithium battery systems. Recently, with the development of data-driven methods, deep learning has been successfully deployed for state-of-health (SOH) estimation. However, existing models trained using raw samples directly usually contain noise due to sensor errors. To enhance the performance of SOH prediction, short-term segments are extracted for SOH estimation based on reasonable SOC ranges. To address the measuring error that exists in the voltage and temperature samples, the reconstructed feature series (RFSs) is designed to restrain the signals’ noise. Then, a CNN-GRU network with attention mechanism is proposed to achieve SOH estimation based on short-term RFSs’ samples. To further enhance accuracy, a parallel structure is designed to fuse the feature information from both streams, raw samples, and RFSs in a reasonable manner. The performance of our proposed method is validated over a wide range of experiments on the Oxford battery degradation dataset, where the RMSE and MAE averaged 0.582% and 0.524%, respectively, demonstrating its forward estimation performance.

Behavioral Studies Using Large-Scale Brain Networks - Methods and Validations.

Mapping human behaviors to brain activity has become a key focus in modern cognitive neuroscience. As methods such as functional MRI (fMRI) advance cognitive scientists show an increasing interest in investigating neural activity in terms of functional connectivity and brain networks, rather than activation in a single brain region. Due to the noisy nature of neural activity, determining how behaviors are associated with specific neural signals is not well-established. Previous research has suggested graph theory techniques as a solution. Graph theory provides an opportunity to interpret human behaviors in terms of the topological organization of brain network architecture. Graph theory-based approaches, however, only scratch the surface of what neural connections relate to human behavior. Recently, the development of data-driven methods, e.g., machine learning and deep learning approaches, provide a new perspective to study the relationship between brain networks and human behaviors across the whole brain, expanding upon past literatures. In this review, we sought to revisit these data-driven approaches to facilitate our understanding of neural mechanisms and build models of human behaviors. We start with the popular graph theory approach and then discuss other data-driven approaches such as connectome-based predictive modeling, multivariate pattern analysis, network dynamic modeling, and deep learning techniques that quantify meaningful networks and connectivity related to cognition and behaviors. Importantly, for each topic, we discuss the pros and cons of the methods in addition to providing examples using our own data for each technique to describe how these methods can be applied to real-world neuroimaging data.

A model-based calibration approach for structural fault diagnosis using piezoelectric impedance measurements and a finite element model

The electromechanical coupling property of piezoelectric transducers gives rise to a promising class of structural fault diagnosis methods often referred to collectively as impedance-based approaches. One active line of research in the related literature is the development of data-driven methods that can leverage the available experimental impedance measurements to accurately pinpoint the location and severity of structural faults. In this article, we offer a new perspective to the problem by casting the impedance-based fault diagnosis into a statistical calibration formulation, which has gained a wide popularity in the industrial statistics community in the past two decades. Specifically, we decide to estimate the values of the fault attributes (e.g. location and severity) that achieve the closest match between the outputs from a finite element model and those experimentally solicited from the host structure. We further propose to couple this statistical formulation with a pre-screening procedure to reduce the calibration search space and mitigate parameter identifiability issues. In addition to the merit of capably diagnosing structural faults, the proposed approach extends various useful concepts from the statistical calibration literature to the structural health monitoring applications, such as the construction of surrogate models for modeling and predicting impedance changes, the explicit use of a bias function to correct for inherent inadequacies in finite element models, as well as the ability to produce continuous probability distributions for quantifying a fault’s severity. These additional benefits substantially enhance both the fault diagnosis capability and computational efficiency. We demonstrate the effectiveness of the proposed approach using two simulated and two experimental case studies from the literature.

Blind source separation problem in GPS time series

A critical point in the analysis of ground displacement time series, as those recorded by space geodetic techniques, is the development of data-driven methods that allow the different sources of deformation to be discerned and characterized in the space and time domains. Multivariate statistic includes several approaches that can be considered as a part of data-driven methods. A widely used technique is the principal component analysis (PCA), which allows us to reduce the dimensionality of the data space while maintaining most of the variance of the dataset explained. However, PCA does not perform well in finding the solution to the so-called blind source separation (BSS) problem, i.e., in recovering and separating the original sources that generate the observed data. This is mainly due to the fact that PCA minimizes the misfit calculated using an $$L_{2}$$ norm ( $$\chi ^{2})$$ , looking for a new Euclidean space where the projected data are uncorrelated. The independent component analysis (ICA) is a popular technique adopted to approach the BSS problem. However, the independence condition is not easy to impose, and it is often necessary to introduce some approximations. To work around this problem, we test the use of a modified variational Bayesian ICA (vbICA) method to recover the multiple sources of ground deformation even in the presence of missing data. The vbICA method models the probability density function (pdf) of each source signal using a mix of Gaussian distributions, allowing for more flexibility in the description of the pdf of the sources with respect to standard ICA, and giving a more reliable estimate of them. Here we present its application to synthetic global positioning system (GPS) position time series, generated by simulating deformation near an active fault, including inter-seismic, co-seismic, and post-seismic signals, plus seasonal signals and noise, and an additional time-dependent volcanic source. We evaluate the ability of the PCA and ICA decomposition techniques in explaining the data and in recovering the original (known) sources. Using the same number of components, we find that the vbICA method fits the data almost as well as a PCA method, since the $$\chi ^{2}$$ increase is less than 10 % the value calculated using a PCA decomposition. Unlike PCA, the vbICA algorithm is found to correctly separate the sources if the correlation of the dataset is low ( $$<$$ 0.67) and the geodetic network is sufficiently dense (ten continuous GPS stations within a box of side equal to two times the locking depth of a fault where an earthquake of $$M_\mathrm{{w}} >6$$ occurred). We also provide a cookbook for the use of the vbICA algorithm in analyses of position time series for tectonic and non-tectonic applications.