Deep Learning-Based Residual Control Chart for Binary Response

A residual (r) control chart of asymmetrical and non-normal binary response variable with highly correlated explanatory variables is proposed in this research. To avoid multicollinearity between multiple explanatory variables, we employ and compare a neural network regression model and deep learning regression model using Bayesian variable selection (BVS), principal component analysis (PCA), nonlinear PCA (NLPCA) or whole multiple explanatory variables. The advantage of our r control chart is able to process both non-normal and correlated multivariate explanatory variables by employing a neural network model and deep learning model. We prove that the deep learning r control chart is relatively efficient to monitor the simulated and real binary response asymmetric data compared with r control chart of the generalized linear model (GLM) with probit and logit link functions and neural network r control chart.

Unraveling the time-scale teleconnections between soil moisture and vegetation

<p>Identifying causal relations from observational data is key to understand Earth system interactions. Extensions to spatio-temporal analysis at different scales are of vital importance for better understanding dynamical phenomenon of natural complex systems. Soil moisture-vegetation interactions constitute a central part of ecosystem functioning and health. Here we are interested in uncovering (potentially nonlinear) spatio-temporal causal relations at different time scales between two relevant Earth observation variables: soil moisture (SM) and vegetation optical depth (VOD). To aboard the complexity data problem, we extract relevant and expressive feature components with the nonlinear kernel-based dimensional reduction method ROCK-PCA in [1]. The method yields the main modes of variability of the variables that are then used to study causal relations. To infer causality relations we use the cross-information kernel Granger causality (XKGC) method introduced in [2], which accounts for nonlinear cross-relations between the involved variables and generalizes nonlinear GC methods. Results are succesfully compared to standard correlation analysis, transfer entropy and convergent cross-mapping alternative methods. In general XKGC identifies a sparser connectivity than correlation. Also, well-known wet and dry patterns are identified as reported in the literature, but other interesting unreported connections and spatio-temporal SM<-->VOD emerge.</p><p>REFERENCES<br>[1] D. Bueso, M. Piles and G. Camps-Valls, "Nonlinear PCA for Spatio-Temporal Analysis of<br>Earth Observation Data," in IEEE Transactions on Geoscience and Remote Sensing, accepted (2020).<br>[2] Brajard, J., Charantonis, A., Chen, C., & Runge, J. (Eds.). (2019). Proceedings of the<br>9th International Workshop on Climate Informatics: CI 2019 (No. NCAR/TN-561+PROC).</p>

Strain Field Pattern Recognition for Structural Health Monitoring Applications

Strain field pattern recognition, also known as strain mapping, is a structural health monitoring approach based on strain measurements gathered through a network of sensors (i.e., strain gauges and fiber optic sensors such as FGBs or distributed sensing), data-driven modeling for feature extraction (i.e., PCA, nonlinear PCA, ANNs, etc.), and damage indices and thresholds for decision making (i.e., Q index, T2 scores, and so on). The aim is to study the correlations among strain readouts by means of machine learning techniques rooted in the artificial intelligence field in order to infer some change in the global behavior associated with a damage occurrence. Several case studies of real-world engineering structures both made of metallic and composite materials are presented including a wind turbine blade, a lattice spacecraft structure, a UAV wing section, a UAV aircraft under real flight operation, a concrete structure, and a soil profile prototype.