Application of Machine Learning to Optimize Management of Children in Hospital with Lower Respiratory Tract Infection.

Effective triage can help optimize the use of limited healthcare resources for managing paediatric patients with lower respiratory tract infection (LRTI), the primary cause of death worldwide for under 5 years old children. However, triage decisions do not consider medium to long term needs of hospitalized children. In this study, we aim to leverage data-driven methods using objective measures to predict the type of hospital stay (short or long). We used vital signs (heart rate, oxygen saturation, breathing rate, and temperature) recorded from 12,881 children admitted to paediatric intensive care units in China. We generated multiple features from each vital sign, and then used regularized logistic regression with 10-fold cross validation to test the generalizability of our models. We investigated the minimum number of recording days needed to provide a reliable estimate. We assessed model performance with Area Under the Curve (AUC) using Receiver Operating Characteristic. Our results show that each vital sign independently helps predict hospital stay and the AUC increases further when vital signs are combined. In addition, early prediction of the type of stay of a patient admitted for LRTI using vital signs is possible, even with using only one day of recordings. There is now a need to apply these predictive models to other populations to assess the generalizability of the proposed methods.

Supervised learning strategy for classification and regression tasks applied to aeronautical structural health monitoring problems.

This paper presents the use of a kernel-based machine learning strategy targeting classification and regression tasks in view of automatic flaw(s) detection, localization and characterization. The studied use-case is a structural health monitoring configuration with an array of piezoelectric sensors integrated on aluminium panels affected by flaws of various positions and dimensions. The measured guided wave signals are post processed with a guided wave imaging algorithm in order to obtain an image representing the health of each specimen. These images are then used as inputs to build classification and regression models. In this paper, an extensive numerical validation campaign is conducted to validate the process. Then the inversion is applied to an experimental campaign, which demonstrate the ability to use a numerically-built model to invert experimental data.

Guided wave imaging of composite plates using passive acquisitions by fiber Bragg gratings.

In this paper, imaging results of defects in composite plates using guided wave-based algorithms, such as delay and sum and Excitelet, are presented. Those algorithms are applied to passive data for which the signal corresponding to each emitter-receiver couple is recovered as a result of the cross correlation of the ambient noise measured simultaneously by the two sensors. The transition to passive imaging allows the use of lighter sensors that are unable to emit ultrasonic waves, such as fiber Bragg gratings (FBGs) sensors on optical fibers, which are used in this study. The imaging results presented here show the feasibility of active and passive imaging in composite plates using FBGs as receivers, reducing the impact of the acquisition system on the structure in the context of structural health monitoring.

Passive guided wave tomography for structural health monitoring.

In this paper the authors present a baseline-free quantitative method for imaging corrosion flaws in thin plates. It only requires an embedded guided wave sensor network used in a fully passive way, i.e., without active emission of waves. This method is called passive guided wave tomography. The aim of this development is the use of this method for the structural health monitoring of critical structures with heavy limitations on both sensor's intrusiveness and diagnostic's reliability because it allows the use of sensors that cannot emit elastic waves such as fiber Bragg gratings, which are less intrusive than piezoelectric transducers. The idea consists in using passive methods in order to retrieve the impulse response from elastic diffuse fields-naturally present in structures-measured simultaneously between the sensors. In this paper, two passive methods are studied: the ambient noise cross-correlation and the passive inverse filter. Once all the impulse responses between the sensors are retrieved, they are used as input data to perform guided wave tomography.

Passive guided waves measurements using fiber Bragg gratings sensors.

Guided elastic waves are often studied as an effective solution for Structural Health Monitoring (SHM) systems of plate-like structures thanks to the capacity to propagate on large distances. In typical applications such as monitoring delaminations in aircraft fuselage, a network made of piezoelectric transducer (PZT) is used to emit and receive such waves in the structure. Fiber Bragg grating (FBG) sensors on optical fibers are a promising alternative to PZT for guided waves measurements in practical applications due to the capacity for dense multiplexing and robustness with respect to the environment. However, unlike conventional PZT transducers, FBG sensors cannot emit waves. It is demonstrated here that FBG sensors can be used in combination with a passive diffuse noise cross-correlation technique in order to extract the coherent guided waves propagating between two sensors. This could lead to a system using only FBG sensors in the near future. The reconstructed signals can then be analyzed with usual guided waves algorithms, like in active SHM systems, keeping all the advantages of this kind of monitoring in terms of fine diagnosis. The experimental demonstration shown in this paper is performed at ultrasonic frequencies (20-100 kHz) typically used in guided waves based SHM systems showing the potential of the approach.

Multichannel Multiple Signal Classification for dispersion curves extraction of ultrasonic guided waves.

Multichannel acquisition of ultrasonic guided waves can be used to extract dispersion curves, from the time-position domain (t-x) to the frequency-wavenumber (f-k), or frequency-velocity domain (f-c). Accurate measurements are needed in order to be able to precisely characterize the specimen, by improving the extraction of low amplitude modes and enhance resolution. The proposed method is based on the MUltiple SIgnal Classification algorithm combined with a multi-emitter and multi-receiver acquisition. In this work, this method is applied on experimental data to extract dispersive information from multilayered bonded specimens.

Excitation and focusing of Lamb waves in a multilayered anisotropic plate.

The radiation of Lamb waves by an axisymmetric source on the surface of an anisotropic plate is investigated. An analytical expression of the Green's function, valid in the far field domain, is derived. This approximation shows that the anisotropy of the propagation medium induces a focusing of Lamb modes in some directions, which correspond to minima of the slowness. Numerical simulations and experiments demonstrate that for the fundamental A(0) and S(0) modes, this phenomenon, analog to the phonon focusing effect, can be very strong in materials such as composite fiber-reinforced polymers. This effect due to the plate anisotropy must be correctly taken into account, for example, in order to develop systems for in situ structural health monitoring. The choice of the most appropriate Lamb mode, the excitation frequency, and the design of the array of piezoelectric disks used as transmitters and receivers depends on such considerations.