Machine learning at the interface of structural health monitoring and non-destructive evaluation.

While both non-destructive evaluation (NDE) and structural health monitoring (SHM) share the objective of damage detection and identification in structures, they are distinct in many respects. This paper will discuss the differences and commonalities and consider ultrasonic/guided-wave inspection as a technology at the interface of the two methodologies. It will discuss how data-based/machine learning analysis provides a powerful approach to ultrasonic NDE/SHM in terms of the available algorithms, and more generally, how different techniques can accommodate the very substantial quantities of data that are provided by modern monitoring campaigns. Several machine learning methods will be illustrated using case studies of composite structure monitoring and will consider the challenges of high-dimensional feature data available from sensing technologies like autonomous robotic ultrasonic inspection. This article is part of the theme issue 'Advanced electromagnetic non-destructive evaluation and smart monitoring'.

Bayesian and Markov chain Monte Carlo methods for identifying nonlinear systems in the presence of uncertainty.

In this paper, the authors outline the general principles behind an approach to Bayesian system identification and highlight the benefits of adopting a Bayesian framework when attempting to identify models of nonlinear dynamical systems in the presence of uncertainty. It is then described how, through a summary of some key algorithms, many of the potential difficulties associated with a Bayesian approach can be overcome through the use of Markov chain Monte Carlo (MCMC) methods. The paper concludes with a case study, where an MCMC algorithm is used to facilitate the Bayesian system identification of a nonlinear dynamical system from experimentally observed acceleration time histories.

Aspects of structural health and condition monitoring of offshore wind turbines.

Wind power has expanded significantly over the past years, although reliability of wind turbine systems, especially of offshore wind turbines, has been many times unsatisfactory in the past. Wind turbine failures are equivalent to crucial financial losses. Therefore, creating and applying strategies that improve the reliability of their components is important for a successful implementation of such systems. Structural health monitoring (SHM) addresses these problems through the monitoring of parameters indicative of the state of the structure examined. Condition monitoring (CM), on the other hand, can be seen as a specialized area of the SHM community that aims at damage detection of, particularly, rotating machinery. The paper is divided into two parts: in the first part, advanced signal processing and machine learning methods are discussed for SHM and CM on wind turbine gearbox and blade damage detection examples. In the second part, an initial exploration of supervisor control and data acquisition systems data of an offshore wind farm is presented, and data-driven approaches are proposed for detecting abnormal behaviour of wind turbines. It is shown that the advanced signal processing methods discussed are effective and that it is important to adopt these SHM strategies in the wind energy sector.

Bayesian sensitivity analysis of a model of the aortic valve.

Understanding the mechanics of the aortic valve has been a focus of attention for many years in the biomechanics literature, with the aim of improving the longevity of prosthetic replacements. Finite element models have been extensively used to investigate stresses and deformations in the valve in considerable detail. However, the effect of uncertainties in loading, material properties and model dimensions has remained uninvestigated. This paper presents a formal statistical consideration of a selected set of uncertainties on a fluid-driven finite element model of the aortic valve and examines the magnitudes of the resulting output uncertainties. Furthermore, the importance of each parameter is investigated by means of a global sensitivity analysis. To reduce computational cost, a Bayesian emulator-based approach is adopted whereby a Gaussian process is fitted to a small set of training data and then used to infer detailed sensitivity analysis information. From the set of uncertain parameters considered, it was found that output standard deviations were as high as 44% of the mean. It was also found that the material properties of the sinus and aorta were considerably more important in determining leaflet stress than the material properties of the leaflets themselves.

Computational model of an infant brain subjected to periodic motion simplified modelling and Bayesian sensitivity analysis.

Non-accidental head injury in infants, or shaken baby syndrome, is a highly controversial and disputed topic. Biomechanical studies often suggest that shaking alone cannot cause the classical symptoms, yet many medical experts believe the contrary. Researchers have turned to finite element modelling for a more detailed understanding of the interactions between the brain, skull, cerebrospinal fluid (CSF), and surrounding tissues. However, the uncertainties in such models are significant; these can arise from theoretical approximations, lack of information, and inherent variability. Consequently, this study presents an uncertainty analysis of a finite element model of a human head subject to shaking. Although the model geometry was greatly simplified, fluid-structure-interaction techniques were used to model the brain, skull, and CSF using a Eulerian mesh formulation with penalty-based coupling. Uncertainty and sensitivity measurements were obtained using Bayesian sensitivity analysis, which is a technique that is relatively new to the engineering community. Uncertainty in nine different model parameters was investigated for two different shaking excitations: sinusoidal translation only, and sinusoidal translation plus rotation about the base of the head. The level and type of sensitivity in the results was found to be highly dependent on the excitation type.

Identification of pre-sliding friction dynamics.

The hysteretic nonlinear dependence of pre-sliding friction force on displacement is modeled using different physics-based and black-box approaches including various Maxwell-slip models, NARX models, neural networks, nonparametric (local) models and dynamical networks. The efficiency and accuracy of these identification methods is compared for an experimental time series where the observed friction force is predicted from the measured displacement. All models, although varying in their degree of accuracy, show good prediction capability of pre-sliding friction. Finally, we show that even better results can be achieved by using an ensemble of the best models for prediction.

Noninvasive investigation of blood oxygenation dynamics of tumors by near-infrared spectroscopy.

The measurement of dynamic changes in the blood oxygenation of tumor vasculature could be valuable for tumor prognosis and optimizing tumor treatment plans. In this study we employed near-infrared spectroscopy (NIRS) to measure changes in the total hemoglobin concentration together with the degree of hemoglobin oxygenation in the vascular bed of breast and prostate tumors implanted in rats. Measurements were made while inhaled gas was alternated between 33% oxygen and carbogen (95% O(2), 5% CO(2)). Significant dynamic changes in tumor oxygenation were observed to accompany respiratory challenge, and these changes could be modeled with two exponential components, yielding two time constants. Following the Fick principle, we derived a simplified model to relate the time constants to tumor blood-perfusion rates. This study demonstrates that the NIRS technology can provide an efficient, real-time, noninvasive means of monitoring the vascular oxygenation dynamics of tumors and facilitate investigations of tumor vascular perfusion. This may have prognostic value and promises insight into tumor vascular development.

A novel method for fast imaging of brain function, non-invasively, with light.

Imaging of the human body by any non-invasive technique has been an appropriate goal of physics and medicine, and great success has been obtained with both Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) in brain imaging. Non-imaging responses to functional activation using near infrared spectroscopy of brain (fNIR) obtained in 1993 (Chance, et al. [1]) and in 1994 (Tamura, et al. [2]) are now complemented with images of pre-frontal and parietal stimulation in adults and pre-term neonates in this communication (see also [3]). Prior studies used continuous [4], pulsed [3] or modulated [5] light. The amplitude and phase cancellation of optical patterns as demonstrated for single source detector pairs affords remarkable sensitivity of small object detection in model systems [6]. The methods have now been elaborated with multiple source detector combinations (nine sources, four detectors). Using simple back projection algorithms it is now possible to image sensorimotor and cognitive activation of adult and pre- and full-term neonate human brain function in times < 30 sec and with two dimensional resolutions of < 1 cm in two dimensional displays. The method can be used in evaluation of adult and neonatal cerebral dysfunction in a simple, portable and affordable method that does not require immobilization, as contrasted to MRI and PET.

Ubiquitous expression of cytokines in diverse leukemias of lymphoid and myeloid lineage.

It has recently been suggested that autocrine production of hematopoietic regulatory molecules can modulate the cardinal features of many leukemic states: excessive proliferation of the neoplastic cells and suppression of the normal elements. We therefore analyzed samples obtained from 57 patients with a variety of hematologic malignancies (21, acute myelogenous leukemia; 14, acute lymphoblastic leukemia; 12, Philadelphia chromosome-positive chronic myelogenous leukemia [blast phase] or acute leukemia; 5, chronic lymphocytic leukemia; and 5, chronic myelomonocytic leukemia) for expression of interleukin-1 beta (IL-1 beta) and tumor necrosis factor-alpha (TNF-alpha) transcripts on Northern blots. TNF-alpha mRNA was discerned in almost half of the samples (47%), and was expressed in some patients with every type of leukemia, except T-cell acute lymphoblastic leukemia (ALL). Expression occurred with great frequency in samples (12 of 15 [80%]) from monocytic (acute or chronic) leukemias, and from advanced chronic lymphocytic leukemia (4 of 5 samples [80%]). IL-1 beta transcripts were detected in 20 of 57 samples (35%). Its presence, like that of TNF-alpha, was ubiquitous, and only chronic lymphocytic leukemia and T-cell acute lymphoblastic leukemia cells consistently failed to produce IL-1 beta message. Therefore it appears that TNF-alpha and/or IL-1 beta mRNA can be found in the leukemic cells from a substantial subset of patients with B cell-derived acute lymphoblastic leukemia as well as with chronic and acute myeloid, monocytic or lymphocytic leukemias. Because these cytokines have potent direct and indirect effects on normal and malignant hematopoiesis, their widespread constitutive expression by neoplastic blood cells may play a fundamental role in driving the leukemic process.

Interferon-stimulated genes in interferon-sensitive and -resistant chronic myelogenous leukemia patients.

alpha-Interferon induces hematological and cytogenetic remissions in some individuals with newly diagnosed Philadelphia-positive chronic myelogenous leukemia. However, interferon-resistant disease occurs in a consistent patient subset (primary resistance) and develops during therapy in additional patients (secondary resistance). Several alpha-interferon-inducible genes have been characterized. In interferon-resistant cell line variants, defects in these genes have been implicated in the mechanisms mediating resistance. We have, therefore, evaluated mRNA expression of four interferon-stimulated genes (ISGs) following alpha-interferon therapy. Twenty-seven chronic myelogenous leukemia patients (ten interferon-sensitive patients, 17 interferon-resistant patients) were studied. Peripheral blood samples were collected prior to and 1 to 7 days after starting interferon therapy and analyzed for the expression of 2'-5' oligoadenylate synthetase, ISG-15, ISG-54, and 6-16 transcripts. Following therapy with alpha-interferon, 2'-5' oligoadenylate synthetase, ISG-54, and 6-16 transcripts were discerned in all patients regardless of their response to interferon. The ISG-15 message was detected in eight of nine interferon-sensitive and in 15 of 16 interferon-resistant patients, as well. Overall, no consistent defect in the ISG system could be identified. Therefore, lack of induction of these genes cannot explain resistance to alpha-interferon in chronic myelogenous leukemia patients. Other mechanisms such as posttranslational modification, leading to defects in the ISG corresponding proteins, may play a role in the development of resistance.

Bronchoalveolar lavage of allergic asthmatic patients following allergen bronchoprovocation.

In these studies, we describe the use of bronchoalveolar lavage (BAL) to study local changes following aerosol bronchoprovocation (BPC) and environmental exposure to antigen in mildly symptomatic asthmatic patients. The BAL was performed in 12 atopic subjects "out of season," and in five normal subjects at baseline, less than or equal to 4, or 24 hours following BPC. Five asthmatic patients were also lavaged during seasonal exposure to allergen. The BAL cells were examined with light and transmission electron microscopy. Bronchoprovocation, by itself, resulted in an average maximal decrease in FEV1 of 13 percent just prior to BAL. There was no significant decrease in FEV1 as a result of BAL. Within four hours after BPC, the number of neutrophils was significantly greater in BAL compared to baseline (1.5 +/- 0.6 X 10(5) vs 3.4 +/- 1.7 X 10(5) cells; p less than 0.01), and the number of eosinophils was significantly greater within four hours and at 24 hours when compared to baseline values (0.4 +/- 0.3 X 10(5) vs 1.9 +/- 0.7 X 10(5) vs 1.2 +/- 0.4 X 10(5) cells; p less than 0.02). Transmission electron micrographs of BAL from lungs of asthmatic patients revealed degranulation of mast cells and loss of core material from eosinophil granules following challenge with aerosolized allergen or with spontaneous environmental exposure. These studies show that in carefully selected, mildly symptomatic asthmatic subjects, BPC and BAL may be useful to evaluate pathogenetic mechanisms in allergic bronchial asthma.