Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “sampling algorithm”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 973 records · Page 54

Receiver Gain Modulation Circuit

A receiver gain modulation circuit (RGMC) was developed that modulates the power gain of the output of a radiometer receiver with a test signal. As the radiometer receiver switches between calibration noise references, the test signal is mixed with the calibrated noise and thus produces an ensemble set of measurements from which ensemble statistical analysis can be used to extract statistical information about the test signal. The RGMC is an enabling technology of the ensemble detector. As a key component for achieving ensemble detection and analysis, the RGMC has broad aeronautical and space applications. The RGMC can be used to test and develop new calibration algorithms, for example, to detect gain anomalies, and/or correct for slow drifts that affect climate-quality measurements over an accelerated time scale. A generalized approach to analyzing radiometer system designs yields a mathematical treatment of noise reference measurements in calibration algorithms. By treating the measurements from the different noise references as ensemble samples of the receiver state, i.e. receiver gain, a quantitative description of the non-stationary properties of the underlying receiver fluctuations can be derived. Excellent agreement has been obtained between model calculations and radiometric measurements. The mathematical formulation is equivalent to modulating the gain of a stable receiver with an externally generated signal and is the basis for ensemble detection and analysis (EDA). The concept of generating ensemble data sets using an ensemble detector is similar to the ensemble data sets generated as part of ensemble empirical mode decomposition (EEMD) with exception of a key distinguishing factor. EEMD adds noise to the signal under study whereas EDA mixes the signal with calibrated noise. It is mixing with calibrated noise that permits the measurement of temporal-functional variability of uncertainty in the underlying process. The RGMC permits the evaluation of EDA by modulating the receiver gain using an external signal. Without the RGMC, samples of calibrated references from radiometers form an ensemble data set of the natural occurring fluctuations within a receiver. By driving the gain of an otherwise stable receiver with an external signal, the conceptual framework and generalization of the mathematics of EDA can be tested. A series of measurements was conducted to evaluate and characterize the performance of the RGMC. Test signals stepped the RGMC across its dynamic range of performance using a radiometer that sampled four noise references; analysis indicates that the RGMC successfully modulated the receiver gain with an external signal. Calibration algorithms applied to four noise references demonstrate the RGMC produced ensemble data sets of the external signal.

Jones, Hollis↗

Heating system of pellet samples integrated with terahertz spectrometer

This article describes automation of temperature-dependent terahertz spectroscopic experiments. The proposed dual-heater temperature controller based on a cascade proportional-integral-derivative algorithm provides smooth temperature changes in the polyethylene-based pharmaceutical pellet samples. The device has been integrated with a terahertz time-domain spectrometer. Thermodynamic experiments can now be performed without any probe inserted into the measured sample. Selected results of temperature-induced evolution in terahertz spectra are presented.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Optical Sampler for Broad-Spectrum Sensing

We investigated the feasibility of using non-uniform optical sampling to measure wide bandwidth RF signals more efficiently. This combined two areas of signal measurement research: optical sampling, which uses optics to alleviate the bandwidth limitations of traditional electronic sampling, and compressed sensing, which seeks to measure signals with fewer samples than typically necessary. This can both increase accuracy and reduce cost over current state-of-the-art sampling devices. However, several different compressed sensing architectures and algorithms offer varying levels of performance, robustness and complexity. In this work, we compared the tradeoffs between several of these methods in simulation of a remote sensing application. We found that while feasible, compressed sensing is inherently less sensitive than fully sampled systems, and so must be paired with a very low noise optical system. We also built an experimental prototype of the system and took initial measurements.

47 OTHER INSTRUMENTATION↗

Quantitative analysis of the reconstruction performance of interpolants

The analysis presented provides a quantitative measure of the reconstruction or interpolation performance of linear, shift-invariant interpolants. The performance criterion is the mean square error of the difference between the sampled and reconstructed functions. The analysis is applicable to reconstruction algorithms used in image processing and to many types of splines used in numerical analysis and computer graphics. When formulated in the frequency domain, the mean square error clearly separates the contribution of the interpolation method from the contribution of the sampled data. The equations provide a rational basis for selecting an optimal interpolant; that is, one which minimizes the mean square error. The analysis has been applied to a selection of frequently used data splines and reconstruction algorithms: parametric cubic and quintic Hermite splines, exponential and nu splines (including the special case of the cubic spline), parametric cubic convolution, Keys' fourth-order cubic, and a cubic with a discontinuous first derivative. The emphasis in this paper is on the image-dependent case in which no a priori knowledge of the frequency spectrum of the sampled function is assumed.

Lansing, Donald L.↗

Learning Spectral Templates for Photometric Redshift Estimation from Broadband Photometry

Estimating redshifts from broadband photometry is often limited by how accurately we can map the colors of galaxies to an underlying spectral template. Current techniques utilize spectrophotometric samples of galaxies or spectra derived from spectral synthesis models. Both of these approaches have their limitations: either the sample sizes are small and often not representative of the diversity of galaxy colors, or the model colors can be biased (often as a function of wavelength), which introduces systematics in the derived redshifts. In this paper, we learn the underlying spectral energy distributions from an ensemble of ∼100 K galaxies with measured redshifts and colors. We show that we are able to reconstruct emission and absorption lines at a significantly higher resolution than the broadband filters used to measure the photometry for a sample of 20 spectral templates. We find that our training algorithm reduces the fraction of outliers in the derived photometric redshifts by up to 28%, bias up to 91%, and scatter up to 25%, when compared to estimates using a standard set of spectral templates. We discuss the current limitations of this approach and its applicability for recovering the underlying properties of galaxies. Our derived templates and the code used to produce these results are publicly available in a dedicated Github repository: https://github.com/dirac-institute/photoz-template-learning.

79 ASTRONOMY AND ASTROPHYSICS↗

The radial profile and flattening of the Milky Way’s stellar halo to 80 kpc from the SEGUE K-giant survey

We characterize the radial density, metallicity, and flattening profile of the Milky Way’s stellar halo, based on the large sample of spectroscopically confirmed giant stars from SDSS/SEGUE-2, spanning galactocentric radii 10 kpc ≤ r{sub GC} ≤ 80 kpc. After excising stars that were algorithmically attributed to apparent halo substructure (including the Sagittarius stream), the sample has 1757 K giants, with a typical metallicity precision of 0.2 dex and a mean distance accuracy of 16%. Compared to blue horizontal branch stars or RR Lyrae variables, giants are more readily understood tracers of the overall halo star population, with less bias in age or metallicity. The well-characterized selection function of the sample enables forward modeling of those data, based on ellipsoidal stellar density models, ν{sub *}(R, z), with Einasto profiles and (broken) power laws for their radial dependence, combined with a model for the metallicity gradient and the flattening profile. Among models with constant flattening, these data are reasonably well fit by an Einasto profile of n = 3.1 ± 0.5 with an effective radius r{sub eff}=15±2 kpc and a flattening of q = 0.7 ± 0.02, or comparably well by an equally flattened broken power law, with radial slopes of α{sub in} = 2.1 ± 0.3 and α{sub out} = 3.8 ± 0.1, with a break radius of r{sub break} = 18 ± 1 kpc; this is largely consistent with earlier work. We find a modest but significant metallicity gradient within the “outer” stellar halo, [Fe/H] decreasing outward. If we allow for a variable flattening q=f(r{sub GC}), we find the distribution of halo giants to be considerably more flattened at small radii, q(10 kpc) = 0.55 ± 0.02, compared to q(>30 kpc) = 0.8 ± 0.03. Remarkably, the data are then very well fit by a single power law with index of 4.2 ± 0.1 on the variable r{sub q}≡√(R{sup 2}+(z/q(r)){sup 2}). In this simple and better-fitting model, there is a break in flattening at ∼20 kpc, instead of a break in the radial density function. While different parameterizations of the radial profile vary in their parameters, their implied density gradient, ∂lnν{sub ∗}/∂lnr, is stable along a direction intermediate between major and minor axis; this gradient is crucial in any dynamical modeling that uses halo stars as tracers.

79 ASTRONOMY AND ASTROPHYSICS↗

Dark Energy Survey Year 3 results: galaxy sample for BAO measurement

Here, we present and validate the galaxy sample used for the analysis of the baryon acoustic oscillation (BAO) signal in the Dark Energy Survey (DES) Y3 data. The definition is based on a colour and redshift-dependent magnitude cut optimized to select galaxies at redshifts higher than 0.5, while ensuring a high-quality determination. The sample covers ~ 4100 square degrees to a depth of i = 22.3 (AB) at 10σ. It contains 7,031,993 galaxies in the redshift range from z = 0.6 to 1.1, with a mean effective redshift of 0.835. Redshifts are estimated with the machine learning algorithm DNF, and are validated using the VIPERS PDR2 sample. We find a mean redshift bias of z bias ~ 0.01 and a mean uncertainty, in units of 1 + z, of σ 68 ~ 0.03$. We evaluate the galaxy population of the sample, showing it is mostly built upon Elliptical to Sbc types. Furthermore, we find a low level of stellar contamination of ≲ 4 %. We present the method used to mitigate the effect of spurious clustering coming from observing conditions and other large-scale systematics. We apply it to the BAO sample and calculate weights that are used to get a robust estimate of the galaxy clustering signal. This paper is one of a series dedicated to the analysis of the BAO signal in DES Y3. In the companion papers, we present the galaxy mock catalogues used to calibrate the analysis and the angular diameter distance constraints obtained through the fitting to the BAO scale.

79 ASTRONOMY AND ASTROPHYSICS↗

SU-G-JeP1-12: Head-To-Head Performance Characterization of Two Multileaf Collimator Tracking Algorithms for Radiotherapy

Purpose: Multi-leaf collimator (MLC) tracking is being clinically pioneered to continuously compensate for thoracic and abdominal motion during radiotherapy. The purpose of this work is to characterize the performance of two MLC tracking algorithms for cancer radiotherapy, based on a direct optimization and a piecewise leaf fitting approach respectively. Methods: To test the algorithms, both physical and in silico experiments were performed. Previously published high and low modulation VMAT plans for lung and prostate cancer cases were used along with eight patient-measured organ-specific trajectories. For both MLC tracking algorithm, the plans were run with their corresponding patient trajectories. The physical experiments were performed on a Trilogy Varian linac and a programmable phantom (HexaMotion platform). For each MLC tracking algorithm, plan and patient trajectory, the tracking accuracy was quantified as the difference in aperture area between ideal and fitted MLC. To compare algorithms, the average cumulative tracking error area for each experiment was calculated. The two-sample Kolmogorov-Smirnov (KS) test was used to evaluate the cumulative tracking errors between algorithms. Results: Comparison of tracking errors for the physical and in silico experiments showed minor differences between the two algorithms. The KS D-statistics for the physical experiments were below 0.05 denoting no significant differences between the two distributions pattern and the average error area (direct optimization/piecewise leaf-fitting) were comparable (66.64 cm2/65.65 cm2). For the in silico experiments, the KS D-statistics were below 0.05 and the average errors area were also equivalent (49.38 cm2/48.98 cm2). Conclusion: The comparison between the two leaf fittings algorithms demonstrated no significant differences in tracking errors, neither in a clinically realistic environment nor in silico. The similarities in the two independent algorithms give confidence in the use of either algorithm for clinical implementation.

60 APPLIED LIFE SCIENCES↗

Mapping Stochastic Devices to Probabilistic Algorithms

Probabilistic and Bayesian neural networks have long been proposed as a method to incorporate uncertainty about the world (both in training data and operation) into artificial intelligence applications. One approach to making a neural network probabilistic is to leverage a Monte Carlo sampling approach that samples a trained network while incorporating noise. Such sampling approaches for neural networks have not been extensively studied due to the prohibitive requirement of many computationally expensive samples. While the development of future microelectronics platforms that make this sampling more efficient is an attractive option, it has not been immediately clear how to sample a neural network and what the quality of random number generation should be. This research aimed to start addressing these two fundamental questions by examining basic “off the shelf” neural networks can be sampled through a few different mechanisms (including synapse “dropout” and neuron “dropout”) and examine how these sampling approaches can be evaluated both in terms of evaluating algorithm effectiveness and the required quality of random numbers.

97 MATHEMATICS AND COMPUTING↗

DISCOVERY OF A LARGE NUMBER OF CANDIDATE PROTOCLUSTERS TRACED BY ∼15 Mpc-SCALE GALAXY OVERDENSITIES IN COSMOS

To demonstrate the feasibility of studying the epoch of massive galaxy cluster formation in a more systematic manner using current and future galaxy surveys, we report the discovery of a large sample of protocluster candidates in the 1.62 deg{sup 2} COSMOS/UltraVISTA field traced by optical/infrared selected galaxies using photometric redshifts. By comparing properly smoothed three-dimensional galaxy density maps of the observations and a set of matched simulations incorporating the dominant observational effects (galaxy selection and photometric redshift uncertainties), we first confirm that the observed ∼15 comoving Mpc-scale galaxy clustering is consistent with ΛCDM models. Using further the relation between high-z overdensity and the present day cluster mass calibrated in these matched simulations, we found 36 candidate structures at 1.6 < z < 3.1, showing overdensities consistent with the progenitors of M{sub z} {sub =} {sub 0} ∼ 10{sup 15} M {sub ☉} clusters. Taking into account the significant upward scattering of lower mass structures, the probabilities for the candidates to have at least M{sub z=} {sub 0} ∼ 10{sup 14} M {sub ☉} are ∼70%. For each structure, about 15%-40% of photometric galaxy candidates are expected to be true protocluster members that will merge into a cluster-scale halo by z = 0. With solely photometric redshifts, we successfully rediscover two spectroscopically confirmed structures in this field, suggesting that our algorithm is robust. This work generates a large sample of uniformly selected protocluster candidates, providing rich targets for spectroscopic follow-up and subsequent studies of cluster formation. Meanwhile, it demonstrates the potential for probing early cluster formation with upcoming redshift surveys such as the Hobby-Eberly Telescope Dark Energy Experiment and the Subaru Prime Focus Spectrograph survey.

79 ASTRONOMY AND ASTROPHYSICS↗

Design and expected performance of a fast neutron attenuation probe for light element density measurements

In this paper, we present the design and expected performance of a proof-of-concept 32 channel material identification system. Our system is based on the energy-dependent attenuation of fast neutrons for four elements: hydrogen, carbon, nitrogen and oxygen. We describe a new approach to obtaining a broad range of neutron energies to probe a sample, as well as our technique for reconstructing the molar densities within a sample. The system's performance as a function of time-of-flight energy resolution is explored using a Geant4-based Monte Carlo. Our results indicate that, with the expected detector response of our system, we will be able to determine the molar density of all four elements to within a 20–30% accuracy in a two hour scan time. In many cases this error is systematically low, thus the ratio between elements is more accurate. This degree of accuracy is enough to distinguish, for example, a sample of water from a sample of pure hydrogen peroxide: the ratio of oxygen to hydrogen is reconstructed to within 8±0.5% of the true value. Lastly, with future algorithm development that accounts for backgrounds caused by scattering within the sample itself, the accuracy of molar densities, not ratios, may improve to the 5–10% level for a two hour scan time.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Enabling Explosives and Contraband Detection

We present the design and performance of a proof-of-concept 32 channel material identification system. Our system is based on the energy-dependent attenuation of fast neutrons for four elements: hydrogen, carbon, nitrogen and oxygen. We describe a new approach to obtaining a broad range of neutron energies to probe a sample, as well as our technique for reconstructing the molar densities within a sample. The system's performance as a function of time-of-flight energy resolution is explored using a Geant4-based Monte Carlo. Our results indicate that, with the expected detector response of our system, we will be able to determine the molar density of all four elements to within a 20-30% accuracy in a two hour scan time. In many cases this error is systematically low, thus the ratio between elements is more accurate. This degree of accuracy is enough to distinguish, for example, a sample of water from a sample of pure hydrogen peroxide: the ratio of oxygen to hydrogen is reconstructed to within 8 0.5% of the true value. Finally, with future algorithm development that accounts for backgrounds caused by scattering within the sample itself, the accuracy of molar densities, not ratios, may improve to the 5-10% level for a two hour scan time. Experimental performance was evaluated with various thicknesses of polyethylene. The detector response in terms of energy, particle identification, and timing are presented as well.

42 ENGINEERING↗

Computational studies of x-ray scattering from three-dimensionally-aligned asymmetric-top molecules

We theoretically and numerically analyze x-ray scattering from asymmetric-top molecules three-dimensionally aligned using elliptically polarized laser light. A rigid-rotor model is assumed. The principal axes of the polarizability tensor are assumed to coincide with the principal axes of the moment of inertia tensor. Several symmetries in the Hamiltonian are identified and exploited to enhance the efficiency of solving the time-dependent Schroedinger equation for each rotational state initially populated in a thermal ensemble. Using a phase-retrieval algorithm, the feasibility of structure reconstruction from a quasiadiabatically aligned sample is illustrated for the organic molecule naphthalene. The spatial resolution achievable strongly depends on the laser parameters, the initial rotational temperature, and the x-ray pulse duration. We demonstrate that for a laser peak intensity of 5 TW/cm{sup 2}, a laser pulse duration of 100 ps, a rotational temperature of 10 mK, and an x-ray pulse duration of 1 ps, the molecular structure may be probed at a resolution of 1 A ring .

74 ATOMIC AND MOLECULAR PHYSICS↗

StarHorse results for spectroscopic surveys and Gaia DR3: Chrono-chemical populations in the solar vicinity, the genuine thick disk, and young alpha-rich stars

The Gaia mission has provided an invaluable wealth of astrometric data for more than a billion stars in our Galaxy. The synergy between Gaia astrometry, photometry, and spectroscopic surveys gives us comprehensive information about the Milky Way. Using the Bayesian isochrone-fitting code StarHorse, we derive distances and extinctions for more than 10 million unique stars listed in both Gaia Data Release 3 and public spectroscopic surveys: 557 559 in GALAH+ DR3, 4 531 028 in LAMOST DR7 LRS, 347 535 in LAMOST DR7 MRS, 562 424 in APOGEE DR17, 471 490 in RAVE DR6, 249 991 in SDSS DR12 (optical spectra from BOSS and SEGUE), 67 562 in the Gaia-ESO DR5 survey, and 4 211 087 in the Gaia RVS part of the Gaia DR3 release. StarHorse can increase the precision of distance and extinction measurements where Gaia parallaxes alone would be uncertain. We used StarHorse for the first time to derive stellar ages for main-sequence turnoff and subgiant branch stars, around 2.5 million stars, with age uncertainties typically around 30%; the uncertainties drop to 15% for subgiant-branch-only stars, depending on the resolution of the survey. With the derived ages in hand, we investigated the chemical-age relations. In particular, the α and neutron-capture element ratios versus age in the solar neighbourhood show trends similar to previous works, validating our ages. We used the chemical abundances from local subgiant samples of GALAH DR3, APOGEE DR17, and LAMOST MRS DR7 to map groups with similar chemical compositions and StarHorse ages, using the dimensionality reduction technique t-SNE and the clustering algorithm HDBSCAN. We identify three distinct groups in all three samples, confirmed by their kinematic properties: the genuine chemical thick disk, the thin disk, and a considerable number of young alpha-rich stars (427) that are also a part of the delivered catalogues. We confirm that the genuine thick disk’s kinematics and age properties are radically different from those of the thin disk and compatible with high-redshift (z ≈ 2) star-forming disks with high dispersion velocities. We also find a few extra chemical populations in GALAH DR3 thanks to the availability of neutron-capture element information.

79 ASTRONOMY AND ASTROPHYSICS↗

Training and Validation of Spectral Gap Filling Algorithm for Cpf-Ceres Intercalibration

The Climate Absolute Radiance and Refractivity Observatory (CLARREO) Pathfinder (CPF) mission is set to launch an SI-traceable reflective solar (RS) spectrometer aboard the International Space Station to measure Earth-reflected solar radiation with a radiometric uncertainty of 0.3% (k=1). The CPF intercalibration team has devised a cutting-edge methodology to accurately transfer the benchmark CPF calibration reference to the shortwave (SW) channel (200-5000 nm) of the Clouds and the Earth’s Radiant Energy System (CERES) instrument. The spectral range of CPF measurements spans from 350-2300 nm, while the CERES SW channel measures the Earth-reflected broadband solar radiances between 200 nm to 5 μm. To conduct precise CPF-CERES intercalibration analysis, the CPF-like spectral radiances outside the CPF spectral range need to be estimated to match the CERES SW spectral range. In response, the team has developed a fast algorithm that leverages spectrally redundant information within the CPF-measured portion through principal component analysis (PCA) and utilizes pre-established spectral correlation relationships among wavelengths to extend the CPF spectrum below 350 nm and above 2300 nm. Our results show that the algorithm achieves excellent accuracy in generating the missing energy in the UV and IR portions. The RMS error in the UV region is less than 4.5x10-3 W/m2/sr/nm, while in the IR region, it is smaller than 8x10-5 W/m2/sr/nm. Our methodology was validated using measured EMIT radiance data, which covers the spectral range from 0.381 μm to 2.493 μm. We employed EMIT radiances within the wavelength range of 0.43 – 2.25 μm to generate radiances for both the shorter wavelength range (0.381 – 0.43 μm) and longer wavelength range (2.25 – 2.493 μm). The generated radiances agree very well with the measured EMIT radiances. The standard deviation in the integrated broadband radiances was about 0.1%, and the bias is less than 0.004% for over 1.5 million EMIT measured samples. These statistics show that the spectral gap filling algorithm is robust and effective in substantially reducing the spectral difference-induced uncertainty in the CPF-CERES intercalibration samples.

Qiguang Yang↗

Waveform processing using neural network algorithms on the front-end electronics

In a multi-channel radiation detector readout system, waveform sampling, digitization, and raw data transmission to the data acquisition system constitute a conventional processing chain. The deposited energy on the sensor is estimated by extracting peak amplitudes, area under pulse envelopes from the raw data, and starting times of signals or time of arrivals. However, such quantities can be estimated using machine learning algorithms on the front-end Application-Specific Integrated Circuits (ASICs), often termed as “edge computing”. Edge computation offers enormous benefits, especially when the analytical forms are not fully known or the registered waveform suffers from noise and imperfections of practical implementations. In this work, we aim to predict peak amplitude from a single waveform snippet whose rising and falling edges containing only 3 to 4 samples. We thoroughly studied two well-accepted neural network algorithms, Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN) by varying their model sizes. Further, to better fit front-end electronics, neural network model reduction techniques, such as network pruning methods and variable-bit quantization approaches, were also studied. By combining pruning and quantization, our best performing model has the size of 1.5 KB, reduced from 16.6 KB of its full model counterpart. It can reach mean absolute error of 0.034 comparing to that of a naive baseline of 0.135. Such parameter-efficient and predictive neural network models established feasibility and practicality of their deployment on front-end ASICs.

47 OTHER INSTRUMENTATION↗

A performance evaluation framework for building fault detection and diagnosis algorithms

Fault detection and diagnosis (FDD) algorithms for building systems and equipment represent one of the most active areas of research and commercial product development in the buildings industry. However, far more effort has gone into developing these algorithms than into assessing their performance. As a result, considerable uncertainties remain regarding the accuracy and effectiveness of both research-grade FDD algorithms and commercial products - a state of affairs that has hindered the broad adoption of FDD tools. This article presents a general, systematic framework for evaluating the performance of FDD algorithms. The article focuses on understanding the possible answers to two key questions: in the context of FDD algorithm evaluation, what defines a fault and what defines an evaluation input sample? The answers to these questions, together with appropriate performance metrics, may be used to fully specify evaluation procedures for FDD algorithms.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Analysis of breathing cracks using vibrations

The influence of half-ellipse-shaped fatigue cracks and machined through-thickness cuts on the impact response of cantilever beam samples were analysed using the Finite Element Method and compared to experimental data for different crack depths. The effect of sample width was analysed for geometries with fatigue cracks (aspect ratio a/c = 0.45) and through-thickness cuts (rectangular cracks). The differences between both types of damage were analysed numerically and experimentally. The effect of sample width was studied and it was found that the behaviour of the cantilever with a fatigue-crack approaches that of the saw cut sample when the fatigue crack reaches the edges of the test piece. Acceleration data were analysed using the Fast Fourier Transform algorithm and Power Density Spectra (PSD) were obtained. For fatigue-cracked samples, harmonics of the frequencies of natural modes of oscillation were found. Analytical modelling was done using a mass-spring-damper system for studying the bilinear behaviour of the stiffness element. Harmonics found were assigned to the breathing behaviour of the crack. The detection threshold using frequency shifts inspection was twice as high as using spectral analysis. This would allow crack detection at half of the fatigue life of the specimen.

42 ENGINEERING↗