A Monte Carlo-Based Method for Assessing the Measurement Uncertainty in the Training and Use of Artificial Neural Networks

Coral, Rodrigo; Flesch, Carlos Alberto; Penz, Cesar Alberto; Roisenberg, Mauro; Pacheco, Antonio Luiz Schalata

doi:10.1515/mms-2016-0015

Cited by 9 publications

(3 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, we used 𝐵 = 12 input quantities 𝑋 1...𝐵 = {𝑝 D , 𝑝 S , ReΓ D , ImΓ D , ReΓ S , ImΓ S , Re𝐸 𝑅 , Im𝐸 𝑅 , Re𝐸 S , Im𝐸 S , Re𝐸 D , Im𝐸 D } in the measurement function (4). For each 𝑓 1...𝑘 correlation matrix of size 𝐵 × 𝐵 was calculated using the Cholesky factorization with the Octave chol function [18], and the correlation degree between input quantities was checked by finding the number of instances of statistically significant (𝛼 < 0.005) correlation using the Octave corrcoef function [18], which is shown in Table 3 as numbers.…”

Section: Validation Of the MCM Software On Example Of Determining Typ...mentioning

confidence: 99%

“…The MCM has an advantage over the GUM when the estimated uncertainty of measurement consists mainly of measurement errors with a PDF significantly deviating from the normal distribution and/or when it is determined from a small number 𝑛 of measurement samples. The low popularity of the MCM was due to the need to use multiple (up to 10 6 ) samples of measured quantities with appropriately chosen PDF distributions, but with the currently available computing power of personal computers, the estimation of uncertainty interval at 𝑝 = 95% confidence level usually takes a few seconds, so the MCM is now growing in popularity [4,5].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Determination of measurement uncertainty by a Monte Carlo method for an RF power sensor calibration system using a VNA

Jaworski,

Szatkowski,

Kossek

2023

Metrology and Measurement Systems

View full text Add to dashboard Cite

This work proposes a systematic assessment of measuring type A uncertainty (caused by random errors) used in RF power sensor calibration. To reduce A type uncertainty, several successive measurements are repeated. The uncertainty arises from repeatability errors in connectors caused by changes in their electrical properties during repeated mating. The suitability of the METAS UncLib software was analysed and we concluded that software should be developed to take into account the shape of probability density function (PDF) using a Monte Carlo method (MCM), which was lacking in METAS UncLib. The self-developed software was then tested on an example taken from the literature and the superiority of the MCM over the analytical method (GUM) was confirmed. During the calibration of the RF sensor using a vector network analyzer (VNA), a series of repeated measurements were performed and, after applying our MCM software, it was found that the measurement uncertainties calculated by the MCM method were several times larger than those by the GUM. The reason for this was that the correlation between the measured input quantities was not taken into account. When this was done using a covariance matrix and assuming a normal PDF of the input quantities, the results obtained with the GUM and the MCM converged. Our main objective was to investigate the influence of the PDF shape of the input measurement samples on the measurement uncertainty. Taking more than a dozen measurements is too costly, on the other hand, the small sample size prevents a reliable determination of the PDF shape. Finally, to overcome this inconvenience, we have developed a special method that uses the histograms of standardized input data taken at all measurement frequencies under fixed conditions without disconnecting the connectors, to increasing the total number of results which were needed to create the PDF histograms of input quantities.

show abstract

Section: Validation Of the MCM Software On Example Of Determining Typ...mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Determination of measurement uncertainty by a Monte Carlo method for an RF power sensor calibration system using a VNA

Jaworski,

Szatkowski,

Kossek

2023

Metrology and Measurement Systems

View full text Add to dashboard Cite

show abstract

“…Auto-associative neural networks are essentially multi-layered perceptrons (MLPs) [6,[27][28] that are trained by presenting to the network target vectors equal to the training ones, i.e. the network is expected to produce the input at its output.…”

Section: Auto-associative Neural Networkmentioning

confidence: 99%

Metrology and Measurement Systems

Dworakowski¹,

Dziedziech²,

Jabłoński³

2018

View full text Add to dashboard Cite

Reliable monitoring for detection of damage in epicyclic gearboxes is a serious concern for all industries in which these gearboxes operate in a harsh environment and in variable operational conditions. In this paper, autonomous multidimensional novelty detection algorithms are used to estimate the gearbox' health state based on vectors of features calculated from the vibration signal. The authors examine various feature vectors, various sources of data and many different damage scenarios in order to compare novel detection algorithms based on three different principles of operation: a distance in the feature space, a probability distribution, and an ANN (artificial neural network)-based model reconstruction approach. In order to compensate for non-deterministic results of training of neural networks, which may lead to different network performance, the ensemble technique is used to combine responses from several networks. The methods are tested in a series of practical experiments involving implanting a damage in industrial epicyclic gearboxes, and acquisition of data at variable speed conditions.

show abstract

Function-oriented defect assessment in hybrid sheet molding compound tensile specimen using surrogate models

Bretz

Koch

Dębowski

et al. 2022

Prod. Eng. Res. Devel.

View full text Add to dashboard Cite

Glass fiber (GF) Sheet Molding Compound (SMC) composites are popular lightweight materials due to their good processability. Hybrid SMCs expand the field of operation, combining the high stiffness of unidirectional carbon fibers (CF) with the economic efficiency of GF. Combinations of manufacturing deviations (delamination, varying GF content, CF misorientation) occur during the production of hybrid SMCs and impede the mechanical performance of the part. A function-oriented quality assurance instead of strict tolerances is proposed. Finite element (FE) simulations are computationally too expensive for an assessment within the cycle time. Hence, surrogate models are trained on multiple parameterized FE simulations. The surrogate models shall allow for an individual functional assessment in real-time based on integrated measurement inputs. This work focuses on the generation of parametrized FE simulations for measurement inputs and surrogate modeling. Simulations and surrogate models show acceptable deviations from tensile tests for multiple combinations of manufacturing deviations. The measurement uncertainty of the stiffness prediction is assessed for both the FE simulation and the surrogate models in accordance with the Guide to the Expression of Uncertainty in Measurement (GUM).

show abstract

A Monte Carlo-Based Method for Assessing the Measurement Uncertainty in the Training and Use of Artificial Neural Networks

Cited by 9 publications

References 13 publications

Determination of measurement uncertainty by a Monte Carlo method for an RF power sensor calibration system using a VNA

Determination of measurement uncertainty by a Monte Carlo method for an RF power sensor calibration system using a VNA

Metrology and Measurement Systems

Function-oriented defect assessment in hybrid sheet molding compound tensile specimen using surrogate models

Contact Info

Product

Resources

About