Noisy Data Conditions Research Articles

SCAE-Stacked Convolutional Autoencoder for Fault Diagnosis of a Hydraulic Piston Pump with Limited Data Samples.

Deep learning (DL) models require enormous amounts of data to produce reliable diagnosis results. The superiority of DL models over traditional machine learning (ML) methods in terms of feature extraction, feature dimension reduction, and diagnosis performance has been shown in various studies of fault diagnosis systems. However, data acquisition can sometimes be compromised by sensor issues, resulting in limited data samples. In this study, we propose a novel DL model based on a stacked convolutional autoencoder (SCAE) to address the challenge of limited data. The innovation of the SCAE model lies in its ability to enhance gradient information flow and extract richer hierarchical features, leading to superior diagnostic performance even with limited and noisy data samples. This article describes the development of a fault diagnosis method for a hydraulic piston pump using time-frequency visual pattern recognition. The proposed SCAE model has been evaluated on limited data samples of a hydraulic piston pump. The findings of the experiment demonstrate that the suggested approach can achieve excellent diagnostic performance with over 99.5% accuracy. Additionally, the SCAE model has outperformed traditional DL models such as deep neural networks (DNN), standard stacked sparse autoencoders (SSAE), and convolutional neural networks (CNN) in terms of diagnosis performance. Furthermore, the proposed model demonstrates robust performance under noisy data conditions, further highlighting its effectiveness and reliability.

Damage assessment in plate structures using an enhanced Jaya algorithm

The article proposes a new method called enhanced Jaya (EJaya) to identify and evaluate damage in plate structures. The method uses a nonlinear optimization task to solve the inverse damage identification problem. The damage magnitudes of the plate elements are taken as design variables, and the objective function to be minimized uses the variations in the plate flexibility matrix and includes a penalization term to avoid false alarm detection. EJaya is developed by incorporating three modifications in the traditional Jaya algorithm: generating an initial population of candidate solutions using prior knowledge of the damage scenario, integrating a component that takes into account the individual distribution through the search space, and implementing a new local search technique. The simulation results show that EJaya is efficient in both solution quality and computational effort and can detect and evaluate damage even in large-sized problems and under noisy data conditions.

Deep learning in attosecond metrology.

Time-resolved photoelectron spectroscopy provides a versatile tool for investigating electron dynamics in gaseous, liquid, and solid samples on sub-femtosecond time scales. The extraction of information from spectrograms recorded with the attosecond streak camera remains a difficult challenge. Common algorithms are highly specialized and typically computationally heavy. In this work, we apply deep neural networks to map from streaking traces to near-infrared pulses as well as electron wavepackets and extensively benchmark our results on simulated data. Additionally, we illustrate domain-shift to real-world data. We also attempt to quantify the model predictive uncertainty. Our deep neural networks display competitive retrieval quality and superior tolerance against noisy data conditions, while reducing the computational time by orders of magnitude.

Hybrid Framework for Speaker-Independent Emotion Conversion Using i-Vector PLDA and Neural Network

Expressive speech can be synthesized using acoustic feature modeling by mapping the spectral and fundamental frequency parameters between neutral speech and target emotions based on context. Speaker and text-independent emotion conversion are challenging modeling problems in this paradigm. In this paper, spectral mapping using an i-vector-based framework of fixed dimensions is proposed for the speaker-independent emotion conversion, considering the entire problem in the utterance domain, rather than the existing approaches using frame-level processing. The high dimensionality of i-vectors and reduced utterances for i-vector training necessitate the use of Probabilistic Linear Discriminant Analysis (PLDA) to derive the emotion dependent latent vector. The i-vector setup does not require parallel data or alignment procedures at any stage of training. $F_{0}$ training is conducted on a multilayer feed-forward neural network using limited aligned seed parallel data. The framework is tested on three different languages (datasets) viz. German (EmoDB), Telugu (IITKGP), and English (SAVEE). The proposed approach delivered superior performance compared to the baseline under both clean and noisy data conditions considered for analysis. Under clean data conditions, the proposed model was found to perform better than the baseline with a Mel Cepstral Distortion as low as 3.8 (fear), an $F_{0}$ -RMSE of 26.31 (happiness), and a Perceptual Evaluation of Speech Quality (PESQ) of 3.64 (anger) across datasets. Subjective testing provided a maximum CMOS of 4.10 (anger), 4.44 (fear), and 3.43 (happiness).

Mixed integer optimization approach to groundwater pollution source identification problems

ABSTRACTIn this study, we consider an approach that employs 0–1 type integer programming in a simulation-optimization model to identify the characteristics of groundwater pollutant sources. The advantage of the proposed solution model is that it reduces the dimensions of the unknown variable, where 0–1 type integer variables denote the source position. If the estimated location variable is equal to 0, this means that the corresponding location is not the actual source location and there is no need to estimate its source flux. If the estimated location variable is equal to 1, this means that the corresponding location is the actual source location and the source flux needs to be identified. We evaluated the performance of the proposed model based on four scenarios with different potential source location sets, incomplete data conditions, and noisy data conditions. The results obtained indicated that the proposed solution model is an effective approach for solving groundwater pollution source identification problems.

Finite element model calibration using frequency responses with damping equalization

Model calibration is a cornerstone of the finite element verification and validation procedure, in which the credibility of the model is substantiated by positive comparison with test data. The calibration problem, in which the minimum deviation between finite element model data and experimental data is searched for, is normally characterized as being a large scale optimization problem with many model parameters to solve for and with deviation metrics that are nonlinear in these parameters. The calibrated parameters need to be found by iterative procedures, starting from initial estimates. Sometimes these procedures get trapped in local deviation function minima and do not converge to the globally optimal calibration solution that is searched for. The reason for such traps is often the multi-modality of the problem which causes eigenmode crossover problems in the iterative variation of parameter settings. This work presents a calibration formulation which gives a smooth deviation metric with a large radius of convergence to the global minimum. A damping equalization method is suggested to avoid the mode correlation and mode pairing problems that need to be solved in many other model updating procedures. By this method, the modal damping of a test data model and the finite element model is set to be the same fraction of critical modal damping. Mode pairing for mapping of experimentally found damping to the finite element model is thus not needed. The method is combined with model reduction for efficiency and employs the Levenberg–Marquardt minimizer with randomized starts to achieve the calibration solution. The performance of the calibration procedure, including a study of parameter bias and variance under noisy data conditions, is demonstrated by two numerical examples.

Directivity Estimations for Short Dipole Antenna Arrays Using Radial Basis Function Neural Networks

The role of directivity is very important in the operation of an array as it gives a measure of the effectiveness of the array in pointing the radiations in a specific direction. Traditional methods used for the computation of directivity are although effective but may be time consuming. Artificial neural networks (ANNs) do not require the complex mathematical procedures and are therefore faster. Being nonlinear in nature, ANNs adapt to the nonlinear behavior of antenna arrays easily. In this letter, directivity estimations for the uniform linear arrays of collinear short dipoles and parallel short dipoles, using radial basis function neural networks (RBF-NNs) have been presented. The algorithm has also been applied for a planar array with short dipoles. The robustness of the method has been tested by evaluating its performance for noisy data conditions. The highlight features of the study are the accuracy and speed shown by the method in estimating results for the unseen inputs even in noisy data conditions.

Locally multidimensional scaling by creating neighborhoods in diffusion maps

This paper analyzes and improves an advanced multidimensional scaling method, known as locally multidimensional scaling, which assumes that high-dimensional data lie on a low-dimensional manifold. The method preserves local distances in the manifold by using classical scaling on a set of clusters in the high-dimensional data. These clusters are called neighborhoods, and the success of the method depends on the proper selection of these neighborhoods. At present, a neighborhood set is difficult to tune, and even if done well, the method may not function properly in dealing with noisy data. Our proposal utilizes clustering in a diffusion map, and thereby improves the original method in two ways. First, neighborhood selection is easier to tune, and second, the neighborhoods chosen enable the improved method to work under noisy data conditions. Our experiments demonstrate better tuning and robustness-to-noise results compared with the original method and some other existing multidimensional scaling methods on synthetic and real data sets.

An implicit radial basis function based reconstruction approach to electromagnetic shape tomography

In a reconstruction problem for subsurface tomography (modeled by the Helmholtz equation), we formulate a novel reconstruction scheme for shape and electromagnetic parameters from scattered field data, based upon an implicit Hermite interpolation based radial basis function (RBF) representation of the boundary curve. An object's boundary is defined implicitly as the zero level set of an RBF fitted to boundary parameters comprising the locations of few points on the curve (the RBF centers) and the normal vectors at those points. The electromagnetic parameter reconstructed is the normalized (w.r.t. the squared ambient wave number) difference of the squared wave numbers between the object and the ambient half-space. The objective functional w.r.t. boundary and electromagnetic parameters is set up and required Frechet derivatives are calculated. Reconstructions using a damped Tikhonov regularized Gauss–Newton scheme for this almost rank-deficient problem are presented for 2D test cases of subsurface landmine-like dielectric single and double-phantom objects under noisy data conditions. The double phantom example demonstrates the capability of our present scheme to separate out the two objects starting from an initial single-object estimate. The present implicit-representation scheme thus enjoys the advantages (and conceptually overcomes the respective disadvantages) of current implicit and explicit representation approaches by allowing for topological changes of the boundary curve, while having few unknowns respectively. In addition, the Hermite interpolation based RBF representation is a powerful method to represent shapes in three dimensions, thus conceptually paving the way for the algorithm to be used in 3D.

A Nonlinear Iterative Reconstruction and Analysis Approach to Shape-Based Approximate Electromagnetic Tomography

A nonlinear Helmholtz-equation-modeled electromagnetic tomographic reconstruction problem is solved for the object boundary and inhomogeneity parameters in a damped Tikhonov-regularized Gauss-Newton (DTRGN) solution framework. In this paper, the object is represented in a suitable global basis, whereas the boundary is expressed as the zero level set of a signed-distance function. For an explicit parameterized boundary-representation-based reconstruction scheme, analytical Jacobian and Hessian calculations are made to express the changes in scattered field values w.r.t. changes in the inhomogeneity parameters and the control points in a spline representation of the object boundary, via the use of a level-set representation of the object. Even though, in this paper, a homogeneous dielectric is considered and a spline representation has been used to represent the boundary, the formulation can be used for a general global basis representation of the inhomogeneity as well as arbitrary parameterizations of the boundary, and is generalizable to three dimensions. Reconstruction results are presented for test cases of landminelike dielectric objects embedded in the ground under noisy data conditions. To confirm convergence and, at times, to know which of the obtained iterates are closer to the actual unknown solution, using a perturbation theory framework, a local (Hessian-based) convergence analysis is applied to the DTRGN scheme for the reconstruction, yielding estimates of convergence rates in the residual and parameter spaces.

Analysis of electrical conductivity imaging

We investigate the feasibility of imaging the electrical conductivity in a cross‐section of an object (such as a core sample) by numerical inversion of low‐frequency, electromagnetic (EM) boundary data. Current flow is assumed to be confined to the cross‐section, which is modeled as a network of resistors. The network serves as a discrete approximation of the distributed‐parameter system that is described mathematically by Maxwell’s equations for steady current flow in a nonhomogeneous medium. A complete set of linearly independent voltage vectors is applied to the peripheral nodes, and the resulting node currents serve as the measured data for estimating the internal conductivity pattern (image). We generate estimates of this conductivity image by using an iterative process on network equations that are linearized in the unknown conductance variables. The mathematical feasibility of this approach is demonstrated by computer simulation studies using data generated from the network model. Reconstructed images are presented for sample conductance patterns under both ideal and noisy data conditions. An error analysis is performed to relate data noise to image‐estimation error.