Showing papers by "Sudip Dey published in 2021"

Stochastic Oblique Impact on Composite Laminates: A Concise Review and Characterization of the Essence of Hybrid Machine Learning Algorithms

Abstract: Due to the absence of adequate control at different stages of complex manufacturing process, material and geometric properties of composite structures are often uncertain. For a secure and safe design, tracking the impact of these uncertainties on the structural responses is of utmost significance. Composite materials, commonly adopted in various modern aerospace, marine, automobile and civil structures, are often susceptible to low-velocity impact caused by various external agents. Here, along with a critical review, we present machine learning based probabilistic and non-probabilistic (fuzzy) low–velocity impact analyses of composite laminates including a detailed deterministic characterization to systematically investigate the consequences of source- uncertainty. While probabilistic analysis can be performed only when complete statistical description about the input variables are available, the non-probabilistic analysis can be executed even in the presence of incomplete statistical input descriptions with sparse data. In this study, the stochastic effects of stacking sequence, twist angle, oblique impact, plate thickness, velocity of impactor and density of impactor are investigated on the crucial impact response parameters such as contact force, plate displacement, and impactor displacement. For efficient and accurate computation, a hybrid polynomial chaos based Kriging (PC-Kriging) approach is coupled with in-house finite element codes for uncertainty propagation in both the probabilistic and non- probabilistic analyses. The essence of this paper is a critical review on the hybrid machine learning algorithms followed by detailed numerical investigation in the probabilistic and non-probabilistic regimes to access the performance of such hybrid algorithms in comparison to individual algorithms from the viewpoint of accuracy and computational efficiency.

Compound influence of topological defects and heteroatomic inclusions on the mechanical properties of SWCNTs

Abstract: The utility of carbon nanotubes as the reinforcement agents in polymer and metal matrix composites has opened up a new avenue in the development of novel composite materials with exceptional strength and stiffness to weight ratios. Such exploitation of superior mechanical properties of carbon nanotubes depends on their inherent irregularities and structural integration. The nanotubular structures of carbon are prone to topological defects and heteroatom dopants due to the inevitable complexities in nano-synthesis. The objective of this article is to quantify the compound influence of such inherent structural irregularities (such as single vacancy defects and nanopores) and foreign atom inclusions (such as nitrogen and boron atoms) on the mechanical characteristics (like constitutive relation, fracture strength, failure strain and Young’s moduli) of single-walled carbon nanotubes (SWCNT) under various multi-physical influences (such as temperature, strain rate, diameter and chirality) based on molecular dynamics (MD) simulations. The current investigation also includes a detailed analysis on the variation in mechanical characteristics of CNTs under different spatial distributions of defects and doping.

Probing the Stochastic Dynamics of Coronaviruses: Machine Learning Assisted Deep Computational Insights with Exploitable Dimensions

Abstract: A machine learning assisted efficient, yet comprehensive characterization of the dynamics of coronaviruses, in conjunction with finite element (FE) approach, is presented. Without affecting the accuracy of prediction in low-frequency vibration analysis, an equivalent model for the FE analysis is proposed, based on which the natural frequencies corresponding to first three non-rigid modes are analyzed. To quantify the inherent system-uncertainty efficiently, Monte Carlo simulation is proposed in conjunction with the machine learning based FE computational framework for obtaining complete probabilistic descriptions considering both individual and compound effect of stochasticity. A variance based sensitivity analysis is carried out to enumerate the relative significance of different material parameters corresponding to various constituting parts of the coronavirus structure. Using the modal characteristics like natural frequencies and mode shapes of the virus structure including their stochastic bounds, it is possible to readily identify coronaviruses by comparing the experimentally measured dynamic responses in terms of the peaks of frequency response function. Results from this first of its kind study on coronaviruses along with the proposed generic machine learning based approach will accelerate the detection of viruses and create efficient pathways toward future inventions leading to cure and containment in the field of virology.

Modeling and combined application of MOEA/D and TOPSIS to optimize WEDM performances of A286 superalloy

Abstract: Superalloys are categorized as difficult to process materials with a broad spectrum of applications in industries. Process modeling and optimization of WEDM performances on nickel- and titanium-based superalloys are widely investigated. However, such investigations on iron-based superalloy are still lacking and hence probed in the present article. Thus, the paper targets modeling the correlation between the performance parameters and the control parameters with two popular techniques: response surface methodology (RSM) and artificial neural network (ANN) for WEDM of a typical iron-based superalloy, i.e., A286 superalloy. A comparison between the model estimates and the experimental values is made to check ANN and RSM's prediction accuracy. The estimates by the ANN model are exact and consistent with the experimental results. An analysis of variance (ANOVA) test is performed to perceive the degree of statistical significance of parameters. Moreover, a novel two-stage procedure, i.e., MOEA/D in collaboration with TOPSIS method, is implemented to search the optimal condition for process performances. The quality of Pareto-optimal solutions acquired using MOEA/D is compared to that of Pareto-optimal solutions obtained using NSGA II, PESA II, and MMOPSO through the use of a hypervolume (HV) parameter. Wilcoxon’s test is performed to identify the statistical difference between MOEA/D and competing algorithms. The optimal parametric combination recommended by the proposed optimization approach is Ton = 130 µs, Toff = 52 µs, Ipeak = 12 A, Wf = 5 m/min and SV = 30 V. The proposed optimization technique can also be exploited in other manufacturing processes.

Machine learning-based performance analysis of two-axial-groove hydrodynamic journal bearings:

Abstract: The precise prediction of a rotor against instability is needed for avoiding the degradation or failure of the system’s performance due to the parametric variabilities of a bearing system In gener

Comparative evaluation on probabilistic performance of journal bearing: a surrogate-based approach

Abstract: This paper presents a comparative assessment of surrogate models in probabilistic analysis of finite plain journal bearing. Traditional Monte Carlo simulation (MCS) is employed dealing with a large number of data for validating the constructed surrogate model dealing with a limited number of data. In the case of a complex engineering system like journal bearing with finite length wherein no analytical solution exists and the system needs to be solved based on expensive numerical techniques such as finite difference and finite element method due to unavailability of a large number of experimental data. Hence, surrogate models show their computational efficiency complying with the accuracy of the models. Thus, the present study aims to investigate the applicability of five surrogate models such as moving least square, support vector machine, radial basis function, polynomial neural network and multivariate adaptive regression splines in terms of their efficiency and accuracy. A probabilistic analysis approach combining the finite difference method, surrogate models and MCS is presented in this work. The validation and parametric results corresponding to the comparison of the constructed surrogate models are presented. Substantial intuitive new results are conferred in the probabilistic surrogate schemes.

Support Vector Model Based Thermal Uncertainty on Stochastic Natural Frequency of Functionally Graded Cylindrical Shells

Abstract: This paper presents the effect of temperature on stochastic natural frequencies of cylindrical shells, composed of functionally graded materials (FGM) by using machine learning quadratic Support Vector Machine (SVM). An eight noded isoperimetric quadratic element is considered for the finite element formulation. The power law is employed to construct the material modelling of FGM cylindrical shells. Monte Carlo Simulation (MCS) is carried out in conjunction with stochastic eigenvalue solution. In the present study, zirconia (ceramic) and aluminium (metal) are considered to compose the FGM. The machine learning SVM model is constructed to reduce the computational iteration time and cost and validated with the traditional MCS model. The statistical analyses are conducted to portray the first three modes of frequencies. The results show that due to the increase of the temperature, the values of both deterministic as well as the stochastic mean of the first three natural frequencies decreases along with the decrease in sparsity. Sensitivity analysis is also carried out to enumerate the significant important input parameters contributing to influence the output quantity of interest (QoI). The statistical results obtained are the first known results.

Interactions- Readdressing the issue

Abstract: Broadly drugs include all the chemical substances excluding food that affect the bodily processes. The drug is considered to be a medicine if it benefits the body. Whereas, if the drug is injurious to the body, it’s considered as a poison. Therefore, the same chemical can be a boon or curse with respect to the situation, condition of use, dosage and the individual using it. In this contemporary healthcare era, a huge number of medications are formulated each year and new interactions between drugs are reported every now and then. As a result, it is no more practical for doctors to be dependent on the memory alone to avoid possible drug interactions. Changes in absorption, distribution, metabolism or elimination of drugs are referred to as pharmacokinetic interactions, resulting in alteration in the level of drugs and its metabolites. The effect of drug changes from person to person than expected because it causes different reaction when a drug reacts with the food or dietary supplements they take (drug -food interaction). So, the effect of the drug is altered by means of increasing, decreasing, or producing a new effect which cannot be produced on its own the effect caused by food or dietary supplements. These interactions may occur due to accidental misuse or due to other factors such as lack of knowledge about it. This review provides a comprehensive literature review on various drug interaction. Generally, drug food interactions are neglected and not well defined but it can cause mild to serious effects. However, all clinicians, pharmacists and nurses should be aware of drug interaction to avoid the consequences caused by drug interactions.

Effect of Eccentricity and Surface Roughness on Probabilistic Performance of Two Axial Groove Bearing

Abstract: This paper presents the effect of eccentricity and surface roughness on the probabilistic performance of two axial groove hydrodynamic journal bearing. In general, it is difficult to quantify experimentally the variabilities involved in dynamic responses of the hydrodynamic bearing due to the randomness involved in surface asperity and eccentricity ratio. The deterministic models available for the analysis of the bearings are not capable to include such uncertainties. These uncertainties arise from the manufacturing imperfections, misalignment of the bearing, frictional wear, uncertain operating condition, model inaccuracy. To simulate such variabilities, Monte Carlo simulation (MCS) is carried out. Stochastic steady-state and dynamic coefficients are obtained by solving the Reynolds equation using the surrogate-based finite difference method. Sensitivity analysis of the performance parameters with respect to stochastic input parameters is portrayed. The moving least square (MLS) model is constructed as the surrogate to increase the computational efficiency. The significant influences of stochastic input parameters such as surface roughness and eccentricity ratio are observed on the random hydrodynamic performance of two axial groove journal bearing.

ANN-Based Random First-Ply Failure Analyses of Laminated Composite Plates

Abstract: This paper presents the random first-ply failure analyses of laminated composite plates by using an artificial neural network (ANN)-based surrogate model. In general, materials and geometric uncertainties are unavoidable in such structures due to their inherent anisotropy and randomness in system configuration. To map such variabilities, stochastic analysis corroborates the fact of inevitable edge towards the quantification of uncertainties. In the present study, the finite element formulation is derived based on the consideration of eight-noded elements wherein each node consists of five degrees of freedom (DOF). The five failure criteria namely, maximum stress theory, maximum strain theory, Tsai-Hill (energy-based criterion) theory, Tsai-Wu (interaction tensor polynomial) theory and Tsai-Hill’s Hoffman failure criteria are considered in the present study. The input parameters include the ply orientation angle, assembly of ply, number of layers, ply thickness and degree of orthotropy, while the first-ply failure loads for five criteria representing output quantity of interest. The deterministic results are validated with past experimental results. The results obtained from the ANN-based surrogate model are observed to attain fitment with the results obtained by Monte Carlo Simulation (MCS). The statistical results are presented for both deterministic, as well as stochastic domain.

Surrogate Model Validation and Verification for Random Failure Analyses of Composites

Abstract: In the present chapter, multivariate adaptive regression splines (MARS) is explored as a surrogate model in conjunction to Monte Carlo simulation (MCS) to analyse the random first-ply failure loads of graphite–epoxy laminated composite plates. The five failure criteria, namely maximum strain theory, maximum stress theory, Tsai–Hill theory, Tsai–Wu theory, and Hoffman theory, are considered. The numerical validation of deterministic failure load is presented first. Thereafter, a concise investigation is carried out to examine the capability of MARS model for efficiently predicting the first-ply failure loads. Comparative results are presented using scatter plots and probability density function plots to access the prediction capability with respect to direct MCS. The current results portray the successful application of MARS as the surrogate model to achieve computational efficiency and analyse the first-ply failure loads.

Dynamic Sensitivity Analysis of Random Impact Behaviour of Hybrid Cylindrical Shells

Abstract: The present chapter investigates the moment-independent sensitivity analysis for hybrid sandwich structures (having cylindrical shell geometry) subjected to low-velocity impact. These hybrid structures are extensively used in lightweight applications where thermal exposure/resistance is of prime importance. Here, the FG facesheet is placed at the upper layer of core whereas laminated composite facesheet is kept at lower layer so that the structure can sustain high-temperature exposure at reduced weight because of laminated composite facesheet at the inner layer of the core. The probabilistic study is performed for the transient impact response of the structure which in turn utilized to assess the sensitiveness of the parameters. The computational efficiency is achieved by implementing polynomial chaos expansion (PCE) metamodel in conjunction with Monte Carlo simulation (MCS). The results illustrate the parameters which significantly affect the transient impact response of the structure.

Experimental Investigation of Dimensional Deviation of Square Punches Cut by Wire EDM of A286 Superalloy

Abstract: A286 superalloy is a Fe–Ni-based superalloy is widely applicable in superchargers, gas turbines, jet engines, fasteners, after burner and turbine wheels because of its innate properties like high thermal resistance, high mechanical strength and substantial corrosion resistance. This study is particularly devoted to suggest optimal process parameter settings for this superalloy, where the dimensional deviation becomes minimum value after wire EDM cut. Measurement of dimensional deviation (DD) is important as it suggests the practitioners to set a proper wire offset during the CNC programming of wire tool path so that the dimension of the product after the Wire EDM cut and the required dimension of the product matches properly. Five important parameters such as pulse on time (Ton), pulse off time (Toff), peak current (Ipeak), wire feed rate (Wf) and spark gap set voltage (SV) are controlled during the experiments, and the experimental layout is designed by L27 orthogonal array. Taguchi method in conjunction with ANOVA is adopted to obtain the significant control parameters. Parametric effect of the control factors on the dimensional deviation is explained. Furthermore, the optimal levels of the control factors are recommended based on higher signal-to-noise ratio values. Comparison between the experimental and the predicted values is evaluated to show reproducibility.

Probabilistic Oblique Impact Analysis of Functionally Graded Plates – A Multivariate Adaptive Regression Splines Approach

Abstract: Purpose: To investigate the probabilistic low-velocity impact of functionally graded (FG) plate using the MARS model, considering uncertain system parameters. Design/methodology/application: The distribution of various material properties throughout FG plate thickness is calculated using power law. For finite element (FE) formulation, isoparametric elements with eight nodes are considered, each component has five degrees of freedom. The combined effect of variability in material properties such as elastic modulus, modulus of rigidity, Poisson’s ratio, and mass density are considered. The surrogate model is validated with the FE model represented by the scatter plot and the probability density function (PDF) plot based on Monte Carlo simulation (MCS). Findings: The outcome of the degree of stochasticity, impact angle, impactor’s velocity, impactor’s mass density, and point of impact on the maximum value of contact force (CFmax ), plate deformation (PDmax), and impactor deformation (IDmax ) are determined. A convergence study is also performed to determine the optimal number of the constructed MARS model’s sample size. Originality/value: The results illustrate the significant effects of uncertain input parameters on FGM plates’ low-velocity impact responses by employing a surrogate-based MARS model.

Temperature-Dependent Random Frequency of Functionally Graded Spherical shells—A PCE Approach

Abstract: This paper presents the effect of temperature on random natural frequencies of spherical shells, composed of functionally graded materials (FGM) with zirconia (ceramic rich) and aluminium (metal rich). An eight noded isoperimetric quadratic element is considered for the finite element formulation. The power law is employed to construct the material modelling of Functionally Graded (FG) spherical shells. Monte Carlo Simulation (MCS) is carried out in conjunction to standard eigenvalue problems. The polynomial chaos expansion (PCE) model is constructed to reduce the computational iteration time and cost and validated it with the traditional MCS model. The statistical analyses are conducted to portray the first three random modes of frequencies. In the present analysis, the statistical results obtained are the first known results.

Quantifying Uncertainty in Structural Responses of Polymer Sandwich Composites: A Comparative Analysis of Neural Networks

Abstract: The manufacturing and fabrication of complex polymer sandwich composite plates involve various processes and parameters, and the lack of control over them causes uncertain system parameters. It is essential to consider randomness in varying parameters to analyse polymer sandwich composite plates. The present study portrays uncertainty quantification in structural responses (such as natural frequencies) of polymer sandwich composite plates using the surrogate model. The comparative study of artificial neural network (ANN) and polynomial neural network (PNN) for uncertain structural responses of the sandwich plate is presented. The proposed ANN as well as PNN algorithm is found to be convergent with intensive Monte Carlo simulation (MCS) for uncertain vibration responses. The predictability of PNN is observed to be more efficient than that of ANN. Typical material properties, skew angle, fibre orientation angle, number of laminate and core thickness are randomly varied to quantify the uncertainties. The use of both the surrogate models (PNN and ANN) results in a significant saving of computational time and cost compared to that of full-scale intensive finite element-based MCS approach.

Effect of Delamination on Natural Frequencies of E-glass and S-glass Epoxy Composite Plates

Abstract: The delamination is one of the major modes of failure occurring in the laminated composite due to insufficient bonding between the layers. In this paper, the natural frequencies of delaminated S-glass and E-glass epoxy cantilever composite plates are presented by employing the finite element method (FEM) approach. The rotary inertia and transverse shear deformation are considered in the present study. The effect of parameters such as the location of delamination along the length, across the thickness, the percentage of delamination, and ply-orientation angle on first three natural frequencies of the cantilever plates are presented for S-glass and E-glass epoxy composites. The standard eigenvalue problem is solved to obtain the natural frequencies and corresponding mode shapes. First three mode shape of S-Glass and E-Glass epoxy laminated composites are portrayed corresponding to different ply angle of lamina.

Polynomial Neural Network Based Stochastic Natural Frequency Analysis of Functionally Graded Plates

Abstract: The present article deals with the stochastic approach for natural frequency (NF) analysis of functionally graded (FG) plates by employing polynomial neural network (PNN) surrogate model combined with finite element (FE) method. The surrogate model for NF analysis of FG plates is validated with the original FE method. Both individual and mixed variation of material properties are taken into account. The present PNN model significantly rises the computational efficiency, and the computational cost decreased in comparison to Monte Carlo Simulation (MCS).