- Research Article
42
- 10.1016/s0893-6080(96)00089-5
On Fault Tolerant Training of Feedforward Neural Networks
- Apr 01, 1997
- Neural Networks
- Behnam S Arad + 1 more +1
On Fault Tolerant Training of Feedforward Neural Networks
In this study, a data-driven deep learning model for fast and accurate prediction of temperature evolution and melting pool size of metallic additive manufacturing processes are developed. The study focuses on bulk experiments of the M4 high-speed steel material powder manufactured by Direct Energy Deposition. Under non-optimized process parameters, many deposited layers (above 30) generate large changes of microstructure through the sample depth caused by the high sensitivity of the cladding material on the thermal history. A 2D finite element analysis (FEA) of the bulk sample, validated in a previous study by experimental measurements, is able to achieve numerical data defining the temperature field evolution under different process settings. A Feed-forward neural networks (FFNN) approach is trained to reproduce the temperature fields generated from FEA. Hence, the trained FFNN is used to predict the history of the temperature fields for new process parameter sets not included in the initial dataset. Besides the input energy, nodal coordinates, and time, five additional features relating layer number, laser location, and distance from the laser to sampling point are considered to enhance prediction accuracy. The results indicate that the temperature evolution is predicted well by the FFNN with an accuracy of 99% within 12 seconds.
Loading PDF
On Fault Tolerant Training of Feedforward Neural Networks
On Fault Tolerant Training of Feedforward Neural Networks
3D FEA of Interfacial Stress in Tensile of Rubber-modified Polypropylene
The interfacial stress and its distribution affect directly the mechanical properties of polymeric blends and composites. Polymers are usually modified with rubber to improve their toughness. The three-dimensional (3D) finite element analysis (FEA) of the interfacial stress distribution between the particulate and the matrix during tensile of a rubber-modified polypropylene (PP) blend is made by means of an ANSYS software in this article, and the results are compared with those of the two-dimensional (2D) FEA simulation published earlier. The results show that the shear effect subjected in the pole zone of the rubber particle is maximal, the equator zone of the rubber particle is strongly pressed and pulled, and the tensile stress reaches the maximum. Comparatively, the stress distribution of 3D FEA is more gentle than that of 2D FEA.
Read moreThermal Modeling in Metal Additive Manufacturing Using Graph Theory
The goal of this work is to predict the effect of part geometry and process parameters on the instantaneous spatiotemporal distribution of temperature, also called the thermal field or temperature history, in metal parts as they are being built layer-by-layer using additive manufacturing (AM) processes. In pursuit of this goal, the objective of this work is to develop and verify a graph theory-based approach for predicting the temperature distribution in metal AM parts. This objective is consequential to overcome the current poor process consistency and part quality in AM. One of the main reasons for poor part quality in metal AM processes is ascribed to the nature of temperature distribution in the part. For instance, steep thermal gradients created in the part during printing leads to defects, such as warping and thermal stress-induced cracking. Existing nonproprietary approaches to predict the temperature distribution in AM parts predominantly use mesh-based finite element analyses that are computationally tortuous—the simulation of a few layers typically requires several hours, if not days. Hence, to alleviate these challenges in metal AM processes, there is a need for efficient computational models to predict the temperature distribution, and thereby guide part design and selection of process parameters instead of expensive empirical testing. Compared with finite element analyses techniques, the proposed mesh-free graph theory-based approach facilitates prediction of the temperature distribution within a few minutes on a desktop computer. To explore these assertions, we conducted the following two studies: (1) comparing the heat diffusion trends predicted using the graph theory approach with finite element analysis, and analytical heat transfer calculations based on Green’s functions for an elementary cuboid geometry which is subjected to an impulse heat input in a certain part of its volume and (2) simulating the laser powder bed fusion metal AM of three-part geometries with (a) Goldak’s moving heat source finite element method, (b) the proposed graph theory approach, and (c) further comparing the thermal trends predicted from the last two approaches with a commercial solution. From the first study, we report that the thermal trends approximated by the graph theory approach are found to be accurate within 5% of the Green’s functions-based analytical solution (in terms of the symmetric mean absolute percentage error). Results from the second study show that the thermal trends predicted for the AM parts using graph theory approach agree with finite element analyses, and the computational time for predicting the temperature distribution was significantly reduced with graph theory. For instance, for one of the AM part geometries studied, the temperature trends were predicted in less than 18 min within 10% error using the graph theory approach compared with over 180 min with finite element analyses. Although this paper is restricted to theoretical development and verification of the graph theory approach, our forthcoming research will focus on experimental validation through in-process thermal measurements.
Read moreElectricity Demand Forecasting Based on Feedforward Neural Network Training by a Novel Hybrid Evolutionary Algorithm
Electricity demand forecasting is an important index to make power development plan and dispatch the loading of generating units in order to meet system demand. In order to improve the accuracy of the forecasting, we apply the feedforward neural network for electricity demand forecasting. Inspired by the idea of Artificial Fish Swarm Algorithm, in this paper we proposed one hybrid evolutionary algorithm which based on PSO and AFSA methods through crossing over the PSO and AFSA algorithms to train the feedforward neural network. This proposed method has been applied in a real electricity load forecasting, the results show that the proposed approach has a better generalization performance and is also more accurate and effective than the feedforward neural network trained by particle swarm optimization.
Read moreDesign and Steady-State Performance of a Novel Winding Type Permanent Magnet Coupling with Slip Power Recovery Function
A novel winding type permanent magnet coupling (WTPMC) is proposed to work as an adjustable speed drive with slip power recovery function. As a kind of dual-mechanical-port electric machine with radial-flux configuration, the WTPMC consists of an outer rotor embedded with three-phase windings, an inner rotor populated with permanent magnets, and a slip power recovery circuit comprising a rectifier, a boost converter, and an ultracapacitor. The working principle of the WTPMC is presented, and its mathematical model is derived. To develop a WTPMC prototype for automotive applications, two-dimensional (2D) finite element analysis (FEA) is conducted using Ansoft Maxwell software to study the steady-state (constant slip speed) performance. For the experimental validation, the WTPMC prototype is manufactured and tested on a test bench. To show the accuracy of the 2D FEA, the computed results are compared with those obtained from experimental measurements. It is shown that the agreement between the 2D FEA and experimental results is good. Moreover, the WTPMC prototype can operate in the output speed range under different load torque conditions. The slip power recovery efficiency for the 2D FEA is 66.7%, while, for experimental measurements, it is 57.2%.
Read moreTraining feedforward neural networks with Bayesian hyper-heuristics
The process of training feedforward neural networks (FFNNs) can benefit from an automated process where the best heuristic to train the network is sought out automatically by means of a high-level probabilistic-based heuristic. This research introduces a novel population-based Bayesian hyper-heuristic (BHH) that is used to train feedforward neural networks (FFNNs). The performance of the BHH is compared to that of ten popular low-level heuristics, each with different search behaviours. The chosen heuristic pool consists of classic gradient-based heuristics as well as meta-heuristics (MHs). The empirical process is executed on fourteen datasets consisting of classification and regression problems with varying characteristics. The BHH is shown to be able to train FFNNs well and provide an automated method for finding the best heuristic to train the FFNNs at various stages of the training process.
Read moreHeterogeneous sensor-based condition monitoring in directed energy deposition
Heterogeneous sensor-based condition monitoring in directed energy deposition
An evaluation of quantum neural networks in the detection of epileptic seizures in the neonatal electroencephalogram
This paper presents the results of an experimental study that evaluated the ability of quantum neural networks (QNNs) to capture and quantify uncertainty in data and compared their performance with that of conventional feedforward neural networks (FFNNs). In this work, QNNs and FFNNs were trained to classify short segments of epileptic seizures in neonatal EEG. The experiments revealed significant differences between the internal representations created by trained QNNs and FFNNs from sample information provided by the training data. The results of this experimental study also confirmed that the responses of trained QNNs are more reliable indicators of uncertainty in the input data compared with the responses of trained FFNNs.
Read moreA study on mechanical properties of CuNi2SiCr layered on nickel–aluminum bronze via directed energy deposition
A study on mechanical properties of CuNi2SiCr layered on nickel–aluminum bronze via directed energy deposition
Homogenized tensile and bending properties of plain weave single‐Ply E‐glass/epoxy
This study empirically, numerically, and analytically analyzes the tensile and bending behavior of a plain weave single‐ply E‐glass/epoxy composite. In the empirical part of the study, tensile and simple bending experiments are conducted. Finite element method is used in the modeling part of the numerical study. Transverse and longitudinal fibers of the plain woven composite are modeled in one and three dimensions by using the ABAQUS software. Finally, mixture ratio approach and composite beam approach are considered in the analytical part of the study. Elastic modulus, Poisson's ratio, bending stiffness, and critical bending radius are obtained experimentally and are compared with the results of the finite element analyses and the results of the analytical study where applicable. The results closest to the empirically obtained ones are obtained by three‐dimensional (3D) finite elements analysis. Moreover, the mechanical characteristics obtained by the composite beam approach used in the analytical study are also very close to the values obtained by 3D finite elements analysis. POLYM. COMPOS., 2012. © 2012 Society of Plastics Engineers
Read moreA novel approach to thermal modelling and analysis in direct energy deposition for M2 steel alloy
Metal additive manufacturing (MAM) with its high material utilization efficiency to achieve the manufacturing of metal parts of complex shapes has great potential in the field of aerospace, medical, automotive and energy fields. MAM process generates large residual stress in the fabricated sample due to extreme mechanical, thermal and metallurgical interaction. Residual stress has a significant effect on dimensional stability. Thermal modelling of the MAM process is required to evaluate and implement the correct optimized process parameters for the quality fabrication module of industry. Thermal modelling is utmost necessary to fulfil the design for additive manufacturing (AM) of direct energy deposition (DED). Here, the fast design feedback loop is compiled by thermal modelling which reduces error in manufacturing. Thermal modelling of the DED AM process is carried out. As the material is temperature dependent, this depicts all the mechanical properties of the material given in the thermal modelling. Residual stress is generated, which has a significant effect on the dimension stability, corrosion resistance, crack growth resistance and mechanical properties of the fabricated sample through DED MAM. The present study performs a numerical simulation modelling for temperature-dependent M2 steel material as powder and SS304 as the substrate for the residual stress and temperature distribution of the defined laser power. The heat generated by laser power, i.e. Double Ellipsoidal Goldak's heat model is defined using Python script in the study. The study uses 800 W laser powers which generated a maximum residual stress of 179.9 MPa and yield stress of 321.7 MPa for the time interval of 50 s and also gives a maximum nodal temperature of 2233 K. The simulation will give us the details about the stresses of the material prior to development, i.e. without printing the actual component and will help to remove defects or printing failure to produce high-quality parts.
Read moreAccurate Prediction of Casing Collapse Resistance by 3D Finite Element Modeling
Three-dimensional (3D) outside diameter (OD) and wall thickness mapping of a casing joint include synchronized OD laser scanning and wall UT (ultrasonic testing) measurement at a resolution of every 1-inch along the length and every 1 or 2 degrees around the circumference. This full-length mapping data can then be used to construct a 3D finite element analysis (FEA) model that captures the geometric attributes, including OD, wall thickness, OD ovality, and wall eccentricity of a casing joint. Together with the actual nonlinear material plasticity model and the measured residual stress, the 3D FEA model can accurately predict the casing collapse strength. Also, the Klever-Tamano ultimate limit-state (KT ULS) collapse equation per API TR 5C3 [1] together with a calibrated moving-average scheme was used to calculate the joint's collapse strength. Results predicted by 3D FEA and KT ULS collapse models were then compared to the actual full-scale collapse test results. A 3D FEA model generally produces more accurate results than the KT collapse equation. In addition, both 3D FEA and KT collapse models predict slightly lower than actual collapse pressure, which therefore is on the conservative side. The accurate prediction of casing collapse strength can be beneficial to the end user in sorting and ranking the casing joints by collapse resistance, and therefore provides valuable information that can be used in the end user's risk assessment with the consideration of the worst-case discharge scenario. Furthermore, the end user may potentially select the joints with the highest predicted collapse strength for the bottom portion of the casing string where the highest collapse resistance is desired. In addition, the OD and wall mapping data together with the actual material tensile properties (yield strength, ultimate tensile strength, and strain hardening parameters) can be used in estimating the internal yield and ductile rupture pressures and other performance properties. This is outside the scope of this paper and can be pursued in future work. These predicted casing performance properties may then be used in sorting and ranking the joints based on a specific performance property of interest or combination of performance properties.
Read moreA 3D nonlinear FEA model for simulation of microbubble behavior
Microbubbles provide the basis for perfusion measurement, early detection of molecular signatures of disease and as a vehicle for drug, or gene, delivery. Although the response of a single microbubble to ultrasound has been characterized by radially symmetric 1D models and high speed camera-based experiments, a plethora of mechanisms, such as bubble-bubble interaction, and adherent bubble response are poorly understood. With few exceptions, 1D model are generally valid at moderate acoustic pressures, but are incapable of predicting phenomena such as higher order mode oscillations or onset of bubble shell rupture. It has been hypothesized that second or higher order mode oscillations are a precursor to bubble shell rupture and subsequent cavitation. In addition, parameters, such as shell stress and shell strain may have threshold values related to bubble shell fracture. In this work, we propose a full 3D FEA (finite element analysis) model that can simulate temporal variations in asymmetric 3D radial, translational bubble motion, various shell parameters such as regional stress /strain, and bubble pulse echo response. Preliminary simulations were conducted to assess the performance of the model for broadband cases, including: 1) single cycle pulse excitation, 2) single cycle of free and wall-bound bubble, 3) predicting translational displacements in response to primary radiation force, and 4) tracking variations in shell stresses and strain for low (180 KPa) and high (330 KPa) peak negative pressure acoustic waveforms. For the acoustic and bubble parameters published by Dayton, the maximum displacement predicted using the 3D FEA model was within 10% of the previously published experimental data. For approximately similar bubble and acoustic parameters (using inverted pulses), the spectrum of the backscattered data from a free bubble was found to be similar to the experimental and simulated results The spectrum of the adherent bubble response was found to be downshifted with respect to that of a free bubble; similar results have been reported by Payne et al . The adherent bubble was found to oscillate asymmetrically in the planes orthogonal to the azimuth-elevation plane or the plane perpendicular to the face of the transducer (edge of the model where an external pulse is applied).
Read moreThe stress distribution of implants made of Polyaryletherketones: A 3D Finite Element Analysis
OBJECTIVE: To explore the stress distribution on the bone-implant structures caused by the Polyetheretherketone (PEEK), Polyetherketoneketone (PEKK), and titanium fixture/abutments by using the three-dimensional (3D) finite element analysis (FEA). MATERIALS AND METHODS: Six models composed of titanium, PEEK, and PEKK implant/abutments under vertical (250 N) and 45˚ oblique (100 N) loading were studied. The obtained principal and von Mises stress values from the models were evaluated. RESULTS: von Mises stresses were found to be highest in titanium implants and abutments under vertical and oblique loads. Extremely increased stress values were observed in the screws of the Polyaryletherketone (PAEK) models compared to titanium models. Lower principal stresses were observed in titanium models than PAEK models in the cancellous and cortical bone under vertical and oblique loads. CONCLUSION: PAEKs transmitted more stress to the peri-implant bone. Stress distribution in titanium models were more homogenous while stress concentrated in the bone adjacent to the coronal part of the implants and neck of the implants in PAEKs.
Read moreStarting Characteristic Analysis of the V-Shaped Interior Permanent Magnet Motor
This research paper deals with the electromagnetic analysis of the V-shaped interior permanent magnet (IPM) motor for the hybrid electric vehicle (HEV). Like Toyota prius III traction motor, a V-typed single-layered IPM motor is investigated. A two-dimensional (2D) finite element analysis (FEA) methodology is discussed and adopted to compute the electromagnetic characteristics of the IPM motor, at no load conditions, with and without stator current excitation, and accelerating from starting to constant speed conditions. The cogging torque of the IPM motor at no load and zero stator current is calculated to ascertain the dynamic stability as well as instability positions of the motor that is of great essence in understanding the startup behavior and the overall performance of the motor. An existing control method modulates the stator supply current, depending on the position of the rotor of the IPM motor. This control approach is used to provide a fixed phase difference between the rotor and the stator magnetic fields so that the IPM motor can smoothly accelerate as it goes into the constant speed. The electromagnetic torque oscillates during the IPM motor's acceleration and despite of the fluctuations, the V-shaped IPM motor demonstrates the reliable performance, and accelerating and stabilizing at its target speed. The 2D FEA is utilized to investigate the IPM motor's dynamic behavior during the startup for its potential application in hybrid electric vehicle (HEV) traction systems.
Read more