Artificial Neural Network-based Research Articles

An artificial neural network-based model for roping prediction in aluminum alloy sheet

Roping is a common macroscopic surface defect in AA6XXX aluminum alloy sheets resulting from the three-dimensional (3-D) spatial distribution of specific textures. The crystal plasticity finite element model (CPFEM) has been applied to study roping behavior and achieves good agreement with experimental observations but suffers from expensive computational consumption due to large-area and multi-layer characteristics of the roping phenomenon. In the present work, an artificial neural network-based (ANN) model is developed to predict thickness strain and corresponding surface roping under tensile deformation. A 3-D artificial orientation map generation algorithm is proposed and combined with CPFEM to produce reliable datasets to train, test, and validate the ANN model. The layer-by-layer random texture replacement strategy is adopted for dataset generation and facilitates the ANN model to capture the effects of individual layer textures on the thickness strain. A novel exponential weight loss function is also introduced to solve the imbalance problem of texture components. The ANN-based model demonstrates good prediction capability and generalizability in multiple validation cases with various artificial and experimental textures. The proposed ANN-based model provides an efficient and accurate alternative to the conventional physics-based method for roping analysis in aluminum alloy and can be also applied to similar microstructure-related deformation prediction in other materials.

Modeling the Quantitative Assessment of the Condition of Bridge Components Made of Reinforced Concrete Using ANN

Bridges in Ukraine are one of the most important components of the infrastructure, requiring attention from government agencies and constant funding. The object of the study was the methodology for quantifying the condition of bridge components. The Artificial Neural Network-based (ANN) tool was developed to quantify the technical condition of bridge components. The literature analysis showed that in most cases the datasets were obtained during the inspection of bridges to solve the problems of assessing the current technical condition. The lack of such a database prompted the creation of a dataset on the basis of the Classification Tables of the Operating Conditions of the Bridge Components (CT). Based on CTs, five datasets were formed to assess the condition of the bridge components: bridge span, bridge deck, pier caps beam, piers and abutments, approaches. The next step of this study was creating, training, validating and testing ANN models. The network with ADAM loss function and softmax activation showed the best results. The optimal values of MAPE and R2 were achieved at the 100th epoch with 64 neurons in the hidden layer and were equal to 0.1% and 0.99998, respectively. The practical application of the ANN models was carried out on the most common type of bridge in Ukraine, namely, a road beam bridge of small length, made of precast concrete. The novelty of this study consists of the development of a tool based on the use of ANN model, and the proposal to modify the methodology for quantifying the condition of bridge components. This will allow minimizing the uncertainties associated with the subjective judgments of experts, as well as increasing the accuracy of the assessment.

Artificial Neural Network Model for Estimating Ocean Heat Content in the Sea Ice-Covered Arctic Regions Using Satellite Data

Ocean heat content (OHC) is an essential parameter to assess Earth’s energy imbalance, global warming, and climate change over the historical record. An accurate estimate of the OHC in the Arctic sea ice regions is challenging due to the lack of in-situ data and satellite-based algorithms. In this study, an artificial neural network-based (ANN) novel approach is presented for estimating OHC changes at various standard depth extents (by World Ocean Atlas) in the Arctic sea ice regions based on the relationships of the sea ice thermodynamic parameters from the satellite measurements and in-situ OHC estimates. Because of the potential uncertainty that arises from the inaccessible near-surface oceanic layer in the in-situ OHC estimates, a preliminary ANN model was developed with a set of approximations to account for the in-situ OHC stored in the respective depths within the inaccessible near-surface oceanic layer. The ANN model architecture was optimized for a depth extent of 700 m and adopted for the remaining depths of 20 m, 30 m, 40 m, 50 m, 100 m, 150 m, 200 m, 250 m, 300 m, 350 m, 400 m, 450 m, 500 m, 550 m, 600 m, and 650 m. The new model was robust in capturing the spatial, temporal, and depth variabilities of OHC in the sea ice-covered Arctic regions with greater accuracy (mean bias error 0.022 GJ m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-2</sup> , mean bias percentage error 0.015%, mean absolute error 0.182 GJ m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-2</sup> , mean absolute percentage error 0.148%, root mean square error 0.24 GJ m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-2</sup> , R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> 0.94, slope 0.93, and intercept 25.05 GJ m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-2</sup> ). This model is capable of estimating OHC and its temporal trends from satellite data which will have implications for understanding the global climate change and its impacts in the Polar Oceans.

Combustion Tuning for a Gas Turbine Power Plant Using Data-Driven and Machine Learning Approach

Abstract Conventional physics-based or experimental-based approaches for gas turbine combustion tuning are time consuming and cost intensive. Recent advances in data analytics provide an alternative method. In this paper, we present a cross-disciplinary study on the combustion tuning of an F-class gas turbine that combines machine learning with physics understanding. An artificial-neural-network-based (ANN) model is developed to predict the combustion performance (outputs), including NOx emissions, combustion dynamics, combustor vibrational acceleration, and turbine exhaust temperature. The inputs of the ANN model are identified by analyzing the key operating variables that impact the combustion performance, such as the pilot and the premixed fuel flow, and the inlet guide vane angle. The ANN model is trained by field data from an F-class gas turbine power plant. The trained model is able to describe the combustion performance at an acceptable accuracy in a wide range of operating conditions. In combination with the genetic algorithm, the model is applied to optimize the combustion performance of the gas turbine. Results demonstrate that the data-driven method offers a promising alternative for combustion tuning at a low cost and fast turn-around.

Partial pruning strategy for a dual-branch multilayer perceptron-based post-equalizer in underwater visible light communication systems.

We demonstrate a partial pruning strategy for a post-equalizer based on a dual-branch multilayer perceptron-based post-equalizer (DBMLP PE) in an underwater visible light communication (UVLC) system. The partial pruning strategy produces a sparse DBMLP PE (SDBMLP PE) with less space complexity than the Volterra PE and bit error rate (BER) performance similar to the DBMLP PE. We experimentally prove the effectiveness and necessity of the partial pruning algorithm in multilayer perceptron-based (MLP) PE. The partial pruning strategy consists of two parts: (i) preventing the pruning of connections to the output nodes, and (ii) avoiding the linear mapping branches of the DBMLP during the pruning process. Our experiments prove that the SDBMLP PE further reduces the BER of the UVLC system by 36.5%, with only 33.8% parameters of the Volterra PE. To the best of our knowledge, this paper presents the first deep neural network-based PE with lower complexity and higher BER performance than the Volterra PE in the UVLC system, which dramatically increases the applicability of artificial neural network-based (ANN) PE in the field of UVLC systems.

Kankanet: An artificial neural network-based object detection smartphone application and mobile microscope as a point-of-care diagnostic aid for soil-transmitted helminthiases.

BackgroundEndemic areas for soil-transmitted helminthiases often lack the tools and trained personnel necessary for point-of-care diagnosis. This study pilots the use of smartphone microscopy and an artificial neural network-based (ANN) object detection application named Kankanet to address those two needs.Methodology/Principal findingsA smartphone was equipped with a USB Video Class (UVC) microscope attachment and Kankanet, which was trained to recognize eggs of Ascaris lumbricoides, Trichuris trichiura, and hookworm using a dataset of 2,078 images. It was evaluated for interpretive accuracy based on 185 new images. Fecal samples were processed using Kato-Katz (KK), spontaneous sedimentation technique in tube (SSTT), and Merthiolate-Iodine-Formaldehyde (MIF) techniques. UVC imaging and ANN interpretation of these slides was compared to parasitologist interpretation of standard microscopy.Relative to a gold standard defined as any positive result from parasitologist reading of KK, SSTT, and MIF preparations through standard microscopy, parasitologists reading UVC imaging of SSTT achieved a comparable sensitivity (82.9%) and specificity (97.1%) in A. lumbricoides to standard KK interpretation (97.0% sensitivity, 96.0% specificity). The UVC could not accurately image T. trichiura or hookworm. Though Kankanet interpretation was not quite as sensitive as parasitologist interpretation, it still achieved high sensitivity for A. lumbricoides and hookworm (69.6% and 71.4%, respectively). Kankanet showed high sensitivity for T. trichiura in microscope images (100.0%), but low in UVC images (50.0%).Conclusions/SignificanceThe UVC achieved comparable sensitivity to standard microscopy with only A. lumbricoides. With further improvement of image resolution and magnification, UVC shows promise as a point-of-care imaging tool. In addition to smartphone microscopy, ANN-based object detection can be developed as a diagnostic aid. Though trained with a limited dataset, Kankanet accurately interprets both standard microscope and low-quality UVC images. Kankanet may achieve sensitivity comparable to parasitologists with continued expansion of the image database and improvement of machine learning technology.

A Holistic Approach Towards Intelligent Hotspot Prevention in Network-on-Chip-Based Multicores

Traffic hotspots, a severe form of network congestion, can be caused unexpectedly in a network-on-chip (NoC) due to the immanent spatio-temporal unevenness of application traffic. Hotspots reduce the NoC's effective throughput, where in the worst-case scenario, network traffic flows can be frozen indefinitely. To alleviate this problematic phenomenon several adaptive routing algorithms employ online load-balancing schemes, aiming to reduce the possibility of hotspots arising. Since most are not explicitly hotspotagnostic, they cannot completely prevent hotspot formation(s) as their reactive capability to hotspots is merely passive. This paper presents a pro-active Hotspot-Preventive Routing Algorithm (HPRA) which uses the advance knowledge gained from network embedded artificial neural network-based (ANN) hotspot predictors to guide packet routing in mitigating any unforeseen near-future hotspot occurrences. First, these ANN-based predictors are trained offline and during multicore operation they gather online statistical data to predict about-to-be-formed hotspots, promptly informing HPRA to take appropriate hotspot-preventive action(s). Next, in a holistic approach, additional ANN training is performed with data acquired after HPRA interferes, so as to further improve hotspot prediction accuracy; hence, the ANN mechanism does not only predict hotspots, but is also aware of changes that HPRA imposes upon the interconnect infrastructure. Evaluation results, including utilizing real application traffic traces gathered from parallelized workload executions onto a chip multiprocessor architecture, show that HPRA can improve network throughput up to 81 percent when compared with prior-art. Hardware synthesis results affirm the HPRA mechanism's moderate overhead requisites.

A new approach to predict the excitation current and parameter weightings of synchronous machines based on genetic algorithm-based k-NN estimator

This paper presents a novel and efficient solution to overcome difficulties in excitation current estimation and parameter weighting of synchronous motors. Weighting the parameters or searching the best coefficients of problems is commonly accomplished through intuitive/heuristic approaches. For this reason, in this study, a genetic algorithm-based k-nearest neighbor estimator (also called intuitive k-NN estimator, IKE) is adapted to explore the optimum parameters and this algorithm estimates the excitation current of a synchronous motor with having small prediction errors. The motor parameters such as load current, power factor, error and excitation current changes are weighted depending on the effects on the excitation current. The experimental results are compared with the estimation results in consideration with standard deviations of the well-known Artificial Neural Network-based (ANN) method and k-NN-based estimator with that of the proposed IKE method. The results have shown that the proposed IKE estimator achieves the tasks in high accuracies, stabilities, robustness and low error rates other two well-known methods presented in the literature.

Intelligent Hotspot Prediction for Network-on-Chip-Based Multicore Systems

Hotspots are network-on-chip (NoC) routers or modules in multicore systems which occasionally receive packetized data from other networked element producers at a rate higher than they can consume it. This adverse phenomenon may greatly reduce the performance of NoCs, especially when wormhole flow-control is employed, as backpressure can cause the buffers of neighboring routers to quickly fill-up leading to a spatial spread in congestion. This can cause the network to saturate prematurely where in the worst scenario the NoC may be rendered unrecoverable. Thus, a hotspot prevention mechanism can be greatly beneficial, as it can potentially enable the interconnection system to adjust its behavior and prevent the rise of potential hotspots, subsequently sustaining NoC performance. The inherent unevenness of traffic patterns in an NoC-based general-purpose multicore system such as a chip multiprocessor, due to the diverse and unpredictable access patterns of applications, produces unexpected hotspots whose appearance cannot be known a priori, as application demands are not predetermined, making hotspot prediction and subsequently prevention difficult. In this paper, we present an artificial neural network-based (ANN) hotspot prediction mechanism that can be potentially used in tandem with a hotspot avoidance or congestion-control mechanism to handle unforeseen hotspot formations efficiently. The ANN uses online statistical data to dynamically monitor the interconnect fabric, and reactively predicts the location of an about to-be-formed hotspot(s), allowing enough time for the multicore system to react to these potential hotspots. Evaluation results indicate that a relatively lightweight ANN-based predictor can forecast hotspot formation(s) with an accuracy ranging from 65% to 92%.

Application of an artificial neural network for predicting the sizing curve of stand-alone photovoltaic system: a case study

In this paper, an artificial neural network-based (ANN) prediction of the sizing curve of stand-alone PV systems (SAPV) was developed. Numerical method has been used for constructing the sizing curves for different loss of load probability (LLP) corresponding to 40-sites located in Algeria. The feed-forward neural network (FFNN) is trained considering the geographical coordinate and LLP as inputs, the desired data are the sizing curve represented by CA = f(CS) (i.e., 30 points where taken from each sizing curves). Subsequently, the FFNN is tested with results obtained by the numeric method. The results show that the FFNN model can be applied effectively and afford high accuracy. The coefficient of determination (r) is arranged between 85% and 98% and the mean relative error do not exceed 2%.

Modeling of surface roughness in precision machining of metal matrix composites using ANN

Characteristics of machined surfaces are known to influence the product performance significantly since they are directly linked to the ability of the material to withstand stresses, temperature, friction and corrosion. This paper presents an experimental work on the analysis of machined surface quality on Al/SiCp composites leading to an artificial neural network-based (ANN) model to predict the surface roughness. The predicted roughness of machined surfaces based on the ANN model was found to be in very good agreement with the unexposed experimental data set.

Neural network kinetic prediction of coke burn-off on spent MnO2/CeO2 wet oxidation catalysts

Combustion kinetics of coke laydown on wet oxidation catalysts is studied by means of temperature-programmed oxidation “TPO” and mass spectrometry “MS” in the temperature range from 30 to 600°C. The study is designed to allow a better understanding of the influence of wet oxidation conditions (catalyst-phenol contacting time and temperature) on the combustion kinetics of coke deposited on MnO2/CeO2 and 1wt.% Pt-MnO2/CeO2 catalysts during phenol degradation. In this respect, the experimental procedure involves the continuous monitoring of carbon oxides and O2 fluxes resulting form the combustion of carbonaceous deposits in a 5% O2/He mixture. Based on the experimental data, an artificial neural network-based (ANN) modeling approach is implemented to represent, as accurately as possible, the complex combustion phenomenon so as to provide the opportunity to predict its evolution.In this context, the resulting ANN model is used as a “black-box” to approximate the complex non-linear conversion rate of the wet oxidation coke. The conversion rate is thus, expressed in terms of the TPO ramp temperature, running oxygen concentration, wet oxidation temperature, and phenol oxidation time.The proposed ANN-based modeling approach proved to be an accurate, reliable and effective tool for the quantification of the coke burn-off kinetics. The method has then great potential as a means to compensate for the lack of efficient phenomenological kinetic modeling techniques. It can also be used in the design of regenerative units downstream of catalytic wastewater treatment reactors based on wet oxidation.

A parallel implementation of the backward error propagation neural network training algorithm: Experiments in event identification

An artificial neural-network-based (ANN) event detection and alarm generation system has been developed to aid clinicians in the identification of critical events commonly occurring in the anesthesia breathing circuit. To detect breathing circuit problems, the system monitored CO 2 gas concentration, gas flow, and airway pressure. Various parameters were extracted from each of these input waveforms and fed into an artificial neural network. To develop truly robust ANNs, investigators are required to train their networks on large training data sets, requiring enormous computing power. We implemented a parallel version of the backward error propagation neural network training algorithm in the widely portable parallel programming language C-Linda. A maximum speedup of 4.06 was obtained with six processors. This speedup represents a reduction in total run-time from 6.4 to 1.5 h. By reducing the total run time of the computation through parallelism, we were able to optimize many of the neural network's initial parameters. We conclude that use of the master-worker model of parallel computation is an excellent method for speeding up the backward error propagation neural network training algorithm.