Optimal Architecture Research Articles

Architectural design and optimization of internal structures in 3D printed electrodes for superior supercapacitor performance

The architecture of electrodes plays a pivotal role in the transfer and transportation of charges during electrochemical reactions. Selecting optimal electrode materials and devising well-conceived electrode structures can substantially enhance the electrochemical performance of devices. This manuscript leverages 3D printing technology to fabricate asymmetric supercapacitor devices featuring regular layered configurations. By investigating the impact of various materials on the internal architecture of printed electrodes, we establish a stratified electrode structure with an orderly arrangement, thereby significantly improving asymmetric charge transfer between electrodes. The application of 3D printing technology to construct electrode structures effectively mitigates the agglomeration of electrode materials. The 3D-printed VCG//MXene devices demonstrate exceptional areal capacitance (205.57 mF cm−2) and energy density (60.03 μWh cm−2), with a power density of 0.174 W cm−2. Consequently, selecting appropriate materials for fabricating printable electrode structures and achieving efficient 3D printing is anticipated to offer novel insights into the construction and enhancement of miniature asymmetric micro-supercapacitor (MSCs) devices.

Architecture Optimization for Hybrid Deep Residual Networks in Liver Tumor Segmentation Using a GA

Nowadays, liver tumor segmentation has been widely applied for vital medical objectives such as illness diagnosis, treatment, and evaluation of liver function. In this work, we aim to improve the performance of hybrid convolution neural network (CNN) models for liver tumor segmentation. A new automatic design method is introduced to build a hybrid CNN model with an optimal architecture using a combination of the original U_Net and the Res-UNet. In the proposed design method, the Res-UNet model is divided into different blocks, and then many different candidate architectures for the hybrid CNN model can be created using gradually these blocks to replace their corresponding blocks of the original U-Net. The replacement is executed if the candidate architecture using a residual block gives better performance than the corresponding one that uses the original U-Net block. We use a genetic algorithm (GA) to get the optimal architecture among the candidate architectures that gives the best performance for the hybrid CNN model. During the optimization process, in addition to the optimal architecture, other crucial configurations such as the learning algorithm, learning rate, and batch size are also determined. We refer to the hybrid CNN model that has the best architecture as GA_Res_UNet. Comparison has been made with other CNN models that are commonly used in image segmentation of liver tumors. We use the 3D-IRCADb01 liver tumor dataset that contains 2800 magnetic resonance (MR) images of 20 patients. Each image has a liver tumor with its corresponding mask. All MR images are resized to 256 × 256 in DICOM format. The findings indicate that GA_Res_UNet outperformed the other compared models over the considered dataset. It achieves 98.5% of dice coefficient and 99.8% of accuracy. Moreover, two other liver tumor segmentation datasets (LiTS17 and CHAOS) are utilized in another comparison to show the superior performance of our proposed model. The reported results show that the proposed model outperformed the other compared state-of-the-art models.

Lung Cancer Classification Using Combination Of Efficientnet And Visual Geometry Group Algorithm

Lung cancer is one of the leading causes of mortality All around the world. It is classified into three main types: Adenocarcinoma of the lung (ACA), Non-small cell lung cancer (N), and Squamous Cell Carcinoma of the lung (SCC). Lung Cancer Classification is crucial on development of effective treatments. This study aims to improve the accuracy of lung cancer classification through the integration of a hybrid model, which combines two Convolutional Neural Networks architectures, namely EfficientNet-B7 and VGG-16. A set of histopathology images was subjected to testing, with the data split into three categories: 60% for training, 30% for validation, and 10% for testing. Prior to use, each image underwent a preprocessing process, wherein it was resized to 256x256 pixels. The model test results achieved an accuracy, precision, recall, and F1-score of 98.73%, which is superior to the EfficientNet-B7 base model. The findings of this study demonstrate the potential of hybrid models to improve accuracy in lung cancer classification. The utilization of hybrid models has the potential to contribute significantly to the beginning diagnosis and appropriate Lung Cancer Therapies. Future research will focus on improving the model through the application of image segmentation techniques and expanding the scope of classification to other types of lung cancer. Optimization of the hybrid model architecture using novel techniques such as the attention mechanism or transfer learning will be conducted to improve the efficiency and accuracy of the model. Additionally, a system that can be integrated into clinical practice will be developed

A methodology for direct parameter identification for experimental results using machine learning — Real world application to the highly non-linear deformation behavior of FRP

In this work, we demonstrate how Machine learning (ML) techniques can be employed to externalize the knowledge and time intensive process of material parameter identification. This is done on the example of a recent data driven material model for the non-linear shear behavior of glass fiber reinforced polypropylene (GF/PP) (Gerritzen, 2022). A convolutional neural network (CNN) based model architecture is trained to predict material modeling parameters based on the input of stress–strain-curves. The optimal model architecture and training setup are determined by hyperparameter optimization (HPO). Solely virtual data, generated using the target material model, is used throughout the training and HPO. The final CNN is capable of calculating model parameter combinations from experimental stress–strain-curves which lead to excellent agreement between experimental and associated model curve.

Calculation of the main operational characteristics of a tethered high-altitude ship-based system

Objectives. Currently, UAVs are actively used in many military and civilian fields such as object surveillance, telecommunications, radar, photography, video recording, and mapping, etc. The main disadvantage of autonomous UAVs is their limited operating time. The long-term operation of UAVs on ships can be ensured by tethered high-altitude systems in which the power supply of engines and equipment is provided from the onboard energy source through a thin cable tether. This paper aims to select and justify the appearance of such system, as well as to calculate the required performance characteristics.Methods. The study used methods of systemic and functional analysis of tethered system parameters, as well as methods and models of the theory of relations and measurement.Results. The issues of design and implementation of new generation tethered high-altitude ship-based systems were considered. A rational type of aerodynamic design for unmanned aerial vehicles was determined based on existing tethered platforms. The optimal architecture of the tethered system was defined and justified. The paper presents the appearance and solution for placement onboard the ship, and describes its operation. The main initial parameters for designing high-altitude systems such as take-off weight, optimal lift altitude, maximum power required for operation, structure of the energy transfer system, as well as deployment and lift time to the design altitude were selected and calculated.Conclusions. The methodology for calculating the necessary characteristics described in the paper can be used for developing and evaluating tethered high-altitude systems. These systems are capable of performing a wide range of tasks, without requiring a separate storage and launch location, which is especially important in the ship environment. The system presented herein possesses significant advantages over well-known analogues.

Direct operational data-driven workflow for dynamic voltage prediction of commercial alkaline water electrolyzers based on artificial neural network (ANN)

The thermal inertia of alkaline water electrolysis (AWE) system under dynamic operation reduces efficiency and poses security risks. To mitigate this, predictive model control (MPC) emerges as a potential solution, aiming to precisely regulate temperature through temperature trend prediction, which relies on accurate voltage predictions. However, the current MPC implementation for AWE system utilizes empirical model for voltage prediction, requiring extensive experimental data for calibration, making it time-consuming and costly for large-scale water electrolysis projects. Data-driven models are a promising alternative for voltage prediction, but research is limited by the lack of long-term commercial AWE operational data and the ambiguity in developing effective data-driven models. In this study, a 10-kW level AWE test system underwent 300 h of simulated solar-driven dynamic operation to assess the resilience of AWE system to dynamic conditions. Subsequently, a robust data-driven workflow using artificial neural networks (ANNs) was developed for precise dynamic voltage prediction based on obtained operational data. The data-driven workflow incorporates automated processes for voltage prediction, including data importing, data cleaning, ANN architecture optimization, model training, and ultimately voltage prediction, avoiding the need for manual parameters fitting. The model’s performance was evaluated and compared with empirical model across four dynamic operation types (cold-start, ramp, stepwise, and random response), achieving an impressive accuracy over 98 %. Furthermore, the ANN model also exhibits application potential in U-I curves prediction (accuracy > 98 %), influential features assessment, operational optimization and MPCs. This ANN-based workflow serves as a versatile framework for precise voltage prediction under dynamic operations and provides an analytical platform for operational data analysis, system optimization, and advanced control system development.

Modeling Fluid Flow in Ship Systems for Controller Tuning Using an Artificial Neural Network

Flow processes onboard ships are common in order to transport fluids like oil, gas, and water. These processes are controlled by PID controllers, acting on the regulation valves as actuators. In case of a malfunction or refitting, a PID controller needs to be re-adjusted for the optimal control of the process. To avoid experimenting on operational real systems, models are convenient alternatives. When real-time information is needed, digital twin (DT) concepts become highly valuable. The aim of this paper is to analyze and determine the optimal NARX model architecture in order to achieve a higher-accuracy model of a ship’s flow process. An artificial neural network (ANN) was used to model the process in MATLAB. The experiments were performed using a multi-start approach to prevent overtraining. To prove the thesis, statistical analysis of the experimental results was performed. Models were evaluated for generalization using mean squared error (MSE), best fit, and goodness of fit (GoF) measures on two independent datasets. The results indicate the correlation between the number of input delays and the performance of the model. A permuted k-fold cross-validation analysis was used to determine the optimal number of voltage and flow delays, thus defining the number of model inputs. Permutations of training, test, and validation datasets were applied to examine bias due to the data arrangement during training.

Modeling Consumer Overall Acceptance for Traditional Spice-Based Ready-to-Drink Using Artificial Neural Network and Kansei Engineering

Measuring consumer acceptance of food products is challenging, primarily because the process is susceptible to bias. Several studies have reported that this challenge can be addressed through Kansei engineering through verbal and nonverbal response measurements. Therefore, this study aimed to predict consumer overall acceptance using artificial neural networks (ANN) and Kansei engineering. A total of 30 respondents participated in this study to test nine different samples of traditional spice-based ready-to-drink (RTD). The overall acceptance score and Kansei responses, including verbal and nonverbal, were then measured. Each sample was served cold in a 60-mL cup labeled with a three-digit random code, and the panelists were successfully presented with the nine drinks. All participants were asked to rank each Kansei word scale based on the intensity of their feelings during the assessment. The heart rate (HR) and skin temperature (ST) were also measured as nonverbal responses in real time. The results showed that Kansei's responses and respondent background best predicted overall acceptance. The optimal model architecture had ten input neurons, two hidden neurons, and one output neuron (10-2-1). The training, validation, and testing data showed that the performance of ANN was satisfactory, with a low error rate (RMSE) and a high coefficient of correlation value (R2). Based on the findings, the developed model could inspire and motivate further studies and development in industries to develop appropriate products for potential consumers, thereby revolutionizing the food industry.

Synthesis of activity-based costing and deep learning to support cost management: A case study in the automotive industry

Activity based costing (ABC) models for cost management and big data analytics are deeply rooted in academia and practice to efficiently manage operational costs. Inspired by cost management and deep learning (DL) theories we develop a multilayer perceptron (MLP) using a nested particle swarm optimized (PSO) neural network algorithm for self-controlled architecture and weight optimization. This innovative approach enables to mimic ABC while increasing prediction accuracy. In a real-world case study we use wheel manufacturing data provided by a large original equipment manufacturer (OEM) and follow a full factorial experimental design. Furthermore, we benchmark our novel DL approach with results derived from a traditional ABC analysis. We demonstrate that this intelligent cost estimation model can mimic ABC using few cost drivers contributing to efficient and transparent interorganizational cost management. Major quantitative findings within a conducted field experiment demonstrate a high forecast accuracy with low absolute cost percentage error (CPE) deviation of the novel cost estimation approach in industry. The results extend the scope of PSO for the simultaneous optimization of the architecture and weights of neural networks. Additionally, we prove that the PSO algorithm is a suitable alternative to traditional optimization methods in DL.

Nanoarchitectonics in Advanced Membranes for Enhanced Osmotic Energy Harvesting.

Osmotic energy, often referred to as "blue energy", is the energy generated from the mixing of solutions with different salt concentrations, offering a vast, renewable, and environmentally friendly energy resource. The efficacy of osmotic power production considerably relies on the performance of the transmembrane process, which depends on ionic conductivity and the capability to differentiate between positive and negative ions. Recent advancements have led to the development of membrane materials featuring precisely tailored ion transport nanochannels, enabling high-efficiency osmotic energy harvesting. In this review, ion diffusion in confined nanochannels and the rational design and optimization of membrane architecture are explored. Furthermore, structural optimization of the membrane to mitigate transport resistance and the concentration polarization effect for enhancing osmotic energy harvesting is highlighted. Finally, an outlook on the challenges that lie ahead is provided, and the potential applications of osmotic energy conversion are outlined. This review offers a comprehensive viewpoint on the evolving prospects of osmotic energy conversion.

Advances in neural architecture search.

Automated machine learning (AutoML) has achieved remarkable success in automating the non-trivial process of designing machine learning models. Among the focal areas of AutoML, neural architecture search (NAS) stands out, aiming to systematically explore the complex architecture space to discover the optimal neural architecture configurations without intensive manual interventions. NAS has demonstrated its capability of dramatic performance improvement across a large number of real-world tasks. The core components in NAS methodologies normally include (i) defining the appropriate search space, (ii) designing the right search strategy and (iii) developing the effective evaluation mechanism. Although early NAS endeavors are characterized via groundbreaking architecture designs, the imposed exorbitant computational demands prompt a shift towards more efficient paradigms such as weight sharing and evaluation estimation, etc. Concurrently, the introduction of specialized benchmarks has paved the way for standardized comparisons of NAS techniques. Notably, the adaptability of NAS is evidenced by its capability of extending to diverse datasets, including graphs, tabular data and videos, etc., each of which requires a tailored configuration. This paper delves into the multifaceted aspects of NAS, elaborating on its recent advances, applications, tools, benchmarks and prospective research directions.

Integrating Experimental and Numerical Data for Improved Steam Reforming Simulation with Deep Learning

Abstract In this paper, a data-driven methane steam reforming simulation is developed and used to predict the post-reaction mixture composition. Until today, methane steam reforming remains a predominant hydrogen production method, yet modeling its complex reactions remains a significant challenge due to the intricate interplay of process variables. Here, we show an artificial neural network simulator that can effectively model these reactions, offering precise predictions based on parameters like temperature, inlet gas composition, methane flow, and nickel catalyst mass. Our approach to data curation integrates experimental, interpolated, and theoretically calculated values and refining the model by assessing the relative importance of each data category. Various neural network structures were tested before ultimately identifying an optimal architecture with a 5-6-8-6-4 network structure. The network underwent 6000 epochs of training, leading to a model that demonstrates excellent agreement with experimental observations, as evidenced by the mean squared error of 0.000217 and the Pearson correlation coefficient of 0.965. Moreover, all process trajectories predicted by the network are characterized by a smooth course and are within a physical range of values. Therefore, this work overcomes a common challenge in chemical process simulation using neural networks and also sets a possible direction for future research in this field.

Ultra-Wideband Phased Antenna Array Time Delay Unit Architecture Optimization in Presence of Component Non-Idealities

Ultra-Wideband Phased Antenna Array Time Delay Unit Architecture Optimization in Presence of Component Non-Idealities

Redefining Neural Architecture Search of Heterogeneous Multinetwork Models by Characterizing Variation Operators and Model Components.

With neural architecture search (NAS) methods gaining ground on manually designed deep neural networks-even more rapidly as model sophistication escalates-the research trend is shifting toward arranging different and often increasingly complex NAS spaces. In this conjuncture, delineating algorithms which can efficiently explore these search spaces can result in a significant improvement over currently used methods, which, in general, randomly select the structural variation operator, hoping for a performance gain. In this article, we investigate the effect of different variation operators in a complex domain, that of multinetwork heterogeneous neural models. These models have an extensive and complex search space of structures as they require multiple subnetworks within the general model in order to answer different output types. From that investigation, we extract a set of general guidelines whose application is not limited to that particular type of model and are useful to determine the direction in which an architecture optimization method could find the largest improvement. To deduce the set of guidelines, we characterize both the variation operators, according to their effect on the complexity and performance of the model; and the models, relying on diverse metrics which estimate the quality of the different parts composing it.

Flexible and High-Efficiency LDPC Encoder Architecture for CCSDS Standard

Abstract The Consultative Committee for Space Data Systems (CCSDS) has adopted quasi-cyclic low-density parity-check (QC-LDPC) codes for use in near-Earth (C2) and deep space (AR4JA) communications. Existing encoder architectures for C2 codes, however, fall short in efficiency for high-throughput applications. This paper introduces a comprehensive approach combining algorithmic and architectural optimizations to enhance hardware usage efficiency (HUE) while offering flexibility. We propose an integrated inter-block and intra-block parallel (IIB-IBP) encoding algorithm that leverages the unique matrix structure to significantly enhance performance. Additionally, a matrix-specific command register pretreatment (MSCRP) technique is developed to effectively handle the special dimensions of the generator matrix. Furthermore, we detail an offline design process for the automated generation of the encoder core’s HDL description, facilitating fine-tuning of encoding parallelism, latency, FPGA resource utilization, and overall throughput. Hardware implementation on a Virtex XC5VLX110T FPGA demonstrates that our encoder reaches an impressive throughput of 10.6 Gb/s with only 2531 LUTs and 1040 FFs, achieving a HUE of 2.97 Mbps/logic unit. This performance marks a 70.6% increase in HUE when compared to state-of-the-art designs.

Advanced computer architecture optimization for machine learning/deep learning

Abstract The recent progress in Machine Learning (Géron, 2022) and particularly Deep Learning (Goodfellow, 2016) models exposed the limitations of traditional computer architectures. Modern algorithms demonstrate highly increased computational demands and data requirements that most existing architectures cannot handle efficiently. These demands result in training speed, inference latency, and power consumption bottlenecks, which is why advanced methods of computer architecture optimization are required to enable the development of ML/DL-dedicated efficient hardware platforms (Engineers, 2019). The optimization of computer architecture for applications of ML/DL becomes critical, due to the tremendous demand for efficient execution of complex computations by Neural Networks (Goodfellow, 2016). This paper reviewed the numerous approaches and methods utilized to optimize computer architecture for ML/DL workloads. The following sections contain substantial discussion concerning the hardware-level optimizations, enhancements of traditional software frameworks and their unique versions, and innovative explorations of architectures. In particular, we discussed hardware including specialized accelerators, which can improve the performance and efficiency of a computation system using various techniques, specifically describing accelerators like CPUs (multicore) (Hennessy, 2017), GPUs (Hwu, 2015) and TPUs (Contributors, 2017), parallelism in multicore architectures, data movement in hardware systems, especially techniques such as caching and sparsity, compression, and quantization, other special techniques and configurations, such as using specialized data formats, and measurement sparsity. Moreover, this paper provided a comprehensive analysis of current trends in software frameworks, Data Movement optimization strategies (A.Bienz, 2021), sparsity, quantization and compression methods, using ML for architecture exploration, and, DVFS (Hennessy, 2017),, which provides strategies for maximizing hardware utilization and power consumption during training, machine learning, dynamic voltage, and frequency scaling, runtime systems. Finally, the paper discussed research opportunity directions and the possibilities of computer architecture optimization influence in various industrial and academic areas of ML/DL technologies. The objective of implementing these optimization techniques is to largely minimize the current gap between the computational needs of ML/DL algorithms and the current hardware’s capability. This will lead to significant improvements in training times, enable real-time inference for various applications, and ultimately unlock the full potential of cutting-edge machine learning algorithms.

Self-adaptive rotational electromagnetic energy generation as an alternative to triboelectric and piezoelectric transductions

Triboelectric and piezoelectric energy harvesters can hardly power most microelectronic systems. Rotational electromagnetic harvesters are very promising alternatives, but their performance is highly dependent on the varying mechanical sources. This study presents an innovative approach to significantly increase the performance of rotational harvesters, based on dynamic coil switching strategies for optimization of the coil connection architecture during energy generation. Both analytical and experimental validations of the concept of self-adaptive rotational harvester were carried out. The adaptive harvester was able to provide an average power increase of 63.3% and 79.5% when compared to a non-adaptive 16-coil harvester for harmonic translation and harmonic swaying excitations, respectively, and 83.5% and 87.2% when compared to a non-adaptive 8-coil harvester. The estimated energy conversion efficiency was also enhanced from ~80% to 90%. This study unravels an emerging technological approach to power a wide range of applications that cannot be powered by other vibrationally driven harvesters.

Optimization of Neural Network Architectures for Image Recognition

Optimization of Neural Network Architectures for Image Recognition

Improving the crashworthiness of bi-tubular architectures with ABS cores under axial loading: Experimental and numerical investigation

In this study, quasi-static axial compression tests were conducted on mild steel bi-tubular architectures with rectangular nested tube (RNT) and square nested tube (SNT) geometries to evaluate their crushing and crashworthiness performance. A multi-criteria decision-making approach was employed to identify the optimal energy-absorbing architecture. The SNT structure, with the smallest gap size between the inner and outer tubes, exhibited the most desirable energy absorption characteristics among the considered cases. Acrylonitrile butadiene styrene (ABS) cores, with either rhombic or square cell configurations, were used to enhance the energy absorption performance of the SNT structure. A finite element model was created to evaluate the responses of the SNT structure filled with ABS cores. The validity of finite element simulations of the ABS cores and optimal architecture under axial compression were confirmed by comparing them with experimental results. The integration of the cores into the nested architecture enhanced crashworthiness performance and contributed to the control of the structure deformation. The SNT structure filled with rhombic ABS core exhibited superior crashworthiness performance compared to the counterpart filled with square core. The energy absorption of nested SNT structures filled with rhombic ABS core can be 116.93% greater than the corresponding non-filled structure. The crashworthiness indices of ABS-filled structures were highly sensitive to the number of cells and wall thickness of the core. A nested architecture with an ABS core could serve as a novel architecture for energy-absorbing devices.

A Deep Reinforcement Learning Approach to DC-DC Power Electronic Converter Control with Practical Considerations

In recent years, there has been a growing interest in using model-free deep reinforcement learning (DRL)-based controllers as an alternative approach to improve the dynamic behavior, efficiency, and other aspects of DC–DC power electronic converters, which are traditionally controlled based on small signal models. These conventional controllers often fail to self-adapt to various uncertainties and disturbances. This paper presents a design methodology using proximal policy optimization (PPO), a widely recognized and efficient DRL algorithm, to make near-optimal decisions for real buck converters operating in both continuous conduction mode (CCM) and discontinuous conduction mode (DCM) while handling resistive and inductive loads. Challenges associated with delays in real-time systems are identified. Key innovations include a chattering-reduction reward function, engineering of input features, and optimization of neural network architecture, which improve voltage regulation, ensure smoother operation, and optimize the computational cost of the neural network. The experimental and simulation results demonstrate the robustness and efficiency of the controller in real scenarios. The findings are believed to make significant contributions to the application of DRL controllers in real-time scenarios, providing guidelines and a starting point for designing controllers using the same method in this or other power electronic converter topologies.