Separate Neural Networks Research Articles

Physics-informed neural networks for heterogeneous poroelastic media

This study presents a novel physics-informed neural network (PINN) framework for modeling poroelasticity in heterogeneous media with material interfaces. The approach introduces a composite neural network (CoNN) where separate neural networks predict displacement and pressure variables for each material. While sharing identical activation functions, these networks are independently trained for all other parameters. To address challenges posed by heterogeneous material interfaces, the CoNN is integrated with the Interface-PINNs (I-PINNs) framework (Sarma et al., Comput. Methods Appl. Mech. Eng. 429: 117135, 2024), allowing different activation functions across material interfaces. This ensures accurate approximation of discontinuous solution fields and gradients. Performance and accuracy of this combined architecture were evaluated against the conventional PINNs approach, a single neural network (SNN) architecture, and the eXtended PINNs (XPINNs) framework through two one-dimensional benchmark examples with discontinuous material properties. The results show that the proposed CoNN with I-PINNs architecture achieves an RMSE that is two orders of magnitude better than the conventional PINNs approach and is at least 40 times faster than the SNN framework. Compared to XPINNs, the proposed method achieves an RMSE at least one order of magnitude better and is 40% faster.

Big Brother Is Watching You: Leveraging Artificial Intelligence for Automated Fuel Cell Monitoring (Technical Report)

Precise monitoring and anomaly detection play a critical role in the successful and continuous operation of fuel cells. Especially in durability testing, it is important to detect failures at an early stage and prevent external influences from affecting the test results or damaging the fuel cell stack. Because most external influencing factors are not directly sensed, they only exhibit in the currently unmonitored cell voltage. This emphasises the need for a sensitive cell voltage monitoring concept.In [1] we presented a machine learning based approach for cell voltage monitoring using neural networks as digital twins. It was successfully shown that this approach enables the detection of anomalies in cell voltage measurements. However, the combination of both steady-state and degradation effects in a single digital twin model hindered the transfer of the model to new fuel cells.Building on this approach, we present a second-generation of the fuel cell digital twin model. As shown in the Figure, the combined model of the previous approach is split into a steady-state fuel cell model and a corrective degradation model. The stationary fuel cell model predicts the stationary cell voltage based on the current operating conditions of the fuel cell. The degradation model accounts for fuel cell degradation and predicts a degradation correction based on historical fuel cell operating data. The combination of the predicted steady-state cell voltage and the degradation correction finally results in a prediction of the degraded cell voltage. This final prediction can be subsequently used for cell voltage monitoring. By decoupling stationary effects and degradation effects into two separate neural networks, each model can be adapted much more specifically to new data using transfer learning. This allows simulation data to be used to pre-train the steady-state model, significantly reducing the amount of real training data required to less than 100 data points. The degradation model on the other hand benefits from the reduced model complexity since only the few operating parameters that influence the fuel cell degradation remain as model inputs.To demonstrate the accuracy and adaptability of this approach, the new digital twin model is applied to real fuel cell data. In particular, the model’s ability to precisely detect previously undetected anomalies in the test data is shown.Finally, we present a framework for automated fuel cell monitoring. The framework provides a web-based graphical user interface to set up the monitoring configurations and to initiate the automated training of digital twins. Subsequently, the trained models are continuously applied to detect anomalies in new measured data. Upon detection of an anomaly, a notification will be sent to the test station operator.[1] L. Klass, A. Kabza, F. Sehnke, K. Strecker and M. Hölzle, “Lifelong performance monitoring of PEM fuel cells using machine learning models,” Journal of Power Sources, vol. 580, 2023. Figure 1

AI-Powered Approaches for Hypersurface Reconstruction in Multidimensional Spaces

The present article explores the possibilities of using artificial neural networks to solve problems related to reconstructing complex geometric surfaces in Euclidean and pseudo-Euclidean spaces, examining various approaches and techniques for training the networks. The main focus is on the possibility of training a set of neural networks with information about the available surface points, which can then be used to predict and complete missing parts. A method is proposed for using separate neural networks that reconstruct surfaces in different spatial directions, employing various types of architectures, such as multilayer perceptrons, recursive networks, and feedforward networks. Experimental results show that artificial neural networks can successfully approximate both smooth surfaces and those containing singular points. The article presents the results with the smallest error, showcasing networks of different types, along with a technique for reconstructing geographic relief. A comparison is made between the results achieved by neural networks and those obtained using traditional surface approximation methods such as Bézier curves, k-nearest neighbors, principal component analysis, Markov random fields, conditional random fields, and convolutional neural networks.

Enhancing HVDC transmission line fault detection using disjoint bagging and bayesian optimization with artificial neural networks and scientometric insights

DC grid fault protection techniques have previously faced challenges such as fixed thresholds, insensitivity to high-resistance faults, and dependency on specific threshold settings. These limitations can lead to elevated fault currents in the grid, particularly affecting multi-modular converters (MMCs) vulnerability to large fault current transients. This paper proposes a novel approach that combines the disjoint-based Bootstrap Aggregating (Bagging) technique and Bayesian optimization (BO) for fault detection in DC grids. Disjoint partitions reduce variance and enhance Ensemble Artificial Neural Network (EANN) performance, while BO optimizes EANN architecture. The proposed approach uses multiple transient periods instead of a fixed time to train the model. Transient periods are segmented into multiple 1 ms intervals, and each interval trains a separate neural network. In this way, a robust local relay is created that does not require high-speed communication systems. Additionally, a discrete wavelet transform (DWT) is applied to select detailed coefficients of the transient fault current, measured at the DC line’s sending terminal for fault protection. EANN is trained in comprehensive offline data that considers noise impact. Simulation results demonstrate the scheme’s ability to detect faults as high as 400 Ω accurately. This makes it a robust, reliable, and effective solution for fault detection on high-voltage direct current (HVDC) transmission lines. Lastly, this research provides the first-ever scientometric analysis of HVDC transmission line fault protection using neural network algorithms, highlighting major research themes and trends. The scientometric analysis was based on a dataset of 136 available research articles from the Scopus database from the last ten years. Therefore, this research provides valuable insights into the use of ANN for HVDC transmission line fault protection.

Nested object detection using mask R-CNN: application to bee and varroa detection

In this paper, we address an essential problem related to object detection and image processing: detecting objects potentially nested in other ones. This problem exists particularly in the beekeeping sector: detecting varroa parasites on bees. Indeed, beekeepers must ensure the level of infestation of their apiaries by the varroa parasite which settles on the backs of bees. As far as we know, there is no yet a published approach to deal with nested object detection using only one neural network trained on two different datasets. We propose an approach that fills this gap. Therefore, we improve the accuracy and the efficiency of bee and varroa detection task. Our work is based on deep learning, more precisely Mask R-CNN neural network. Instead of segmenting detected objects (bees), we segment internal objects (varroas). We add a branch to Faster R-CNN to segment internal objects. We extract relevant features for internal object segmentation and suggest efficient method for training the neural network on two different datasets. Our experiments are based on a set of images of bee frames, containing annotated bees and varroa mites. Due to differences in occurrence rates, two different sets were created. After carrying out experiments, we ended up with a single neural network capable of detecting two nested objects without decreasing accuracy compared to two separate neural networks. Our approach, compared to traditional separate neural networks, improves varroa detection accuracy by 1.9%, reduces infestation level prediction error by 0.22%, and reduces execution time by 28% and model memory by 23%. In our approach, we extract Res4 (a layer of the ResNet neural network) features for varroa segmentation, which improves detection accuracy by 11% compared to standard FPN extraction. Thus, we suggest a new approach that detects nested objects more accurately than two separate network approaches.

Stabilizing Electric Vehicle Systems Using Proximal Policy-Based Self-structuring Control

An active disturbance rejection control (ADRC) has been developed for stabilizing electric vehicle (EV) systems without the need for model identification. The proximal policy optimization (PPO) algorithm, along with actor and critic neural networks, has been used to fine-tune the adjustable parameters of the ADRC controller to achieve optimal performance in a specific case study. The architecture of PPO implements separate neural networks and ameliorates the PPO adaptability to handle continuous action spaces. By maximizing a reward function based on system output, the PPO agent optimally tunes the gains to reduce undesired speed fluctuations of EVs and improve system stability. Performance evaluation under the new European driving cycle and federal test procedure has been conducted to examine the feasibility of the suggested controller. The disturbance rejection capability of the ADRC controller designed by the PPO algorithm has been tested and compared with prevalent control methodologies. Moreover, real-time examinations of the dynamic behavior of EV systems have been made to identify the capability of the suggested controller in real-world hardware. The results show that the suggested controller outperforms other designed controllers in terms of transient behavior and numerical performance metrics.

Development and validation of a deep learning-based method for automatic measurement of uterus, fibroid, and ablated volume in MRI after MR-HIFU treatment of uterine fibroids

IntroductionThe non-perfused volume divided by total fibroid load (NPV/TFL) is a predictive outcome parameter for MRI-guided high-intensity focused ultrasound (MR-HIFU) treatments of uterine fibroids, which is related to long-term symptom relief. In current clinical practice, the MR-HIFU outcome parameters are typically determined by visual inspection, so an automated computer-aided method could facilitate objective outcome quantification. The objective of this study was to develop and evaluate a deep learning-based segmentation algorithm for volume measurements of the uterus, uterine fibroids, and NPVs in MRI in order to automatically quantify the NPV/TFL. Materials and MethodsA segmentation pipeline was developed and evaluated using expert manual segmentations of MRI scans of 115 uterine fibroid patients, screened for and/or undergoing MR-HIFU treatment. The pipeline contained three separate neural networks, one per target structure. The first step in the pipeline was uterus segmentation from contrast-enhanced (CE)-T1w scans. This segmentation was subsequently used to remove non-uterus background tissue for NPV and fibroid segmentation. In the following step, NPVs were segmented from uterus-only CE-T1w scans. Finally, fibroids were segmented from uterus-only T2w scans. The segmentations were used to calculate the volume for each structure. Reliability and agreement between manual and automatic segmentations, volumes, and NPV/TFLs were assessed. ResultsFor treatment scans, the Dice similarity coefficients (DSC) between the manually and automatically obtained segmentations were 0.90 (uterus), 0.84 (NPV) and 0.74 (fibroid). Intraclass correlation coefficients (ICC) were 1.00 [0.99, 1.00] (uterus), 0.99 [0.98, 1.00] (NPV) and 0.98 [0.95, 0.99] (fibroid) between manually and automatically derived volumes. For manually and automatically derived NPV/TFLs, the mean difference was 5% [-41%, 51%] (ICC: 0.66 [0.32, 0.85]). ConclusionThe algorithm presented in this study automatically calculates uterus volume, fibroid load, and NPVs, which could lead to more objective outcome quantification after MR-HIFU treatments of uterine fibroids in comparison to visual inspection. When robustness has been ascertained in a future study, this tool may eventually be employed in clinical practice to automatically measure the NPV/TFL after MR-HIFU procedures of uterine fibroids.

Toward machine-learning-assisted PW-class high-repetition-rate experiments with solid targets

We present progress in utilizing a machine learning (ML) assisted optimization framework to study the trends in a parameter space defined by spectrally shaped, high-intensity, petawatt-class (8 J, 45 fs) laser pulses interacting with solid targets and give the first simulation-based overview of predicted trends. A neural network (NN) incorporating uncertainty quantification is trained to predict the number of hot electrons generated by the laser–target interaction as a function of pulse shaping parameters. The predictions of this NN serve as the basis function for a Bayesian optimization framework to navigate this space. For post-experimental evaluation, we compare two separate neural network (NN) models. One is based solely on data from experiments, and the other is trained only on ensemble particle-in-cell simulations. Reviewing the predicted and observed trends across the experiment-capable laser parameter search space, we find that both ML models predict a maximal increase in hot electron generation at a level of approximately 12%–18%; however, no statistically significant enhancement was observed in experiments. On direct comparison of the NN models, the average discrepancy is 8.5%, with a maximum of 30%. Since shot-to-shot fluctuations in experiments affect the observations, we evaluate the behavior of our optimization framework by performing virtual experiments that vary the number of repeated observations and the noise levels. Here, we discuss the implications of such a framework for future autonomous exploration platforms in high-repetition-rate experiments.

SANe: Space adaptation network for temporal knowledge graph completion

Temporal Knowledge Graphs (TKGs) model time-dependent facts as relations between entities at specific timestamps, making them well-suited for real-world scenarios. However, TKGs are susceptible to incompleteness, necessitating Temporal Knowledge Graph Completion (TKGC) to predict missing facts. Prior methods often struggle to effectively handle two critical properties of TKGs, time-variability and time-stability, simultaneously, which hinders their performance. In this paper, we propose Space Adaptation Network (SANe), a novel approach for TKGC. SANe adapts facts at different timestamps to distinct latent spaces, effectively addressing time-variability. Our model introduces Parameter Generation Network to produce separate neural networks for each snapshot, which are then encoded into different latent spaces. A dynamic convolutional neural network processes entities and relations, utilizing different learned parameters generated by parameter generation network with respect to timestamps. By handling different temporal snapshots separately, TKGC is transformed into static KGC, enabling the modeling of time-variability. Dynamic convolutional neural network efficiently learns collective knowledge over large periods and supplements more specific knowledge gradually in smaller periods, facilitating time-stability. To strike a balance between learning time-variability and time-stability, we introduce a time-aware parameter generator to produce parameters hierarchically based on year, month, and day timestamps. Long-term knowledge is effectively shared across adjacent snapshots within the same year or month, while short-term knowledge within a day is preserved in specific parameters. However, in unbalanced TKGs, where many facts occur in small intervals, the large number of parameters generated by time-aware parameter generator may remain underutilized. To address this, we propose Adaptive Parameter Generation with a partition tree, ensuring parameter load balancing while maintaining time-stability. We conduct extensive experiments on five benchmark datasets, demonstrating the superiority of SANe over existing methods for TKGC, achieving state-of-the-art performance. Our contributions include pioneering TKGC from the perspective of space adaptation, achieving a balance between time-variability and time-stability through latent space overlap constraints, and substantiating the effectiveness of our model through comprehensive experiments on rich temporal datasets.

A deep learning approach to censored regression

In censored regression, the outcomes are a mixture of known values (uncensored) and open intervals (censored), meaning that the outcome is either known with precision or is an unknown value above or below a known threshold. The use of censored data is widespread, and correctly modeling it is essential for many applications. Although the literature on censored regression is vast, deep learning approaches have been less frequently applied. This paper proposes three loss functions for training neural networks on censored data using gradient backpropagation: the tobit likelihood, the censored mean squared error, and the censored mean absolute error. We experimented with three variations in the tobit likelihood that arose from different ways of modeling the standard deviation variable: as a fixed value, a reparametrization, and an estimation using a separate neural network for heteroscedastic data. The tobit model yielded better results, but the other two losses are simpler to implement. Another central idea of our research was that data are often censored and truncated simultaneously. The proposed losses can handle simultaneous censoring and truncation at arbitrary values from above and below.

A Time-Series-Based Sample Amplification Model for Data Stream with Sparse Samples

The data stream is a dynamic collection of data that changes over time, and predicting the data class can be challenging due to sparse samples, complex interdependent characteristics between data, and random fluctuations. Accurately predicting the data stream in sparse data can create complex challenges. Due to its incremental learning nature, the neural networks suitable approach for streaming visualization. However, the high computational cost limits their applicability to high-speed streams, which has not yet been fully explored in the existing approaches. To solve these problems, this paper proposes an end-to-end dynamic separation neural network (DSN) approach based on the characteristics of data stream fluctuations, which expands the static sample at a given moment into a sequence of sample streams in the time dimension, thereby increasing the sparse samples. The Temporal Augmentation Module (TAM) can overcome these challenges by modifying the sparse data stream and reducing time complexity. Moreover, a neural network that uses a Variance Detection Module (VDM) can effectively detect the variance of the input data stream through the network and dynamically adjust the degree of differentiation between samples to enhance the accuracy of forecasts. The proposed method adds significant information regarding the data sparse samples and enhances low dimensional samples to high data samples to overcome the sparse data stream problem. In VDM the preprocessed data achieve data augmentation and the samples are transmitted to VDM. The proposed method is evaluated using different types of data streaming datasets to predict the sparse data stream. Experimental results demonstrate that the proposed method achieves a high prediction accuracy and that the data stream has significant effects and strong robustness compared to other existing approaches.

An Open-Circuit Fault Diagnosis System Based on Neural Networks in the Inverter of Three-Phase Permanent Magnet Synchronous Motor (PMSM)

Three-phase motors find extensive applications in various industries. Open-circuit faults are a common occurrence in inverters, and the open-circuit fault diagnosis system plays a crucial role in identifying and addressing these faults to enhance the safety of motor operations. Nevertheless, the current open-circuit fault diagnosis system faces challenges in precisely detecting specific faulty switches. The proposed work presents a neural network-based open-circuit fault diagnosis system for identifying faulty power switches in inverter-driven motor systems. The system leverages trained phase-to-phase voltage data from the motor to recognize the type and location of faults in each phase with high accuracy. Employing separate neural networks for each of the three phases in a three-phase permanent magnet synchronous motor, the system achieves an outstanding overall fault detection accuracy of approximately 99.8%, with CNN and CNN-LSTM architectures demonstrating superior performance. This work makes two key contributions: (1) implementing neural networks to significantly improve the accuracy of locating faulty switches in open-circuit fault scenarios, and (2) identifying the optimal neural network architecture for effective fault diagnosis within the proposed system.

Four-Wheeled Vehicle Sideslip Angle Estimation: A Machine Learning-Based Technique for Real-Time Virtual Sensor Development

In the last few decades, the role of vehicle dynamics control systems has become crucial. In this complex scenario, the correct real-time estimation of the vehicle’s sideslip angle is decisive. Indeed, this quantity is deeply linked to several aspects, such as traction and stability optimization, and its correct understanding leads to the possibility of reaching greater road safety, increased efficiency, and a better driving experience for both autonomous and human-controlled vehicles. This paper aims to estimate accurately the sideslip angle of the vehicle using different neural network configurations. Then, the proposed approach involves using two separate neural networks in a dual-network architecture. The first network is dedicated to estimating the longitudinal velocity, while the second network predicts the sideslip angle and takes the longitudinal velocity estimate from the first network as input. This enables the creation of a virtual sensor to replace the real one. To obtain a reliable training dataset, several test sessions were conducted on different tracks with various layouts and characteristics, using the same reference instrumented vehicle. Starting from the acquired channels, such as lateral and longitudinal acceleration, steering angle, yaw rate, and angular wheel speeds, it has been possible to estimate the sideslip angle through different neural network architectures and training strategies. The goodness of the approach was assessed by comparing the estimations with the measurements obtained from an optical sensor able to provide accurate values of the target variable. The obtained results show a robust alignment with the reference values in a great number of tested conditions. This confirms that the adoption of artificial neural networks represents a reliable strategy to develop real-time virtual sensors for onboard solutions, expanding the information available for controls.

SparseVoxNet: 3-D Object Recognition With Sparsely Aggregation of 3-D Dense Blocks.

Automatic recognition of 3-D objects in a 3-D model by convolutional neural network (CNN) methods has been successfully applied to various tasks, e.g., robotics and augmented reality. Three-dimensional object recognition is mainly performed by analyzing the object using multi-view images, depth images, graphs, or volumetric data. In some cases, using volumetric data provides the most promising results. However, existing recognition techniques on volumetric data have many drawbacks, such as losing object details on converting points to voxels and the large size of the input volume data that leads to substantial 3-D CNNs. Using point clouds could also provide very promising results; however, point-cloud-based methods typically need sparse data entry and time-consuming training stages. Thus, using volumetric could be a more efficient and flexible recognizer for our special case in the School of Medicine, Shanghai Jiao Tong University. In this article, we propose a novel solution to 3-D object recognition from volumetric data using a combination of three compact CNN models, low-cost SparseNet, and feature representation technique. We achieve an optimized network by estimating extra geometrical information comprising the surface normal and curvature into two separated neural networks. These two models provide supplementary information to each voxel data that consequently improve the results. The primary network model takes advantage of all the predicted features and uses these features in Random Forest (RF) for recognition purposes. Our method outperforms other methods in training speed in our experiments and provides an accurate result as good as the state-of-the-art.

Model Semantic Attention (SemAtt) With Hybrid Learning Separable Neural Network and Long Short-Term Memory to Generate Caption

Model Semantic Attention (SemAtt) With Hybrid Learning Separable Neural Network and Long Short-Term Memory to Generate Caption

A Hybrid Fault Detection and Fault Tolerant Control Based on kNN and NN Approaches for ABS Speed Sensors

To perform fault tolerance for Anti-lock Braking System (ABS), This paper proposes a hybrid Fault Detection and Fault Tolerant Control (FD-FTC) for ABS speed sensors. It utilizes a Fault Detection (FD) unit and a Data Construction (DC) unit. The first one, the FD unit, is based on a kNN classifier model with 99.9% fault detection accuracy to perform three tasks: early fault detection, fault location diagnosis, and excluding faulty signals from being utilized in further processes. On the other hand, the second one, the DC Unit, is based on two separate neural network models. These models have an MSE of 2.01139e-1 and a R2 of 999880 for the first model and an MSE of 1.12486e-0 and 0.999586 for the second model. They are employed to provide an estimated alternative signal for the ABS speed sensors. These estimated signals are employed to perform two tasks: confirming fault detection declared by the FD model and compensating for the excluded faulty signal to fulfill fault accommodation. Both methods are trained and tested with MATLAB and Simulink. Results demonstrate that the proposed hybrid method has the ability to accurately detect and tolerate sensor faults and fulfill its design purpose, especially during emergency braking. Index Terms— Anti-lock Braking System, Fault Tolerant Control, k-Nearest Neighbour, Neural Networks, Speed Sensor Fault.

Learned Lossless Image Compression Through Interpolation With Low Complexity

With the increasing popularity of deep learning in image processing, many learned lossless image compression methods have been proposed recently. One group of algorithms are based on scale-based auto-regressive models and can provide competitive compression performance while also allowing easily parallelized computations and short encoding/decoding times. However, they use large neural networks and have high computational requirements. This paper presents an interpolation based learned lossless image compression method which falls in the scale-based auto-regressive models group. The method achieves compression performance better than or on par with the recent scale-based auto-regressive models, yet requires more than 10x less neural network parameters (0.19M) and encoding/decoding computation complexity. These achievements are due to the contributions/findings in the overall system and neural network architecture design, such as sharing interpolator neural networks across different scales, using separate neural networks for different parameters of the probability distribution model and performing the processing in the YCoCg-R color space instead of the RGB color space.

Identification and risk stratification of coronary disease by artificial intelligence-enabled ECG

Atherosclerotic cardiovascular disease (ASCVD) is the leading cause of death worldwide, driven primarily by coronary artery disease (CAD). ASCVD risk estimators such as the pooled cohort equations (PCE) facilitate risk stratification and primary prevention of ASCVD but their accuracy is still suboptimal. Using deep electronic health record data from 7,116,209 patients seen at 70+ hospitals and clinics across 5 states in the USA, we developed an artificial intelligence-based electrocardiogram analysis tool (ECG-AI) to detect CAD and assessed the additive value of ECG-AI-based ASCVD risk stratification to the PCE. We created independent ECG-AI models using separate neural networks including subjects without known history of ASCVD, to identify coronary artery calcium (CAC) score ≥300 Agatston units by computed tomography, obstructive CAD by angiography or procedural intervention, and regional left ventricular akinesis in ≥1 segment by echocardiogram, as a reflection of possible prior myocardial infarction (MI). These were used to assess the utility of ECG-AI-based ASCVD risk stratification in a retrospective observational study consisting of patients with PCE scores and no prior ASCVD. The study period covered all available digitized EHR data, with the first available ECG in 1987 and the last in February2023. ECG-AI for identifying CAC ≥300, obstructive CAD, and regional akinesis achieved area under the receiver operating characteristic (AUROC) values of 0.88, 0.85, and 0.94, respectively. An ensembled ECG-AI identified 3, 5, and 10-year risk for acute coronary events and mortality independently and additively to PCE. Hazard ratios for acute coronary events over 3-years in patients without ASCVD that tested positive on 1, 2, or 3 versus 0 disease-specific ECG-AI models at cohort entry were 2.41 (2.14-2.71), 4.23 (3.74-4.78), and 11.75 (10.2-13.52), respectively. Similar stratification was observed in cohorts stratified by PCE or age. ECG-AI has potential to address unmet need for accessible risk stratification in patients in whom PCE under, over, or insufficiently estimates ASCVD risk, and in whom risk assessment over time periods shorter than 10 years is desired. Anumana.

Multi-fidelity neural network for uncertainty quantification of chemical reaction models

Surrogate models are often used to accelerate the uncertainty quantification (UQ) of chemical kinetic models. However, the construction of surrogate models usually requires the repetitive generation of high-fidelity input-output samples, which is time-consuming. In this work, we utilize a multi-fidelity neural network to speed up the surrogate model construction, yielding a multi-fidelity neural network-based surrogate model (MFNNSM). MFNNSM consists of two separate neural networks. The first neural network learns from high and low-fidelity samples, and then generates samples to train the second neural network for the subsequent UQ analysis. Based on the fact that the uncertainty similarity (sensitivity-based similarity) does exist between the predictions from the reduced and detailed combustion kinetics models, or between the predictions under different conditions, MFNNSM can use reduced model predictions as low-fidelity samples to generate detailed model predictions as the high-fidelity samples, or can transfer samples across different simulations conditions. To demonstrate the MFNNSM method, the ignition delay time (IDT) and laminar flame velocity (LFV) of methanol and n-decane are chosen as model prediction targets for UQ analysis. The results show that MFNNSM can achieve acceleration factors up to 6, by utilizing reduced model samples to generate detailed model samples under the same conditions, and when reusing the reduced model samples under different conditions, the acceleration factor can increase to 10. In addition, the influences of the relative error of the reduced models and the uncertainty similarity coefficients between combustion conditions on the performance of MFNNSM are further discussed, providing the guidance for future applications of MFNNSM.

An effective stacked autoencoder based depth separable convolutional neural network model for face mask detection.

The COVID-19 pandemic has been infecting the entire world over the past years. To prevent the spread of COVID-19, people have acclimatised to the new normal, which includes working from home, communicating online, and maintaining personal cleanliness. There are numerous tools required to prepare to compact transmissions in the future. One of these elements for protecting individuals from fatal virus transmission is the mask. Studies have indicated that wearing a mask may help to reduce the risk of viral transmission of all kinds. It causes many public places to take efforts to ensure that its guests wear adequate face masks and keep a safe distance from one another. Screening systems need to be installed at the doors of businesses, schools, government buildings, private offices, and/or other important areas. A variety of face detection models have been designed using various algorithms and techniques. Most of the articles in the previously published research have not worked on dimensionality reduction in conjunction with depth-wise separable neural networks. The necessity of determining the identities of people who do not cover their faces when they are in public is the driving factor for the development of this methodology. This research work proposes a deep learning technique to determine if a person is wearing mask or not and identifies whether it is properly worn or not. Stacked Auto Encoder (SAE) technique is implemented by stacking the following components: Principal Component Analysis (PCA) and Depth-wise Separable Convolutional Neural Network (DWSC-NN). PCA is used to reduce the irrelevant features in the images and resulted high true positive rate in the detection of mask. We achieved an accuracy score of 94.16% and an F1 score of 96.009% by the application of the method described in this research.