High Computational Processing Research Articles

IF-TONIR: Iteration-free Topology Optimization based on Implicit Neural Representations

Topology optimization holds great significance as a research topic in the field of mechanical engineering, aiming to design and optimize structures to achieve desired performance while adhering to specific constraints. However, its high computational complexity and iterative optimization process severely impact the efficiency, which presents substantial obstacles to its practical applications. To tackle this challenge, recent research is dedicated to the advancement of iteration-free topology optimization methods that leverage neural networks and deep learning, aiming to directly predict optimal structures through optimization problem configurations. In this paper, we propose IF-TONIR, a novel data-driven topology optimization method that utilizes implicit neural representations. Our approach employs signed distance fields to represent structures, offering compact and smooth representations that effectively eliminate the checkerboard phenomenon commonly observed in density-based methods. IF-TONIR leverages Conditional Variational Autoencoders, which use a CNN-based encoder and a MLP-based decoder to learn and reconstruct optimal structures. We employ the features extracted from physical information as conditions to guide the decoder in generating optimal structures that adhere to specific design domain shapes and boundary conditions. Furthermore, we propose the integration of a topological loss based on persistent homology to train the model. This loss function effectively penalizes the existence of structural disconnections in the reconstructed output, thereby enhancing the overall physical reliability of the generated structures. Various experiments have demonstrated that our iteration-free topology optimization method based on implicit representations can accurately identify regions of high strain energy and generate continuous structures with low compliance. The methods also holds the theoretical capability of outputting optimal structures at any desired resolution. Our code and dataset are available on https://github.com/jbHu67/IF-TONIR.git

Continual learning for topological geo-localisation

Geo-localisation from a single aerial image for Uncrewed Aerial Vehicles (UAVs) is an alternative to other vision-based methods, such as visual Simultaneous Localisation and Mapping (SLAM), seeking robustness under GPS failure. Due to the success of deep learning and the fact that UAVs can carry a low-cost camera, we can train a Convolutional Neural Network (CNN) to predict position from a single aerial image. However, conventional CNN-based methods adapted to this problem require off-board training that involves high computational processing time and where the model can not be used in the same flight mission. In this work, we explore the use of continual learning via latent replay to achieve online training with a CNN model that learns during the flight mission GPS coordinates associated with single aerial images. Thus, the learning process repeats the old data with the new ones using fewer images. Furthermore, inspired by the sub-mapping concept in visual SLAM, we propose a multi-model approach to assess the advantages of using compact models learned continuously with promising results. On average, our method achieved a processing speed of 150 fps with an accuracy of 0.71 to 0.85, demonstrating the effectiveness of our methodology for geo-localisation applications.

Lite and Efficient Deep Learning Model for Bearing Fault Diagnosis Using the CWRU Dataset

Bearing defects are a common problem in rotating machines and equipment that can lead to unexpected downtime, costly repairs, and even safety hazards. Diagnosing bearing defects is crucial for preventative maintenance, and deep learning models have shown promising results in this field. On the other hand, the high complexity of these models can lead to high computational and data processing costs, making their practical implementation challenging. Recent studies have focused on optimizing these models by reducing their size and complexity, but these methods often compromise classification performance. This paper proposes a new approach that reduces the dimensionality of input data and optimizes the model structure simultaneously. A much lower input data dimension than that of existing deep learning models was achieved by downsampling the vibration sensor signals used for bearing defect diagnosis and constructing spectrograms. This paper introduces a lite convolutional neural network (CNN) model with fixed feature map dimensions that achieve high classification accuracy with low-dimensional input data. The vibration sensor signals used for bearing defect diagnosis were first downsampled to reduce the dimensionality of the input data. Next, spectrograms were constructed using the signals of the minimum interval. Experiments were conducted using the vibration sensor signals from the Case Western Reserve University (CWRU) dataset. The experimental results show that the proposed method could be highly efficient in terms of computation while maintaining outstanding classification performance. The results show that the proposed method outperformed a state-of-the-art model for bearing defect diagnosis under different conditions. This approach is not limited to the field of bearing failure diagnosis, but could be applied potentially to other fields that require the analysis of high-dimensional time series data.

Hybrid Artificial Bee Colony and Improved Simulated Annealing for the Capacitated Vehicle Routing Problem

Capacitated Vehicle Routing Problem (CVRP) is a type of NP-Hard combinatorial problem that requires a high computational process. In the case of CVRP, there is an additional constraint in the form of a capacity limit owned by the vehicle, so the complexity of the problem from CVRP is to find the optimum route pattern for minimizing travel costs which are also adjusted to customer demand and vehicle capacity for distribution. One method of solving CVRP can be done by implementing a meta-heuristic algorithm. In this research, two meta-heuristic algorithms have been hybridized: Artificial Bee Colony (ABC) with Improved Simulated Annealing (SA). The motivation behind this idea is to complete the excess and the lack of two algorithms when exploring and exploiting the optimal solution. Hybridization is done by running the ABC algorithm, and then the output solution at this stage will be used as an initial solution for the Improved SA method. Parameter testing for both methods has been carried out to produce an optimal solution. In this study, the test was carried out using the CVRP benchmark dataset generated by Augerat (Dataset 1) and the recent CVRP dataset from Uchoa (Dataset 2). The result shows that hybridizing the ABC algorithm and Improved SA could provide a better solution than the basic ABC without hybridization.

An Efficient Model for Accurate Evaluation of Consumption Pattern in Distribution System Reconfiguration

Consumption patterns of electric power systems are important for distribution companies, because of their significant impact on energy losses amount. Therefore, some incentives are suggested by distribution companies to energy consumers for correcting their consumption manner. For a specific load pattern, distribution system reconfiguration (DSR) is an effective way to mitigate energy losses. Hence, some research works have included load variations in the DSR problem to show the importance of consumption patterns in reconfiguration decisions. However, some of the specialized literature has ignored load changes in their reconfiguration models due to the high computational burden and processing time. On the other hand, the energy losses are calculated inaccurately if the consumption pattern is neglected. Consequently, the main goal of this article is to investigate load pattern impact on switching sequences to find out how much is load profile important for minimization of energy losses in DSR. The evaluations were carried out for three well-known distribution systems using a classic optimization tool, the A Mathematical Programming Language.

A Randles Circuit Parameter Estimation of Li-Ion Batteries With Embedded Hardware

Accurate modelling of electrochemical sources is very important to predict how a source will perform in specific applications related to the load or environmental parameters. A Randles circuit is considered as a reliable equivalent electrical circuit in studying and modelling various electrochemical systems and processes. The classical parameter estimation approach based on use of software packages (ZSim, MEISP, LEVMW, etc.) requires for the high computational performance processing units, decreasing the reliability as proper maintenance actions can be delayed because of offline analysis. Advancing the state of the art, we propose a low-complexity approach for embedded hardware-based parameter estimation of the Randles circuit. Our non-iterative method uses only the measured real and imaginary part of impedance, with numerical approximation of the first derivative of real/imaginary part quotient, to create closed-form expressions with a unique solution. The initial estimated values are available from partial dataset (after measurement at only 3 frequencies). Moreover, it is not software platform-specific, which enables a high level of portability. The presented method is verified with theoretical, numerical and experimental analysis, with more than 1000 datasets. We also demonstrated the applicability in parameter estimation of the Randles circuit of a Li-ion battery. Finally, we verified suitability for embedded hardware platforms, with deployment on a microcontroller-based platform with a clock speed of 16 MHz and 8 kB of SRAM. Reliable parameter estimation processing of a 100-points dataset was performed in just 106 milliseconds with 1% relative error, requiring less than 53 mJ of energy.

Multilateration Approach for Wide Range Visible Light Indoor Positioning System Using Mobile CMOS Image Sensor

Visible light positioning (VLP) is a cost-effective solution to the increasing demand for real-time indoor positioning. However, owing to high computational costs and complicated image processing procedures, most of the existing VLP systems fail to deliver real-time positioning ability and better accuracy for image sensor-based large-area indoor environments. In this study, an effective method is proposed to receive coordinate information from multiple light-emitting diode (LED) lights simultaneously. It provides better accuracy in large experimental areas with many LEDs by using a smartphone-embedded image sensor as a terminal device and the existing LED lighting infrastructure. A flicker-free frequency shift on–off keying line coding modulation scheme was designed for the positioning system to ensure a constant modulated frequency. We tested the performance of the decoding accuracy with respect to vertical and horizontal distance, which utilizes a rolling shutter mechanism of a complementary metal-oxide-semiconductor image sensor. The experimental results of the proposed positioning system can provide centimeter-level accuracy with low computational time, rendering it a promising solution for the future direction of large-area indoor positioning systems.

Embedded Intelligence on FPGA: Survey, Applications and Challenges

Embedded intelligence (EI) is an emerging research field and has the objective to incorporate machine learning algorithms and intelligent decision-making capabilities into mobile and embedded devices or systems. There are several challenges to be addressed to realize efficient EI implementations in hardware such as the need for: (1) high computational processing; (2) low power consumption (or high energy efficiency); and (3) scalability to accommodate different network sizes and topologies. In recent years, an emerging hardware technology which has demonstrated strong potential and capabilities for EI implementations is the FPGA (field programmable gate array) technology. This paper presents an overview and review of embedded intelligence on FPGA with a focus on applications, platforms and challenges. There are four main classification and thematic descriptors which are reviewed and discussed in this paper for EI: (1) EI techniques including machine learning and neural networks, deep learning, expert systems, fuzzy intelligence, swarm intelligence, self-organizing map (SOM) and extreme learning; (2) applications for EI including object detection and recognition, indoor localization and surveillance monitoring, and other EI applications; (3) hardware and platforms for EI; and (4) challenges for EI. The paper aims to introduce interested researchers to this area and motivate the development of practical FPGA solutions for EI deployment.

A Modified Corner Detector for SAR Images Registration

As a first step for image processing operations, detection of corners is a vital procedure where it can be applied for many applications as feature matching, image registration, image mosaicking, image fusion, and change detection. Image registration can be defined as process of getting the misalignment of pixel's position between two or more images. In this paper, a modified corner detector named Synthetic Aperture Radar-Phase Congruency Harris (SAR-PCH) based on a combination between both phase congruency, named later PC, and Harris corner detector is proposed where PC image can supply fundamental and significative features although the complex changes of intensities. Also, the proposed approach overcomes the Harris limitation concerning the noise since the Harris is more sensitive to the noise. The performance was similitude with Shi-Tomasi, FAST, and Harris corner detectors where experiments are conducted first with simulated images and second with real ones. Mean square error (MSE) and peak signal-to-noise ratio (PSNR) are used for the simile. Experimental results, carried out in a standard computer, verify its effectiveness where it utilizes the privileges of image constitutional depicting, allowing extraction of the most powerful key points since it preserves robustness of co-registration process using image frequency properties which are not variant to illumination. Reasonable results compared to the state of art method as Shi-Tomasi, FAST, and Harris algorithms were achieved on the expense of high computational processing time that can be recovered using hardware having high capabilities.

Error-Control Truncated SVD Technique for In-Network Data Compression in Wireless Sensor Networks

In-network data compression plays an important role in the elimination of redundant time-series data in a wireless sensor network (WSN). Inconsistency of data and high computational process in cluster formation remain to be challenging issues of in-network data compression particularly for energy-constraint WSNs. This paper develops a new data clustering technique for in-network data preprocessing and compression called Error-Control Truncated Singular Value Decomposition (ETSVD) to achieve online outlier detection and adaptive data compression. The ETSVD is divided into two modules which are Adaptive Recursive Outlier Detection and Smoothing (ARODS) and Adaptive Data Compression (DC). Firstly, the ARODS pre-processes the collected data for outlier detection and smoothing in order to improve the data quality. Secondly, the DC decomposes the pre-processed data into vector space to compress the spatio-temporal correlated data based on the predefined error threshold at the sending end. After the compression of correlated data, the distinct decomposed data are reconstructed at the receiver end which is performed offline. The simulation results show that the proposed technique is able to compress 91.49% of spatio-temporal environmental temperature data with reconstruction error having a minimum tolerance of $\pm 1.0^\circ $ C. The performance improvement of ETSVD in terms of error and accuracy compared to the performance of conventional SVD are 85.26% and 33.49%, respectively. Moreover, the ETSVD provides efficient error-control data preprocessing and compression solutions within the networks with minimum space and time complexities.

An Integrated Object and Machine Learning Approach for Tree Canopy Extraction from UAV Datasets

Unmanned aerial vehicles (UAV) have emerged as new platforms for acquiring ultra-high resolution images, which are challenging for extraction of features using conventional image processing approaches. Tree canopies are required to be constantly monitored for better planning and management. UAV is currently one way to survey canopies over a large area for precisely estimating their geometry. Conventional segmentation techniques are extensively used for image feature extraction. However, they lack in accuracy and require high computational processing when used for ultra-high resolution UAV datasets. These issues can be handled by superpixel segmentation algorithms which have good boundary adherence and are computationally efficient. Simple linear iterative clustering (SLIC) is a subset of superpixel segmentation technique which uses minimum tuning parameters making it most efficient. As the random forest is known for handling multiple attributes and robustness, it can be used for classifying and extracting features from segmented image generated using SLIC. The present study mainly focuses on the automation for extraction of tree canopies along with their object-based attributes from the UAV dataset. The data acquisition was carried out using Trimble UX5 fixed-wing UAV which was further orthorectified at a spatial resolution of 13 cm. The ortho-image was further segmented using SLIC algorithm. Canopy segments are then identified and classified using random forest, which is then merged into trees objects on the basis of their proximity. Accuracy assessment was then carried out for extracted tree canopies and was found that the aforesaid approach could achieve 93% similarity index. The current study highlights the potential of using SLIC segmentation and random forest classification method for tree canopy extraction from the ultra-high resolution ortho-image derived from UAV platforms.

Wireless differential resonance frequency pressure sensor: Circuital model and prototype validation

This article presents a pressure sensor composed of a set of three Archimedean spiral inductors. One of them is below the other two concentric spirals, where the upper coils are over a flexible membrane. When external pressure is applied over the sensor, the membrane deforms and changes the distance between the lower spirals and the upper inside coil, the capacitance, and its resonance frequency. However, once the sensor is in contact with a conductive medium, a change in the effective permittivity surrounding the sensor takes place reshaping the resonant frequencies. To improve the sensor response, we introduce a reader resonator sensor system. We present the comparative response of the system as a prototype, an analytical model, and a finite element model (FEM). The prototype system is fabricated using a photolithographic process for the elements of the sensor over a polymethylsiloxane and glass substrate, while the reader is fabricated over a FR4 substrate. COMSOL Multiphysics is used for the FEM simulation of the system to evaluate its behavior. The analytical model involves the integration of techniques to obtain the effective permittivity for each element plane and a coupled magnetic resonance to represent the interaction of the parts. We obtain a correspondence between the resonances of the different elements, with a maximum displacement frequency of 240 MHz for the proposed model, and 80 MHz in the FEM case related to the implemented prototype. The accuracy of this model allows a fast analysis to improve the design of this type of topology without high computational processing.

Impact of compressed and down-scaled training images on vehicle detection in remote sensing imagery

Vehicle detection in remote sensing imagery is a prominent issue over the last few years. In this application, the processing of optical remote sensing images becomes critical due to the complex environment, large size, occlusions and color variations. However, several approaches have been proposed to improve the training process but still, the efforts are moving towards optimal solutions. The training process is time-consuming and a large amount of memory is required to store those training images. Numerous traditional state-of-the-art approaches are suffering from problematic high computational time. In this paper, a new training methodology which is based on compressed and down-scaled images is implemented to reduce the training time. The training images are compressed at Quality Factor (QF) of 50 and down-scaled by scale factor of 0.5 to evaluate the performance for vehicle detection. The existing approaches of computer vision are taking advantage of high computational Graphical Processing Units (GPUs) to speed up the training process. The proposed framework is also a better way to reduce the computational time. To compare performance, we have trained the RCNN, Fast-RCNN, Faster-RCNN and Cascade detectors by using three types of training image sets and several experiments have been performed. More specifically, our approach makes the training faster than the training based on original images and training based on compressed images provides optimal results.

Low complexity performance assessment of a sensor array via unscented transformation

Due to the advances on electronics, applications of antenna array signal processing are becoming more frequent. When employing antenna arrays for beamforming, the signal to interference and noise ratio (SINR) should be assessed. Many factors can affect the SINR such as the array element positioning error and the direction of arrival (DOA) estimation error. In these cases, the assessment is traditionally performed via the SINR average obtained using Monte Carlo (MC) simulations. However, this approach requires a great amount of realizations that demand a high computational effort and processing time due to its slow convergence. In this paper, we propose a low complexity performance assessment of the average SINR via unscented transformation. Compared to MC simulations, our proposed method requires only a few trials and has a negligible computational complexity, yet giving a comparable SINR when the DOA estimation is perturbed. When the antenna elements positioning is perturbed, a multivariate scenario arises. For multivariate scenario the proposed scheme has an exponential increase in complexity, therefore, still being advantageous for a small number of antennas.

Optimized Cross-Layered Unequal Error Protection for SPIHT Coded Wireless Video Transmission

Transmission of coded video over erroneous channels (e.g., Internet and wireless networks) is always a challenging task. Error control strategies, to combat the channel errors, are generally applied at individual network layers leading to inefficient utilization of resources. The cross-layer design has emerged as alternative for efficient utilization of the available resources, such as bandwidth, delay, and transmission power to provide best quality of the reconstructed video. Furthermore, to efficiently utilize the resources, cross-layer error control parameters need to be optimized. Since, the optimal error control parameters change with channel conditions, new optimal values need to be calculated every time the channel condition changes. However, optimization of the parameters at the time of video transmission requires very high computational processing power otherwise it may induce a large transmission delay which is intolerable in video communication. In this paper, we propose an optimized cross-layered unequal error protection (UEP) scheme targeted for hand-held devices which have low processing power, uses pre-computed look-up table (LUT) that stores the optimal cross-layer error control parameters for a range of channel conditions. LUT is constructed using offline optimization of the parameters. For real time video communication, given a channel condition, the corresponding error control parameters are read from a LUT and are passed to the corresponding network layers. Thus, the proposed transmission system is simple and delivers best end-to-end quality of the video. Simulation results show that the proposed cross-layered UEP system can achieve up to 14–17 dB and 2–4 dB improvement in video quality over individual application layer optimized UEPs in additive white Gaussian noise and fading channel respectively at very poor channel conditions. Furthermore, the proposed system outperforms other contemporary methods.

Real-time stress detection of monocrystalline silicon by laser irradiation using Mach–Zehnder interferometry

The effect of changes in laser-induced stress upon irradiation of monocrystalline silicon was studied in detail using Mach–Zehnder interferometer, high speed camera and computer processing system, real-time detection of stress distribution and stress evolvement under different laser fluences and pulse widths. After irradiation, the changes of interference fringes were used to calculate the stress value. The results show that the stresses increased with the increase of the laser fluence. The formation of stress could be explained using thermoelastic theory. The cleavage plane’s dislocation appears in the following sequence: (1 1 1) plane appears and then the dislocation slip of (1 1 0) cleavage plane appears. In addition, it is found that cleavage plane of (1 1 1) mainly exist in the spot and cleavage plane of (1 1 0) mainly exists in the vicinity of spot radius. Stress areas mainly exist in the thin layer on the surface of silicon. Furthermore, stress field was analysed by the finite-element method to get a better understanding of the formation and distribution of stress. Under the same experimental conditions, the numerical results are in agreement with the experimental values.

Design and development of automotive blind spot detection radar system based on ROI pre-processing scheme

In the conventional 2D-FFT based target detection method, all range-Doppler cells are computed by FFT (Fast Fourier Transform) and scanned by CA-CFAR (Cell-Averaging Constant False Alarm Rate) detection. This results in high computational complexity and long processing time. In this paper, we developed an automotive 24 GHz BSD (Blind Spot Detection) FMCW (Frequency Modulated Continuous Wave) radar with a low complexity target detection architecture based on a ROI (Region Of Interest) pre-processing scheme. In the real BSD zone, because the number of cars to be detected is limited, the designed method only extracts their velocities corresponding to the range ROIs in which real targets exist. Moreover, the presence probability of vehicles with the same range-bin but different velocities is very low. Thus, in the designed method, some Doppler ROIs cells with a high magnitude are only applied for CA-CFAR detection. This architecture can dramatically reduce the amount of data to be processed compared to that of the conventional 2D FFT based method, resulting in enhanced processing time. We developed a 24 GHz FMCW radar system composed a transceiver, antennas, and signal processing module. The designed algorithm was implemented in a tiny micro-processor of the signal processing module. By implementing our proposed algorithm in the developed 24 GHz FMCW radar system in an anechoic chamber and a real road, we verified that the range and velocity of a car occupying the BSD zone were detected. Compared to that of the conventional method, the reduction ratio of the total processing time was measured to be 52.4 %.

Motion artifact removal from single channel electroencephalogram signals using singular spectrum analysis

Abstract In ambulatory electroencephalogram (EEG) health care systems, recorded EEG signals often contaminated by motion artifacts. In this paper, we proposed a singular spectrum analysis (SSA) technique with new grouping criteria to remove the motion artifact from a single channel EEG signal. In order to remove the motion artifact from a single channel EEG signal, we considered the eigenvectors (basis vectors) corresponding to motion artifact are grouped or identified based on their local mobility, which is a signal complexity measure. However, as the local mobility of eigenvectors associated to the motion artifact are small, a threshold of 0.1 is set to identify them. The motion artifact signal is estimated using the identified eigenvectors and subtracted from the contaminated EEG signal to obtain the corrected EEG signal. The proposed technique is tested on 21 single channel real EEG signals contaminated by motion artifact and compared the results with the existing combined ensemble empirical mode decomposition and canonical correlation analysis (EEMD-CCA) technique. The simulation results show that the proposed modified SSA enjoys an improvement in the signal to noise ratio and the percentage reduction in artifact. Moreover, as the ambulatory EEG systems are battery operated, use of high computational signal processing techniques will reduce the battery lifetime. Hence, low computational signal processing techniques are greatly demanded in such applications. Thus, we have also evaluated the computational complexity of the proposed technique and compared with EEMD-CCA. We found that the proposed modified SSA technique significantly reduces the computational complexity and thereby lower power consumption compared to the EEMD-CCA.

A mid-level video representation based on binary descriptors: A case study for pornography detection

With the growing amount of inappropriate content on the Internet, such as pornography, arises the need to detect and filter such material. The reason for this is given by the fact that such content is often prohibited in certain environments (e.g., schools and workplaces) or for certain publics (e.g., children). In recent years, many works have been mainly focused on detecting pornographic images and videos based on visual content, particularly on the detection of skin color. Although these approaches provide good results, they generally have the disadvantage of a high false positive rate since not all images with large areas of skin exposure are necessarily pornographic images, such as people wearing swimsuits or images related to sports. Local feature based approaches with Bag-of-Words models (BoW) have been successfully applied to visual recognition tasks in the context of pornography detection. Even though existing methods provide promising results, they use local feature descriptors that require a high computational processing time yielding high-dimensional vectors. In this work, we propose an approach for pornography detection based on local binary feature extraction and BossaNova image representation, a BoW model extension that preserves more richly the visual information. Moreover, we propose two approaches for video description based on the combination of mid-level representations namely BossaNova Video Descriptor (BNVD) and BoW Video Descriptor (BoW-VD). The proposed techniques are promising, achieving an accuracy of 92.40%, thus reducing the classification error by 16% over the current state-of-the-art local features approach on the Pornography dataset.

Finite element modelling of hysteresis, degradation and failure of dowel type timber joints

A method for macro modelling timber-to-timber dowel type joints under shear loads using finite elements is presented. The proposed model, which includes a combination of nonlinear one-dimensional elements, is able to properly represent the hysteretic behaviour, stiffness degradation and failure of this type of joints. Experimental data from mechanical tests was used to define the macro model parameters. The proposed methodology does not require a high computational processing cost and, consequently, can be used for the nonlinear analysis of complex structural systems with a large number of connections.