Raw Input Data Research Articles

Strain-based delamination prediction in fatigue loaded CFRP coupon specimens by deep learning and static loading data

Machine learning (ML) methods for the structural health monitoring (SHM) of composite structures rely on sufficient domain knowledge as they typically demand to extract damage-sensitive features from raw data before training the ML model. In practice, prior knowledge is not available in most cases. Deep learning (DL) methods, on the other hand, can obtain higher-level features from raw input data and have proven superior in several applications. This paper proposes a Convolutional Neural Network (CNN) based approach for the delamination prediction in CFRP double cantilever beam (DCB) specimens using raw local array strain measurements via distributed optical fiber sensors. The conventional CNN architecture is modified to perform regression, as the delamination size is a continuous value. 1D and 2D CNN architectures are deployed and compared and different techniques are exploited to encode 1D spatial strain pattern series as 2D images. Raw strain patterns collected during static testing are used to train the CNNs, while testing is performed on unseen raw fatigue strain patterns, showing the CNN ability to automatically extract discriminative features from the non-pre-processed static strain pattern-based signals that generalize to raw fatigue signals as well. This strategy has the potential to reduce fatigue testing expenditures while also shortening the time required to gather training data.

Dynamic mode decomposition with core sketch

With the increase in collected data volumes, either from experimental measurements or high fidelity simulations, there is an ever-growing need to develop computationally efficient tools to process, analyze, and interpret these datasets. Modal analysis techniques have gained great interest due to their ability to identify patterns in the data and extract valuable information about the system being considered. Dynamic mode decomposition (DMD) relies on elements of the Koopman approximation theory to compute a set of modes, each associated with a fixed oscillation frequency and a decay/growth rate. Extracting these details from large datasets can be computationally expensive due to the need to implement singular value decomposition of the input data matrix. Sketching algorithms have become popular in numerical linear algebra where statistical theoretic approaches are utilized to reduce the cost of major operations. A sketch of a matrix is another matrix, which is significantly smaller, but still sufficiently approximates the original system. We put forth an efficient DMD framework, SketchyDMD, based on a core sketching algorithm that captures information about the range and corange (their mutual relationship) of input data. The proposed sketching-based framework can accelerate various portions of the DMD routines, compared to classical methods that operate directly on the raw input data. We conduct numerical experiments using the spherical shallow water equations as a prototypical model in the context of geophysical flows. We show that the proposed SketchyDMD is superior to existing randomized DMD methods that are based on capturing only the range of the input data.

A system for organizing, collecting, and presenting open-source intelligence

Open-source intelligence is a rapidly expanding area of the security and intelligence industry, involving the collection of internet located open data from various sources, turning that data into actionable intelligence, which is reused where possible and relevant. While creating or processing the raw input data capturing and managing the corresponding provenance information, e.g., workflow, state, raw evidence, reports, and summaries, that simplifies its retrieval and reuse is essential. In comparison, scientific workflows and tools that support them are routinely used in the majority of academic research disciplines, managing diverse sets of data resources and their provenance. Based on the techniques established within the academic community, we have developed a system for managing this open-source intelligence data and associated provenance information. This will enhance the efficiency of retrieving stored data products and reusing them to support intelligence-led security decision-making. The open-source intelligence company partnered within this project has an operational envelope that includes collecting and analyzing personal subject information. Therefore, they must understand the scope of their data holdings appropriately, especially in light of obligations under the General Data Protection Regulation. The system developed allows for tracking requests for intelligence products, ownership of the collection, analysis and generation of intelligence briefs, and tracking the delivery of those final products to the customer for future billing. This adds further layers of efficiency to operations and hence reduces the costs of producing intelligence products.

A Hybrid Deep Learning Framework for Intelligent Predictive Maintenance of Cyber-physical Systems

The proliferation of cyber-physical systems (CPSs) and the advancement of the Internet of Things (IoT) technologies have led to explosive digitization of the industrial sector. It offers promising perspectives for high reliability, availability, maintainability, and safety production process, but also makes the systems more complex and challenging for health assessment. To deal with these challenges, one needs to develop a robust approach to monitor and assess the system’s health state. In this article, a practical and effective hybrid deep learning multi-task framework integrating the advantages of convolutional neural network (CNN) and long short-term memory (LSTM) neural network to reflect the relatedness of remaining useful life prediction with health status detection process for complex multi-object systems in CPS environment is developed. The CNN is used as a feature extractor to compress condition monitoring data and directly extract significant spatiotemporal features from raw multi-sensory input data. The LSTM is used to capture long-term temporary dependency features. The advantages of the proposed hybrid deep learning framework have been verified on the popular NASA’s C-MAPSS dataset. The experimental study compares this approach to the existing methods using the same dataset. The results suggest that the proposed hybrid CNN-LSTM model is superior to existing methods, including traditional machine learning and deep learning-based methods. The proposed framework can provide strong support for the health management and maintenance strategy development of complex multi-object systems.

Tucker's congruence regressive feature projected Tversky discriminant multiple instance learning boost data classification for school student dropout prediction

AbstractPrediction of student dropout in high school is a significant concern in education that affects both a state's education system and its financial system. Early prediction of school student dropout is not an easy issue to resolve since many factors that can influence student retention. The traditional classification techniques were used to solve this problem normally but the higher accuracy was not obtained. In order to improve the accuracy, a novel technique called Tucker's Congruence Regressive Target Feature Matching‐based Tversky Discriminant MIL Boost Data Classification (TCRTFM‐TDMBDC) is introduced. The proposed TCRTFM‐TDMBDC technique consists of four different processes namely data preprocessing, feature extraction, feature selection, and classification. At first, the data preprocessing is carried out for cleaning and altering the raw input data into a valuable and understandable format to minimize the complexity of the classification. After the preprocessing, the feature extraction is carried out by applying Modified Tucker's congruence correlative regression. Thirdly, the feature selection process is performed using Gaussian kernelized target projection feature matching to select the feature subset for accurate classification with minimum time consumption. Finally, the ensemble technique called Tversky Indxive generalized discriminant MIL boost is applied for classifying the given input student data with help of the weak learners. Based on the classification results, the student dropout prediction is accurately performed with minimum time. Experimental results reveal that the proposed technique noticeably predicts student dropout by means of prediction accuracy, precision, recall, F‐measure, and prediction time with respect to the number of student data. The discussed results illustrate that the proposed TCRTFM‐TDMBDC technique achieves higher accuracy with minimum prediction time than the state‐of‐the‐art methods.

Optimal prediction of user mobility based on spatio-temporal matching

Position tracking has become a critical key component for a huge variety of devices, ranging from mobile telephone location tracking to biodiversity monitoring. The majority of location-based services rely mostly on the user’s ongoing and prospective position, indicating a growing need of forecasting the user’s future location. Together with position prediction, forecasting the trajectories between two terminals is beneficial, because it enables to optimize the travel direction between them. This study tackles the problem of increasing prediction accuracy to its maximum level. The proposed work undergoes two major phases: feature extraction and prediction. Initially, antecedent and consequent features, spatio-temporal matching based features, and matching users based features can all be generated from the raw input data. For more precise prediction the most relevant features are extracted. The features will then be fed into the prediction algorithm, which will forecast user mobility. The prediction phase is constructed with an optimized convolutional neural network (CNN). Moreover, the weight of CNN is fine-tuned via a new improved butterfly optimization algorithm (IBOA), which is a conceptual improvement of standard BOA. At last, the supremacy of the presented approach is proved over other models with respect to varied measures. The accuracy of the proposed work is 18.33%, 26.67%, 33.33%, 55.2%, and 61.67% better than the existing models like HS–EH, GAF-WO, CNN, and GSTF.

Engineering solutions to breath tests based on an e-nose system for silicosis screening and early detection in miners

This study aims to develop an engineering solution to breath tests using an electronic nose (e-nose), and evaluate its diagnosis accuracy for silicosis. Influencing factors of this technique were explored. 398 non-silicosis miners and 221 silicosis miners were enrolled in this cross-sectional study. Exhaled breath was analyzed by an array of 16 organic nanofiber sensors along with a customized sample processing system. Principal component analysis was used to visualize the breath data, and classifiers were trained by two improved cost-sensitive ensemble algorithms (random forest and extreme gradient boosting) and two classical algorithms (K-nearest neighbor and support vector machine). All subjects were included to train the screening model, and an early detection model was run with silicosis cases in stage I. Both 5-fold cross-validation and external validation were adopted. Difference in classifiers caused by algorithms and subjects was quantified using a two-factor analysis of variance. The association between personal smoking habits and classification was investigated by the chi-square test. Classifiers of ensemble learning performed well in both screening and early detection model, with an accuracy range of 0.817–0.987. Classical classifiers showed relatively worse performance. Besides, the ensemble algorithm type and silicosis cases inclusion had no significant effect on classification (p > 0.05). There was no connection between personal smoking habits and classification accuracy. Breath tests based on an e-nose consisted of 16× sensor array performed well in silicosis screening and early detection. Raw data input showed a more significant effect on classification compared with the algorithm. Personal smoking habits had little impact on models, supporting the applicability of models in large-scale silicosis screening. The e-nose technique and the breath analysis methods reported are expected to provide a quick and accurate screening for silicosis, and extensible for other diseases.

DeepNEC: a novel alignment-free tool for the identification and classification of nitrogen biochemical network-related enzymes using deep learning.

Nitrogen is essential for life and its transformations are an important part of the global biogeochemical cycle. Being an essential nutrient, nitrogen exists in a range of oxidation states from +5 (nitrate) to -3 (ammonium and amino-nitrogen), and its oxidation and reduction reactions catalyzed by microbial enzymes determine its environmental fate. The functional annotation of the genes encoding the core nitrogen network enzymes has a broad range of applications in metagenomics, agriculture, wastewater treatment and industrial biotechnology. This study developed an alignment-free computational approach to determine the predicted nitrogen biochemical network-related enzymes from the sequence itself. We propose deepNEC, a novel end-to-end feature selection and classification model training approach for nitrogen biochemical network-related enzyme prediction. The algorithm was developed using Deep Learning, a class of machine learning algorithms that uses multiple layers to extract higher-level features from the raw input data. The derived protein sequence is used as an input, extracting sequential and convolutional features from raw encoded protein sequences based on classification rather than traditional alignment-based methods for enzyme prediction. Two large datasets of protein sequences, enzymes and non-enzymes were used to train the models with protein sequence features like amino acid composition, dipeptide composition (DPC), conformation transition and distribution, normalized Moreau-Broto (NMBroto), conjoint and quasi order, etc. The k-fold cross-validation and independent testing were performed to validate our model training. deepNEC uses a four-tier approach for prediction; in the first phase, it will predict a query sequence as enzyme or non-enzyme; in the second phase, it will further predict and classify enzymes into nitrogen biochemical network-related enzymes or non-nitrogen metabolism enzymes; in the third phase, it classifies predicted enzymes into nine nitrogen metabolism classes; and in the fourth phase, it predicts the enzyme commission number out of 20 classes for nitrogen metabolism. Among all, the DPC + NMBroto hybrid feature gave the best prediction performance (accuracy of 96.15% in k-fold training and 93.43% in independent testing) with an Matthews correlation coefficient (0.92 training and 0.87 independent testing) in phase I; phase II (accuracy of 99.71% in k-fold training and 98.30% in independent testing); phase III (overall accuracy of 99.03% in k-fold training and 98.98% in independent testing); phase IV (overall accuracy of 99.05% in k-fold training and 98.18% in independent testing), the DPC feature gave the best prediction performance. We have also implemented a homology-based method to remove false negatives. All the models have been implemented on a web server (prediction tool), which is freely available at http://bioinfo.usu.edu/deepNEC/.

Metall: A persistent memory allocator for data-centric analytics

Data analytics applications transform raw input data into analytics-specific data structures before performing analytics. Unfortunately, such data ingestion steps are often more expensive than analytics. In addition, various types of NVRAM devices are already used in many HPC systems today. Such devices will be useful for storing and reusing data structures beyond a single process life cycle.We developed Metall, a persistent memory allocator built on top of the memory-mapped file mechanism. Metall enables applications to transparently allocate custom C++ data structures into various types of persistent memories. Metall incorporates a concise and high-performance memory management algorithm inspired by Supermalloc and the rich C++ interface developed by Boost.Interprocess library.On a dynamic graph construction workload, Metall achieved up to 11.7x and 48.3x performance improvements over Boost.Interprocess and memkind (PMEM kind), respectively. We also demonstrate Metall’s high adaptability by integrating Metall into a graph processing framework, GraphBLAS Template Library. This study’s outcomes indicate that Metall will be a strong tool for accelerating future large-scale data analytics by allowing applications to leverage persistent memory efficiently.

Steam++ An Extensible End-to-end Framework for Developing IoT Data Processing Applications in the Fog

IoT applications usually rely on cloud computing services to perform data analysis such as filtering, aggregation, classification, pattern detection, and prediction. When applied to specific domains, the IoT needs to deal with unique constraints. Besides the hostile environment such as vibration and electricmagnetic interference, resulting in malfunction, noise, and data loss, industrial plants often have Internet access restricted or unavailable, forcing us to design stand-alone fog and edge computing solutions. In this context, we present STEAM++, a lightweight and extensible framework for real-time data stream processing and decision-making in the network edge, targeting hardware-limited devices, besides proposing a micro-benchmark methodology for assessing embedded IoT applications. In real-case experiments in a semiconductor industry, we processed an entire data flow, from values sensing, processing and analysing data, detecting relevant events, and finally, publishing results to a dashboard. On average, the application consumed less than 500kb RAM and 1.0% of CPU usage, processing up to 239 data packets per second and reducing the output data size to 14% of the input raw data size when notifying events.

Data Mining in Employee Healthcare Detection Using Intelligence Techniques for Industry Development.

Background Even in today's environment, when there is a plethora of information accessible, it may be difficult to make appropriate choices for one's well-being. Data mining, machine learning, and computational statistics are among the most popular arenas of training today, and they are all aimed at secondary empowered person in making good decisions that will maximize the outcome of whatever working area they are involved with. Because the degree of rise in the number of patient roles is directly related to the rate of people growth and lifestyle variations, the healthcare sector has a significant need for data processing services. When it comes to cancer, the prognosis is an expression that relates to the possibility of the patient surviving in general, but it may also be used to describe the severity of the sickness as it will present itself in the patient's future timeline. Methodology. The proposed technique consists of three stages: input data acquisition, preprocessing, and classification. Data acquisition consists of input raw data which is followed by preprocessing to eliminate the missed data and the classification is carried out using ensemble classifier to analyze the stages of cancer. This study explored the combined influence of the prominent labels in conjunction with one another utilizing the multilabel classifier approach, which is successful. Finally, an ensemble classifier model has been constructed and experimentally validated to increase the accuracy of the classifier model, which has been previously shown. The entire performance of the recommended and tested models demonstrates a steady development of 2% to 6% over the baseline presentation on the baseline performance. Results Providing a good contribution to the general health welfare of noncommercial potential workers in the healthcare sector is an opportunity provided by this recommended job outcome. It is anticipated that alternative solutions to these constraints, as well as automation of the whole process flow of all five phases, will be the key focus of the work to be carried out shortly. Predicting health status of employee in industry or information trends is made easier by these data patterns. The proposed classifier achieves the accuracy rate of 93.265%.

Human Activity Recognition using CNN and Pretrained Machine Learning Models

People's lives are enriched by human activity recognition (HAR), which extracts action-level details about human behaviour from raw input data. There are a variety of uses for Human Activity Recognition, including elderly surveillance systems, abnormal behaviour, and so forth. However, deep learning models such as Convolutional Neural Networks have outperformed traditional machine learning methods. As a result of CNN, it is possible to extract features and reduce computational costs. Transfer Learning, on the other hand, refers to the use of pre-trained machine learning models that can be used to detect human activity using a special type of Artificial Neural Network known as Leveraging CNN. Resnet-34 Use of a CNN model can provide detection accuracy up to 96.95 percent for human activity recognition Numerous studies and research have been conducted on HAR. There are only a few models in the majority of the paper, however What we know is that the more data we have, the better the model and the more accurate the model will be.

Forewarn Traffic Signal with Ambulance Detection By GPS and Light Standard Signal Using Machine Learning

People's lives are enriched by human activity recognition (HAR), which extracts action-level details about human behaviour from raw input data. There are a variety of uses for Human Activity Recognition, including elderly surveillance systems, abnormal behaviour, and so forth. However, deep learning models such as Convolutional Neural Networks have outperformed traditional machine learning methods. As a result of CNN, it is possible to extract features and reduce computational costs. Transfer Learning, on the other hand, refers to the use of pre-trained machine learning models that can be used to detect human activity using a special type of Artificial Neural Network known as Leveraging CNN. Resnet-34 Use of a CNN model can provide detection accuracy up to 96.95 percent for human activity recognition Numerous studies and research have been conducted on HAR. There are only a few models in the majority of the paper, however What we know is that the more data we have, the better the model and the more accurate the model will be.

Combining Open Source and Commercial Tools in Digital Twin for Cities Generation

Evidently, Smart Cities are on the rise and there is an increasing need for digital twins of these complex environments and their corresponding 3D models. The creation and maintenance of such twins is a time consuming task, since cities are living evolving organisms. In this paper we are presenting work done within the Digital Twin Cities Centre (DTCC) in Sweden in the field of automated 3D city model generation. We showcase a novel method of combining open source and commercial software for creating digital twins of any urban context in a procedural way from raw input data and using Unreal Engine as a visualization front-end. We combine two different workflows, one based on the commercial software suite Feature Manipulation Engine and in parallel we utilize the open source code developed within the Centre, dubbed DTCC Builder. The assets created can be used in urban planning, multi-physics continuum mechanics simulations and for visualization and illustration on a 3D scale. By using, in a complementary way, established commercial software and state of the art open source C++ code we manage to utilize the quality of life features the first provides, while assigning the demanding tasks of comforming mesh generation to the former. The longer term goal is (nearly) real-time mesh digital twinning of any city and user interaction with the 3D assets provided.

Soft Sensor Modeling Method Based on Target-Guided Related Feature Learning and Its Application

Currently, soft sensors have been extensively used in complex industrial processes. How to learn effective features from nonlinear process data is the core task of building precise soft sensors. Deep learning, a type of feature extraction approach, has shown unique advantages in many fields, as well as in soft sensor modeling. In this article, stacked autoencoder (SAE) is introduced for soft sensor modeling. However, most SAE-based methods focus on learning feature representations from raw input data, as well as ignore the impact of redundant feature information on model performance during the hierarchical pre-training stage. To this end, a novel target-guided related feature learning (TRFL) method is proposed to address the aforementioned two issues. First, an improved grey relational analysis (GRA) method is utilized to evaluate the importance of input variable for each autoencoder (AE), and a threshold determination mechanism is developed to remove the unrelated features. Later, in the decoding stage of each AE, the target variable and input layer variables are simultaneously reconstructed at the output layer, and a dynamic trade-off factor is defined to balance the dimension ratio change of the input layer variable with the target variable. In this way, target-related features can be sufficiently extracted by TRFL. Finally, the effectiveness and superiority of the proposed TRFL are validated with a numerical case and the prediction of two key parameters in the blast furnace ironmaking process.

Unsupervised Remaining Useful Life Prediction through Long Range Health Index Estimation based on Encoders-Decoders

The prediction of the Remaining Useful Life (RUL) is a critical step in Prognostics and Health Management (PHM) of systems under degradation. For efficient RUL predictions, most of the Artificial Intelligence (AI-based) methods perform direct mapping between raw sensor data input and RUL data as output targets for supervised learning. However, in the majority of the real-life cases, the available data are either incomplete or unlabeled, which calls for unsupervised methods. This paper proposes such an unsupervised RUL prediction method. Firstly, this method uses an autoencoder model to extract a Virtual Health Index (VHI) from sensors readings. Secondly, an LSTM-based (Long Short-Term Memory) encoder-decoder achieves VHI future predictions. Once the VHI prediction exceeds a pre-determined threshold, the RUL is recursively inferred. Such a method thus allows to obtain RUL predictions without using RUL-labeled data. This method is tested on C-MAPSS dataset. The results obtained are encouraging and offer new perspectives for real industrial applications

A Hybrid ConvLSTM-Based Anomaly Detection Approach for Combating Energy Theft

In a conventional power grid, energy theft is difficult to detect due to limited communication and data transition. The smart meter along with big data mining technology leads to significant technological innovation in the field of energy theft detection. This paper proposes a convolutional long short-term memory (ConvLSTM) based energy theft detection (ETD) model to identify electricity theft users. In this work, electricity consumption data is reshaped quarterly into a two-dimensional matrix and used as the sequential input to the ConvLSTM. The convolutional neural network (CNN) embedded into the long short-term memory (LSTM) can better learn the features of the data on different quarters, months, weeks, and days. Besides, the proposed model incorporates batch normalization. This technique allows the proposed ETD model to support raw format electricity consumption data input, reducing training time and increasing the efficiency of model deployment. The result of the case study shows that the proposed ConvLSTM model exhibits good robustness. It outperforms the multilayer perceptron (MLP) and CNN-LSTM in terms of performance metrics and model generalization capability. Moreover, the result also demonstrates that K-fold cross-validation can improve the ETD prediction accuracy.

Universal Adversarial Attacks on the Raw Data From a Frequency Modulated Continuous Wave Radar

As more and more applications rely on Artificial Intelligence (AI), it is inevitable to explore the associated safety and security risks, especially for sensitive applications where physical integrity is at risk. One of the most interesting challenges that come with AI are adversarial attacks being a well-researched problem in the visual domain, where a small change in the input data can cause the Neural Network (NN) to make an incorrect prediction. In the radar domain, AI is not that widespread yet but the results that AI applications produce are very promising, which is why more and more applications based on it are being used. This work presents three possible attack methods that are particularly suitable for the radar domain. The developed algorithms generate universal adversarial attack patches for all sorts of radar applications based on NN. The main goal of the algorithms, apart from the computation of universal patches, is the identification of sensitive areas in the raw radar data input which than can be examined more closely. To the best of our knowledge, this is the first work that deals with calculating universal patches on raw radar data, which is of great importance especially for interference analysis. The developed algorithms have been verified on two data sets. One in the field of autonomous driving where the attacks lead to a steering misprediction of up to 0.3 for the steering value which is within [-1,1], with the results also being successfully tested on a demonstrator. The other data set originated from a gesture recognition task, where the attacks decreased the accuracy, originally at 97.0% up to a minimum of 16.5%, which is slightly above 12.5% being the accuracy for a purely random prediction.

Spatial Data Application on Health (COVID-19) in Health Region 6 of Thailand

Purpose: A geographic information system (GIS) in a combination with health data is very useful for disease monitoring, prevention and control. The new sciences of the combination of GIS and classical health data, namely spatial epidemiology. Methods: Using the secondary health database (including COVID-19 data); data source is the 43 standard data folders of Thailand health statistic collected by Ministry of Public Health Thailand, collected between 2013 and 2022. The data of the Heath Region 6 of Thailand was used as an example. Twodata sets health data with spatial data Combined to new database and developed dashboard to present the information via web base system. Results: The dash board provides new perspective of disease distribution view. For example, the map on our dashboard reports the density of COVID-19 cases in each area. Based on the records, the densest areas shown are the urban area in each province. It can also be used for resource distribution and access time to health center of the area. Conclusion: The project reports the use of GIS and public health data to develop a dashboard for monitoring health resources and disease distributions in the health region 6 of Thailand. The major limitation of spatial epidemiology is the lack of or incomplete raw data input in the system.

Learning Text-image Joint Embedding for Efficient Cross-modal Retrieval with Deep Feature Engineering

This article introduces a two-phase deep feature engineering framework for efficient learning of semantics enhanced joint embedding, which clearly separates the deep feature engineering in data preprocessing from training the text-image joint embedding model. We use the Recipe1M dataset for the technical description and empirical validation. In preprocessing, we perform deep feature engineering by combining deep feature engineering with semantic context features derived from raw text-image input data. We leverage LSTM to identify key terms, deep NLP models from the BERT family, TextRank, or TF-IDF to produce ranking scores for key terms before generating the vector representation for each key term by using Word2vec. We leverage Wide ResNet50 and Word2vec to extract and encode the image category semantics of food images to help semantic alignment of the learned recipe and image embeddings in the joint latent space. In joint embedding learning, we perform deep feature engineering by optimizing the batch-hard triplet loss function with soft-margin and double negative sampling, taking into account also the category-based alignment loss and discriminator-based alignment loss. Extensive experiments demonstrate that our SEJE approach with deep feature engineering significantly outperforms the state-of-the-art approaches.