Raw Sensor Data Research Articles

A Generalized Autonomous Power Plant Fault Detection Model Using Deep Feature Extraction and Ensemble Machine Learning

Ensuring operational reliability and efficiency in steam power plants requires advanced and generalized fault detection methodologies capable of addressing diverse fault scenarios in boiler and turbine systems. This study presents an autonomous fault detection framework that integrates deep feature extraction through Convolutional Autoencoders (CAEs) with the ensemble machine learning technique, Extreme Gradient Boosting (XGBoost). CAEs autonomously extract meaningful and nonlinear features from raw sensor data, eliminating the need for manual feature engineering. Principal Component Analysis (PCA) is employed for dimensionality reduction, enhancing computational efficiency while retaining critical fault-related information. The refined features are then classified using XGBoost, a robust ensemble learning algorithm, ensuring accurate fault detection. The proposed model is validated through real-world case studies on boiler waterwall tube leakage and motor-driven oil pump failure in steam turbines. Results demonstrate the framework’s ability to generalize across diverse fault types, detect anomalies at an early stage, and minimize operational downtime. This study highlights the transformative potential of combining deep feature extraction and ensemble machine learning for scalable, reliable, and efficient fault detection in power plant operations.

Trajectory Aware Deep Reinforcement Learning Navigation Using Multichannel Cost Maps

Deep reinforcement learning (DRL)-based navigation in an environment with dynamic obstacles is a challenging task due to the partially observable nature of the problem. While DRL algorithms are built around the Markov property (assumption that all the necessary information for making a decision is contained in a single observation of the current state) for structuring the learning process; the partially observable Markov property in the DRL navigation problem is significantly amplified when dealing with dynamic obstacles. A single observation or measurement of the environment is often insufficient for capturing the dynamic behavior of obstacles, thereby hindering the agent’s decision-making. This study addresses this challenge by using an environment-specific heuristic approach to augment the dynamic obstacles’ temporal information in observation to guide the agent’s decision-making. We proposed Multichannel Cost Map Observation for Spatial and Temporal Information (M-COST) to mitigate these limitations. Our results show that the M-COST approach more than doubles the convergence rate in concentrated tunnel situations, where successful navigation is only possible if the agent learns to avoid dynamic obstacles. Additionally, navigation efficiency improved by 35% in tunnel scenarios and by 12% in dense-environment navigation compared to standard methods that rely on raw sensor data or frame stacking.

LSTM‐based real‐time stress detection using PPG signals on raspberry Pi

AbstractStress, widely recognised for its profound adverse effects on both physical and mental health, necessitates the development of innovative real‐time detection methods. In this context, the escalating prevalence of wearable embedded systems, integrated with artificial intelligence (AI) for the continuous monitoring of critical physiological indicators like heart rate and blood pressure, accentuates their growing relevance in the efficient detection of stress. This article presents an innovative methodology employing deep learning algorithms on the Raspberry Pi 3, a platform distinguished by its cost‐effectiveness and limited resources. The authors have developed an advanced AI algorithm that achieves high accuracy in real‐time stress detection using photoplethysmography (PPG) sensors while significantly reducing computational demands. The authors’ method utilises an artificial neural network with long short‐term memory (LSTM) layers, proving highly effective in time‐series data analysis. In this study, the authors implement key TensorFlow toolkit optimisation techniques including quantisation aware training (QAT), Pruning and prune‐preserving quantisation aware training. These techniques are applied to refine the authors’ model, decreasing size and latency without sacrificing accuracy. The results highlight the LSTM‐based model's proficiency in accurately detecting stress using raw PPG sensor data on the Raspberry Pi 3, a comparatively affordable platform. The model attains an accuracy of 89.32% and an F1 score of 89.55% on the diverse wearable stress and affect detection stress‐level dataset. Additionally, the authors’ optimised model exhibits substantial reductions in both size and latency while maintaining high accuracy. This approach shows great potential for various applications, such as stress monitoring in healthcare, sports, and workplace settings. The use of the Raspberry Pi 3 makes the system portable, cost‐effective, and energy‐efficient, enhancing its potential impact and accessibility.

Encoding human activities using multimodal wearable sensory data

Human Activity Recognition (HAR) is the task to automatically analyze and recognize human body gestures or actions. HAR using time-series multi-modal sensory data is a challenging and important task in the field of machine learning and feature engineering due to its increasing demands in numerous real-world applications such as healthcare, sports and surveillance. Numerous daily wearable devices e.g., smartphones, smartwatches, and smart glasses can be used to collect and analyze the human activities on an unprecedented scale. This paper presents a generic framework to recognize the different human activities using continuous time-series multimodal sensory data of these smart gadgets. The proposed framework follows the channel of Bag-of-Features which consists of four steps: (i) Data acquisition and pre-processing, (ii) codebook computation, (iii) feature encoding, and (iv) classification. Each step in the framework plays a significant role to generate an appropriate feature representation of raw sensory data for efficient activity recognition. In the first step, we employed a simple overlapped-window sampling approach to segment the continuous time-series sensory data to make it suitable for activity recognition. Secondly, we build a codebook using k-means clustering algorithm to group the similar sub-sequences. The center of each group is known as codeword and we assume that it represents a specific movement in the activity sequence. The third step consists of feature encoding which transform the raw sensory data of activity sequence into its respective high-level representation for the classification. Specifically, we presented three reconstruction-based encoding techniques to encode sensory data, namely: Sparse Coding, Local Coordinate Coding, and Locality-constrained Linear Coding. The segmented activity sub-sequences are transformed to high-level representation using these techniques and earlier computed codebook. Finally, the encoded features are classified using a simple Random Forest classifier. The proposed HAR framework using three different encoding techniques is evaluated on three large benchmark datasets: UniMiB-SHAR dataset, MHEALTH dataset and WISDM dataset and their results are compared with most recent state-of-the-art techniques. It outperforms the existing techniques and achieves 98.39%, 99.5%, and 99.4% recognition scores on UniMiB-SHAR dataset, MHEALTH dataset, and WISDM dataset respectively. The excellent recognition results and computational analysis confirms the effectiveness and efficiency of the proposed framework.

Diagnosis of Schizophrenia Using EEG Sensor Data: A Novel Approach with Automated Log Energy-Based Empirical Wavelet Reconstruction and Cepstral Features.

Schizophrenia (SZ) is a severe mental disorder characterised by disruptions in cognition, behaviour, and perception, significantly impacting an individual's life. Traditional SZ diagnosis methods are labour-intensive and prone to errors. This study presents an innovative automated approach for detecting SZ acquired through electroencephalogram (EEG) sensor signals, aiming to improve diagnostic efficiency and accuracy. We utilised Fast Independent Component Analysis to remove artefacts from raw EEG sensor data. A novel Automated Log Energy-based Empirical Wavelet Reconstruction (ALEEWR) technique was introduced to reconstruct decomposed modes based on their variability, ensuring effective extraction of meaningful EEG signatures. Cepstral-based features-cepstral activity, cepstral mobility, and cepstral complexity-were used to capture the power, rate of change, and irregularity of the cepstrum of preprocessed EEG signals. ANOVA-based feature selection was applied to refine these features before classification using the K-Nearest Neighbour (KNN) algorithm. Our approach achieved an exceptional accuracy of 99.4%, significantly surpassing previous methods. The proposed ALEEWR and cepstral analysis demonstrated high precision, sensitivity, and specificity in the automated diagnosis of schizophrenia. This study introduces a highly accurate and efficient method for SZ detection using EEG technology. The proposed techniques offer significant improvements in diagnostic accuracy, with potential implications for enhancing SZ diagnosis and patient care through automated systems.

Recognizing human activities using light-weight and effective machine learning methodologies

Background Human activity recognition poses a complex challenge in predicting individuals’ movements from raw sensor data using machine learning models. This paper explores the application of six prominent machine learning techniques – decision tree, random forest, linear regression, Naïve Bayes, k-nearest neighbor, and neural networks – to enhance the accuracy of human activity detection for e-health systems. Despite previous research efforts employing data mining and machine learning, there remains room for improvement in performance. The study focuses on predicting activities such as walking, standing, laying, sitting, walking upstairs, and walking downstairs. Methods The research employs six machine learning algorithms to recognize human activities, including decision tree, random forest, linear regression, Naïve Bayes, k-nearest neighbor, and neural networks. Results Evaluation of the human activity recognition dataset reveals that the random forest classifier, CNN, GRN and neural network yield promising results, achieving high accuracy. However, Naïve Bayes falls short of satisfying outcomes. Conclusions The study successfully classifies activities like SITTING, STANDING, LAYING, WALKING, WALKING_DOWNSTAIRS, and WALKING_UPSTAIRS with a remarkable accuracy of 98%. The contribution lies in the thorough exploration of machine learning techniques, with neural networks emerging as the most effective in enhancing human activity recognition. The findings showcase the potential for advanced applications in e-health systems and beyond.

An Optimal Feature Selection Method for Human Activity Recognition Using Multimodal Sensory Data

Recently, the research community has taken great interest in human activity recognition (HAR) due to its wide range of applications in different fields of life, including medicine, security, and gaming. The use of sensory data for HAR systems is most common because the sensory data are collected from a person’s wearable device sensors, thus overcoming the privacy issues being faced in data collection through video cameras. Numerous systems have been proposed to recognize some common activities of daily living (ADLs) using different machine learning, image processing, and deep learning techniques. However, the existing techniques are computationally expensive, limited to recognizing short-term activities, or require large datasets for training purposes. Since an ADL is made up of a sequence of smaller actions, recognizing them directly from raw sensory data is challenging. In this paper, we present a computationally efficient two-level hierarchical framework for recognizing long-term (composite) activities, which does not require a very large dataset for training purposes. First, the short-term (atomic) activities are recognized from raw sensory data, and the probabilistic atomic score of each atomic activity is calculated relative to the composite activities. In the second step, the optimal features are selected based on atomic scores for each composite activity and passed to the two classification algorithms: random forest (RF) and support vector machine (SVM) due to their well-documented effectiveness for human activity recognition. The proposed method was evaluated on the publicly available CogAge dataset that contains 890 instances of 7 composite and 9700 instances of 61 atomic activities. The data were collected from eight sensors of three wearable devices: a smartphone, a smartwatch, and smart glasses. The proposed method achieved the accuracy of 96.61% and 94.1% by random forest and SVM classifiers, respectively, which shows a remarkable increase in the classification accuracy of existing HAR systems for this dataset.

Evaluating Large Language Models as Virtual Annotators for Time-series Physical Sensing Data

Traditional human-in-the-loop-based annotation for time-series data like inertial data often requires access to alternate modalities like video or audio from the environment. These alternate sources provide the necessary information to the human annotator, as the raw numeric data is often too obfuscated even for an expert. However, this traditional approach has many concerns surrounding overall cost, efficiency, storage of additional modalities, time, scalability, and privacy. Interestingly, recent large language models (LLMs) are also trained with vast amounts of publicly available alphanumeric data, which allows them to comprehend and perform well on tasks beyond natural language processing. Naturally, this opens up a potential avenue to explore the opportunities in using these LLMs as virtual annotators where the LLMs will be directly provided the raw sensor data for annotation instead of relying on any alternate modality. Naturally, this could mitigate the problems of the traditional human-in-the-loop approach. Motivated by this observation, we perform a detailed study in this paper to assess whether the state-of-the-art (SOTA) LLMs can be used as virtual annotators for labeling time-series physical sensing data. To perform this in a principled manner, we segregate the study into two major phases. In the first phase, we investigate the challenges an LLM like GPT-4 faces in comprehending raw sensor data. Considering the observations from phase 1, in the next phase, we investigate the possibility of encoding the raw sensor data using SOTA SSL approaches and utilizing the projected time-series data to get annotations from the LLM. Detailed evaluation with four benchmark HAR datasets shows that SSL-based encoding and metric-based guidance allow the LLM to make more reasonable decisions and provide accurate annotations without requiring computationally expensive fine-tuning or sophisticated prompt engineering.

A novel hybrid deep learning approach with GWO–WOA optimization technique for human activity recognition

The effectiveness of Human Activity Recognition (HAR) models can be largely attributed to the components derived from domain expertise. The classification system swiftly and effectively categorizes human physical activity by utilizing a comprehensive collection of variables. To construct HAR models and categorize different activities, deep learning algorithms have recently seen increased application in academic research for autonomously extracting features from raw sensory data. The primary goal of this study is to develop a strong and accurate activity classification model by first extracting features from an input dataset and then using deep learning models. In this research, we look at how well the model does in the field of HAR. Analyzing time-series data from smartphones and wearable sensors, algorithms based on deep learning (DL) have been used to forecast various human actions. Applying DL-based methods to time-series data presents a variety of challenges, notwithstanding their usage for activity identification. If the suggested approach is put into practice, the concerns mentioned before may be alleviated. To construct HAR classification techniques using data collected from wearable sensors, this study presents two Hybrid Learning Algorithms (HLA). This research employs Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) methods to teach the model sequential and spatial patterns. The process of selecting characteristics has been enhanced through the use of optimization techniques. Together, the Whale Optimization Algorithm and the Grey Wolf Optimizer form a Hybrid Optimization approach, which is introduced in this study. By combining and capitalizing the benefits of both approaches, the hybrid technique enhances the optimization process as a whole. This research may also assist individuals who have undergone amputations or experienced strokes in selecting an appropriate model and developing a personalized gait reference trajectory.

CLARA: A cluster-based node correlation for sampling rate adaptation and fault tolerance in sensor networks

Recently, wireless sensor networks (WSNs) have been proven as an efficient and low-cost solution for monitoring various kind of applications. However, the massive amount of data collected and transmitted by the sensor nodes, which are mostly redundant, will quickly consume their limited battery power, which is sometimes difficult to replace or recharge. Although the huge efforts made by researchers to solve such problem, most of the proposed techniques suffer from their accuracy and their complexity, which is not suitable for limited-resources sensors. Therefore, designing new data reduction techniques to reduce the raw data collected in such networks is becoming essential to increase their lifetime. In this paper, we propose a CLuster-based node correlation for sAmpling Rate adaptation and fAult tolerance, abbreviated CLARA, mechanism dedicated to periodic sensor network applications. Mainly, CLARA works on two stages: node correlation and fault tolerance. The first stage introduces a data clustering method that aims to search the correlation among neighboring nodes. Then, it accordingly adapts their sensing frequencies in a way to reduce the amount of data collected in such networks while preserving the information integrity at the sink. In the second stage, a fault tolerance model is proposed that allows the sink to regenerate the raw sensor data based on two methods: moving average (MA) and exponential smoothing (ES). We demonstrated the efficiency of our technique through both simulations and experiments. The best obtained results show that the first stage can reduce the sensor sampling rate, and accordingly the sensor energy, up to 64% while the second stage can accurately regenerate the raw data with an error loss less than 0.15.

The Design and Implementation of a Semantic-Based Proactive System for Raw Sensor Data: A Case Study for Laboratory Environments

Abstract— In the last decade, raw sensor data from sensor-based systems, the area of use of which has increased considerably, pose a fundamentally new set of research challenges, including structuring, sharing, and management. Although many different academic studies have been conducted on the integration of sets of data emerging from different sensor-based systems until present, these studies have generally focused on the integration of data as syntax. Studies on the semantic integration of data are limited, and still, the area of the study mentioned have problems that await solutions. In this article; parameters (CO2, TVOC, CO, PM2.5, PM10, Temperature, Humidity, Light), affecting laboratory analysis results and threatening the analyst's health, were measured in laboratory environments selected as “use cases”, and semantic-based information management framework was created for different sensor-based systems. Classical machine learning methods, and regression approaches which are frequently used for such sensor data, have been applied to the proposed sensor ontology and it has been measured that machine learning algorithm performs better on ontological sensor data. The most efficient algorithms in terms of accuracy and time were selected, and integrated into the proposed proactive approach, in order to take the selected laboratory environment’s condition under control.

Individual canopy tree species maps for the National Ecological Observatory Network.

The ecology of forest ecosystems depends on the composition of trees. Capturing fine-grained information on individual trees at broad scales provides a unique perspective on forest ecosystems, forest restoration, and responses to disturbance. Individual tree data at wide extents promises to increase the scale of forest analysis, biogeographic research, and ecosystem monitoring without losing details on individual species composition and abundance. Computer vision using deep neural networks can convert raw sensor data into predictions of individual canopy tree species through labeled data collected by field researchers. Using over 40,000 individual tree stems as training data, we create landscape-level species predictions for over 100 million individual trees across 24 sites in the National Ecological Observatory Network (NEON). Using hierarchical multi-temporal models fine-tuned for each geographic area, we produce open-source data available as 1 km2 shapefiles with individual tree species prediction, as well as crown location, crown area, and height of 81 canopy tree species. Site-specific models had an average performance of 79% accuracy covering an average of 6 species per site, ranging from 3 to 15 species per site. All predictions are openly archived and have been uploaded to Google Earth Engine to benefit the ecology community and overlay with other remote sensing assets. We outline the potential utility and limitations of these data in ecology and computer vision research, as well as strategies for improving predictions using targeted data sampling.

Periodic Energy Optimization Using IOT and ML

The rapid expansion of Internet of Things (IoT) applications across various sectors generates an enormous volume of continuous time-series data. However, transmitting this massive amount of sensor data from energy constrained IoT nodes poses a significant challenge. The continuous transmission of such data consumes vast amounts of energy.In this work, we present a solution to this problem by predicting the periodic behavior of sensor data through a higher-level view of continuous transmission data from nodes in IoT at server side. Our system is composed of an IoT sensor network and a data processing unit. The local sensor network: temperature and humidity data is collected using 4 different nodes, as well, which afterward this info is transferred into a data processing unit built on the Raspberry Pi device. We use the machine learning model Autoregressive Integrated Moving Average (ARIMA) on the processing unit. This model is then applied individually to the data from each of the four nodes, predicting processed sensor values in the future accurately. In short, after getting highly accurate prediction, then we settle down proper energy saving pattern which reduces the data transmission requirements hence results in energy saving pattern.By utilizing the predictive capabilities of the ARIMA model, we minimize the need for constant transmission of raw sensor data. Instead, we transmit only essential updates or deviations from the predicted values. This approach substantially reduces energy consumption by eliminating the transmission of redundant information. In summary, our project aims to overcome the energy limitations of IoT sensor nodes by leveraging predictive modelling techniques, specifically the ARIMA model. By accurately predicting periodic patterns in sensor data, we can optimize energy usage by transmitting only the necessary information, while still ensuring effective monitoring of temperature and humidity in the IoT network.

Humanactivity Recognition using Deep Learning

Human activity recognition (HAR) plays a crucial role in various fields, including healthcare, surveillance, and human-computer interaction. This study explores the application of deep learning techniques for accurate and efficient human activity recognition. Leveraging the capabilities of deep neural networks, specifically convolutional neural networks (CNNs) and recurrent neural networks (RNNs), the proposed approach aims to capture both spatial and temporal features from sensor data. The dataset utilized in this research comprises multi-model sensor inputs, such as accelerometer and gyroscope readings, collected from wearable devices. The deep learning model is designed to automatically learn hierarchical representations of the raw sensor data, enabling robust feature extraction and discrimination between different human activities. Transfer learning is employed to enhance model generalization across diverse activity categories and varying sensor setups. Experimental evaluations are conducted on real-world datasets, demonstrating the effectiveness of the proposed deep learning framework in accurately classifying activities, including walking, running, sitting, and standing. Comparative analyses against traditional machine learning methods underscore the superior performance and adaptability of deep learning in handling complex and dynamic activity patterns. The results showcase the potential of deploying deep learning models in real-time human activity recognition systems, highlighting their scalability and efficiency. The study contributes to the advancement of HAR methodologies, paving the way for the development of more reliable and robust systems in applications such as healthcare monitoring, assisted living, and smart environments.

Machine Learning Approaches for Fault Detection and Diagnosis in Electrical Machines: A Comparative Study of Deep Learning and Classical Methods

Fault detection and diagnosis in electrical machines are crucial for ensuring their safe and reliable operation. In recent years, machine learning techniques have emerged as powerful tools for addressing this challenge, offering the potential for more accurate and efficient fault detection and diagnosis compared to traditional methods. Among these techniques, deep learning has gained significant attention due to its ability to automatically learn relevant features from raw data. However, the performance of deep learning models in this domain has not been extensively compared to classical methods. This paper presents a comparative study of deep learning and classical methods for fault detection and diagnosis in electrical machines. The study evaluates the performance of various machine learning algorithms, including deep neural networks, support vector machines, decision trees, and ensemble methods, in detecting and diagnosing faults such as stator winding faults, rotor faults, and bearing faults. The experimental evaluation is conducted using real-world datasets obtained from electrical machines in industrial settings. Performance metrics such as accuracy, precision, recall, and F1-score are used to assess the effectiveness of each approach in detecting and diagnosing faults accurately and efficiently. The results of the study indicate that deep learning approaches, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), outperform classical methods in terms of fault detection and diagnosis accuracy. These deep learning models demonstrate the ability to automatically extract informative features from raw sensor data, enabling them to effectively identify subtle patterns indicative of faults. The study investigates the interpretability of deep learning models compared to classical methods, examining the extent to which the models can provide insights into the underlying causes of faults. While deep learning models typically operate as black boxes, techniques such as layer-wise relevance propagation (LRP) are employed to enhance their interpretability and facilitate the identification of relevant features contributing to fault detection and diagnosis. This comparative study provides valuable insights into the strengths and limitations of deep learning and classical methods for fault detection and diagnosis in electrical machines, offering guidance for practitioners and researchers in selecting appropriate approaches for their specific applications.

AIoT-enabled decentralized sensor fault diagnosis for structural health monitoring

Structural health monitoring (SHM) systems collect and analyze data recorded from civil infrastructure to detect changes in the structural condition and to predict potential damage. However, sensor faults in SHM systems may lead to erroneous structural assessments, thus to inefficient maintenance operations, which could jeopardize the integrity of the structures. Artificial intelligence (AI) algorithms based on analytical redundancy have proven being effective foundations to implement sensor fault diagnosis but are computationally expensive and require large amounts of raw sensor data to be transmitted to centralized servers. Moreover, operating with centralized servers increases the likelihood of single points of failure, which compromises the robustness of SHM systems. To overcome the limitations and to enhance the robustness of SHM systems, this paper proposes a decentralized approach towards fault-tolerant SHM systems based on the concept of Artificial Intelligence of Things (AIoT). Fault diagnosis algorithms based on AI are embedded into sensor nodes deployed for SHM, using Internet of Things technologies to be computed at the edge of civil infrastructure, avoiding large amounts of data to be sent to central servers. In addition, the sensor nodes embed algorithms that collect, process, and analyze the sensor data to assess the condition of the structures being monitored. The result is a smart (i.e., decentralized, fault-tolerant, and AIoT-based) SHM system that is calibrated in a laboratory experiment and then deployed on a bridge for field validation. It is expected that the approach presented in this paper will improve the robustness of SHM systems in assessing the structural condition of civil infrastructure.

A hybrid deep approach to recognizing student activity and monitoring health physique based on accelerometer data from smartphones

Smartphone sensors have gained considerable traction in Human Activity Recognition (HAR), drawing attention for their diverse applications. Accelerometer data monitoring holds promise in understanding students’ physical activities, fostering healthier lifestyles. This technology tracks exercise routines, sedentary behavior, and overall fitness levels, potentially encouraging better habits, preempting health issues, and bolstering students’ well-being. Traditionally, HAR involved analyzing signals linked to physical activities using handcrafted features. However, recent years have witnessed the integration of deep learning into HAR tasks, leveraging digital physiological signals from smartwatches and learning features automatically from raw sensory data. The Long Short-Term Memory (LSTM) network stands out as a potent algorithm for analyzing physiological signals, promising improved accuracy and scalability in automated signal analysis. In this article, we propose a feature analysis framework for recognizing student activity and monitoring health based on smartphone accelerometer data through an edge computing platform. Our objective is to boost HAR performance by accounting for the dynamic nature of human behavior. Nonetheless, the current LSTM network’s presetting of hidden units and initial learning rate relies on prior knowledge, potentially leading to suboptimal states. To counter this, we employ Bidirectional LSTM (BiLSTM), enhancing sequence processing models. Furthermore, Bayesian optimization aids in fine-tuning the BiLSTM model architecture. Through fivefold cross-validation on training and testing datasets, our model showcases a classification accuracy of 97.5% on the tested dataset. Moreover, edge computing offers real-time processing, reduced latency, enhanced privacy, bandwidth efficiency, offline capabilities, energy efficiency, personalization, and scalability. Extensive experimental results validate that our proposed approach surpasses state-of-the-art methodologies in recognizing human activities and monitoring health based on smartphone accelerometer data.

A Deep Learning Approach for Online State of Health Estimation of Lithium-Ion Batteries Using Partial Constant Current Charging Curves

The accurate state of health (SOH) estimation of lithium-ion batteries (LIBs) during operation is crucial to ensure optimal performance, prolonging battery life and preventing unexpected failure or safety hazards. This work presents a storage- and performance-optimised deep learning approach to estimate the capacity-based SOH of LIBs using raw sensor data from partial charging curves under constant current condition. The proposed model is based on a combination of a one-dimensional convolutional and long short-term memory neural network, and processes time, voltage, and incremental capacity of partial charging curves as time series. The model is cross-validated on different ageing scenarios, reaching an overall MAE = 0.418% and RMSE = 0.531%, promising an accurate SOH estimation of LIBs under varying usage and environmental conditions in a real-world application.

Class imbalance in multi-resident activity recognition: an evaluative study on explainability of deep learning approaches

AbstractRecognizing multiple residents’ activities is a pivotal domain within active and assisted living technologies, where the diversity of actions in a multi-occupant home poses a challenge due to their uneven distribution. Frequent activities contrast with those occurring sporadically, necessitating adept handling of class imbalance to ensure the integrity of activity recognition systems based on raw sensor data. While deep learning has proven its merit in identifying activities for solitary residents within balanced datasets, its application to multi-resident scenarios requires careful consideration. This study provides a comprehensive survey on the issue of class imbalance and explores the efficacy of Long Short-Term Memory and Bidirectional Long Short-Term Memory networks in discerning activities of multiple residents, considering both individual and aggregate labeling of actions. Through rigorous experimentation with data-level and algorithmic strategies to address class imbalances, this research scrutinizes the explicability of deep learning models, enhancing their transparency and reliability. Performance metrics are drawn from a series of evaluations on three distinct, highly imbalanced smart home datasets, offering insights into the models’ behavior and contributing to the advancement of trustworthy multi-resident activity recognition systems.

RL-NBV: A deep reinforcement learning based next-best-view method for unknown object reconstruction

The Next-Best-View (NBV) algorithm is a key component in autonomous unknown object reconstruction. It iteratively determines the optimal sensor pose to capture the maximum information about the object under reconstruction. However, prevailing deep reinforcement learning (DRL) based NBV algorithms tend to transform point cloud, the raw sensor data, into different representations, thereby obscuring natural invariances of the data. In this work, we propose an innovative DRL-based method, denoted as RL-NBV, to learn NBV policy directly from the raw point cloud data. Specifically, we interpret the observation space as the current state of the reconstructed object represented by point clouds and current view selection states. Experimental results indicate that our method outperforms existing methods in terms of reconstruction performance. Moreover, our method significantly improves efficiency over ray-casting-based algorithms as time-consuming ray casting and data transformation are unnecessary.