Context Reasoning Technologies in Ubiquitous Computing Environment

Context Aware Systems (CAS) support mobility of users and make context reasoning efficiently to provide relevant and real-time information. CAS should be particularly adaptive and aware by managing and responding properly to occurring events. Due mainly to the presence of numerous sensors and access to several communication networks, Context-awareness has been successfully included in the mobile device applications such as mobile and tablets. According to this fact, CAS are applied in many areas such as healthcare, transportation, education and etc. However, a reference descriptive model with respect to the capability of reusability and real-time processing and delivering the efficient service has not been considered yet. In other words, it's necessary to deploy a robust and multipurpose architecture which supports all requirements of CAS such as on demand processes and real time interactions. In this paper, the main objective is to design general purpose architecture for context aware systems which covers the CAS requirements and facilitates development of context-aware applications. Context-aware processing is a natural result of such architecture which enables the integration of context reasoning in event based systems. This architecture can be used in any environment such as healthcare, transportation, education and etc. In this paper, first we describe our architecture, related components and relationships between these components. Then, the some parts of the architecture are applied and implemented for a healthcare prototype in the hospital or home domain.

Most smartphones today have a rich set of sensors that could be used to infer input (e.g., accelerometer, gyroscope, microphone); however, the primary mode of interaction is still limited to the front-facing touchscreen and several physical buttons on the case. To investigate the potential opportunities for interactions supported by built-in sensors, we present the implementation and evaluation of BeyondTouch, a family of interactions to extend and enrich the input experience of a smartphone. Using only existing sensing capabilities on a commodity smartphone, we offer the user a wide variety of additional inputs on the case and the surface adjacent to the smartphone. Although most of these interactions are implemented with machine learning methods, compact and robust rule-based detection methods can also be applied for recognizing some interactions by analyzing physical characteristics of tapping events on the phone. This article is an extended version of Zhang et al. [2015], which solely covered gestures implemented by machine learning methods. We extended our previous work by adding gestures implemented with rule-based methods, which works well with different users across devices without collecting any training data. We outline the implementation of both machine learning and rule-based methods for these interaction techniques and demonstrate empirical evidence of their effectiveness and usability. We also discuss the practicality of BeyondTouch for a variety of application scenarios and compare the two different implementation methods.

Color demosaicing and digital zooming are common processes in digital cameras and they often employ similar interpolation concepts based on the information extracted from the raw sensor data. Realizing them independently is not efficient as separate extraction processes are required. It may also cause inconsistent utilization of the raw sensor data in different stages. This paper presents a low-complexity combined algorithm which directly extracts edge information from raw sensor data and exploits it consistently and efficiently in both demosaicing and zooming. The proposed algorithm can produce zoomed full-color images and zoomed CFA images with outstanding performance as compared with conventional approaches.

Due to the limitation of industrial conditions in production, raw sensor data are always shown as an unbalanced dataset, characterized by abundant normal operational data and scarce fault instances. This unbalance can degrade the performance of conventional fault diagnosis methods, leading to reduced accuracy and unstable model training. To address this challenge in bearing fault diagnosis, this paper proposes a Multi-Fault Feature Wasserstein Generative Adversarial Network (MFF-WGAN) to enhance diagnostic precision. First, the framework employs a multi-encoder denoising autoencoder architecture to mitigate noise interference in raw sensor data. Subsequently, the proposed MFF-WGAN integrates label information into its adversarial loss function to enable simultaneous generation of diverse fault categories, while incorporating inter-class feature discrepancies to refine sample quality. Finally, the developed multi-fault feature Wasserstein generation adversarial network is tested on the Case Western Reserve University bearing dataset and the laboratory bearing dataset. Computational results show that the proposed method can generate high-quality bearing samples with multiple faults effectively, which can obtain a higher diagnosis accuracy of 99.01% and 97.71% compared with the existing methods.

Prognostic health management (PHM) systems are designed to predict impending faults and to determine remaining useful life of machinery. An efficient prognostic system can speed up fault diagnosis by providing an indication of what parts of the machinery or vehicle are most likely to fail and will need maintenance in the near future. In this paper, we discuss the essential steps involved in building an effective PHM system. We describe time and frequency domain features that can be extracted from raw sensor data. These features or condition indicators can help summarize the information in raw data and extract critical clues that reflect the health of the machinery. Analytical models can then be used to learn the essential health indicators and how they relate to fault conditions. In addition, we describe a case study of implementing a PHM system for a high speed face milling CNC cutter. We describe features that were analyzed from sensor data. For the analytical engine, we used a Neural Network model for learning the association of the extracted features and the magnitude of wear in the cutter. The neural network was able to determine remaining useful life of cutters in terms of number of remaining cuts for a given wear limit based on extracted features.

Prognostics and health management (PHM) is an important part of ensuring reliable operations of complex safety- critical systems. System-level remaining useful life (RUL) estimation is a much more complex problem than making estimations at the component level, and system-level RUL methodologies remain sparse in the literature. Model-based approaches have traditionally worked in the past for components such as capacitors, MOSFETs, batteries, or hard-drives (to name a few examples), but developing high fidelity dynamics models of cyber physical systems that can be used to study the effects of multiple degrading components in the system remains a challenging task. Some initial work on model-based System RUL predictions was demonstrated in Khorasgani, et al [1], but, to generalize the system-level prognostics problem, we have to resort to pure data driven and hybrid approaches. In this work, we propose an end-to-end data- driven framework for developing deep learning models to predict remaining useful life of cyber physical systems operating under unknown faulty conditions. The raw data is organized with a data schema that improves the model development process anddown stream data analysis tasks. Due to the unknown faulty conditions, the raw sensor data is transformed into signals that expose the underlying degradation processes, which are then used for model development. Bayesian Optimization is used to tune the model parameters prior to training and validation. We show that this approach results in accurate predictions within 3 cycles to end of life (EOL). We demonstrate the effectiveness of our approach by applying it to the N-CMAPSS turbofan engine dataset recently released by NASA, which includes high fidelity degradation modeling, real world operating conditions, and a large set of fault operating modes.

Internet of Things (IoT) is a platform which gives the billions of computing devices, sensory devices that constantly produce and exchange the huge amounts of data over the network. Several interesting applications, like Greenhouse, involving extracting the higher level information from the raw data and representing it in human readable format, exists. An effective mechanism is required to process streaming data and inferring it, to get insight about the data and get some actionable information from processing and measurements. The aim is to represent the data from device point of view to user-centric point of view using Data analytics and Machine learning mechanism. Raw sensor data is created, based on the sensor threshold values. Parameters considered are temperature, humidity, co2 concentration, luminosity, radiation, soil moisture and soil temperature. Numerical values are converted to the strings to create higher level abstraction by applying set of rules. Abstracted data is cleaned by some cleaning processes. Then Latent Dirichlet Allocation (LDA),a topic extraction model, is applied to extract the hidden correlation among the data

Human activity recognition with smartphone sensors using deep learning neural networks

Recently, context aware has become a popular topic in ubiquitous computing. Most context-aware research combines embedded system devices with context-aware systems to realize a ubiquitous environment. This study proposes a Daily Exercise Context Aware Ontology Model (DEOCAM) and builds a Context Aware Exercise System (CAES), which is free of space limitation. CAES is composed of three parts: Personal Exercise Management System (PEMS), Context Aware System (CAS) and Context Sharing Network Platform (CSNP). The purpose of the system is to recognize the daily activities of users, which is helpful in exercise monitoring for health management. Our experimental results show that the system is effective and feasible for daily activity contexts recognition and moving distance computation.

Smart sensor system design requires intelligent data processing, which analyzes raw time-series sensor data to efficiently and precisely discriminate and quantify target gases. The work presented here utilizes the response of twin gas sensor arrays for gases such as ethanol, ethylene, methane, and carbon monoxide to discriminate and quantify the target gases. We propose a 3D convolution neural-based regression dual network (3D-CNRDN) for both gas quantification and discrimination. The spatiotemporal correlation of sensor array responses inspired us to design the deep neural network for the gas concentration estimation model. The sensor array set is spatially correlated, and all the twin array responses are temporally related. 3D-CNRDN uses raw time-series gas sensor array data. The data is fed to the network as the 3D pattern, which contains 2D spatial information varying with third dimension time to recognize patterns that eventually predict the concentration. The model evaluation shows that the proposed methods are an effective technique for gas quantification and identification with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{RMSE}=0.3179$</tex> and classification accuracy 94.37%. Furthermore, the proposed method outperforms and provides higher discrimination accuracy and lower RMSE than other machine learning and deep learning methods.

The wealth of sensor data generated by advanced vehicular sensors that are fitted in new, connected vehicles enables new applications for driver behavior, road health monitoring, and incident reporting. However, standard access mechanisms to the data restricts the services and insights that can be provided by vehicular applications. For those applications to have full access to all of the vehicular sensory data, custom hardware fitted with proprietary automaker software is needed to process raw sensor values. This limits rapid deployment at scale because of the time and costs needed for mass development across proprietary platforms. In this paper, we propose a system to provide access to the raw sensory data using the vehicle's identification number in order to retrieve the vehicles sensors identification numbers, their description, and their related software libraries that house the data processing algorithms specific to the vehicle's make and model. Smartphones can collect raw sensory data through the vehicle's CAN-Bus interface, and use those software libraries to transform raw data into standard formats that can be used by vehicular applications. This way, many applications can be developed without having to worry about customization of hardware to process the data produced by each automaker's sensor platforms.

To determine the position and navigation of an unknown environment, UAVs rely on sensors that provide information regarding position, speed, and orientation. There are sensors to provide direct navigation information such as the Global Navigation Satellite System (GNSS) by providing position data, or indirect sensors such as inertial sensors which provide speed and orientation data. An inertial sensor or commonly known as an Inertial Measurement Unit (IMU) is a combination of data acceleration (accelerometer) and angular velocity (gyroscope). By performing GNSS/IMU sensor fusion at UAV Quadrotor will increase the accuracy of aircraft localization based on its mathematical model involving the Kalman Filter approach. The main goal is to improve the coordinates obtained from Quadrotor UAV measurements, so that position of UAV Quadrotor aircraft is more accurate. Raw data of sensors GNSS/IMU is obtained during the flight of the aircraft. Visual comparison is used to determine whether the coordinate of the processed data has better accuracy than the raw data. The results showed that the Unscented Kalman Filter (UKF) simulation gave 3D position accuracy of 0.403 m to the measurement data. It can improve 23,47% fprm EKF Estimation which give 3D position accuracy of 16.598 m.

The evolution of industry Artificial intelligence (AI) technology in today’s manufacturing system has the potential to be extremely helpful in applications such as quality assurance, process optimization, and predictive maintenance. Particularly, predictive maintenance in manufacturing uses data to forecast potential failures and optimize maintenance schedules, leading to increased efficiency, productivity, and sustainability. Industrial AI is an interdisciplinary area of research, encompassing fields such as Machine Learning (ML), natural language processing and robotics. Association rule learning, a rule-based ML method for discovering interesting relationships between variables in large databases, has been widely used across business intelligent applications for decision-making. However, there are few studies on predictive maintenance in manufacturing systems. This paper proposes Apriori-Based learning in R for providing the predictive analytics to improve fault detection and support predictive maintenance, thereby enhancing production stability and efficiency. The case study is conducted to illustrate the feasibility of the proposed method is demonstrated.

IntroductionWearables with proprietary scoring protocols are typically used to assess sleep and physical activity, but the field is shifting to wearables with raw sensor data accessible and open-source scoring methods to enhance rigor and reproducibility. The data infrastructure to process raw sensor data in clinical research is underdeveloped; we therefore developed the Modular Actigraphy Platform (MAP).MethodsMAP is a cloud-based computational platform that processes high-resolution time series sensor data to derive sleep and physical activity metrics. It was engineered to be modular, providing flexibility in data processing and enabling the seamless integration of open-source sleep and physical activity scoring methodologies as they become available (currently, GGIR and MIMS processing algorithms have been integrated). A structured Software Development Life Cycle (SDLC) approach was used to guide the development of MAP, with a multi-level testing framework consisting of unit testing (verification of modules), integration testing (interaction among modules), and system testing (validating specifications). Following these foundational tests, we then completed user acceptance testing in two phases - alpha (17 test files) and beta (686 files from 4 pediatric cohorts) to assess processing performance.ResultsFor beta testing, MAP leveraged up to 60 CPU cores and 500 GiB of memory. The pre-processing module was the most computationally demanding and was more efficient in MAP compared to offline processing (up to 8 CPU cores and 23.2 GiB of memory). For example, the preprocessing GGIR part 1 container was completed at a speed of 0.29–0.49 minutes per file (1.6–2.9 times faster than offline processing) and the pre-processing MIMS container was completed at a speed of 0.49–4.66 minutes per file (2.4 to 14.0 times faster than offline processing).ConclusionMAP is an efficient computational platform that integrates open-source scoring algorithms to efficiently process raw sensor data for wearable sleep and physical activity estimation in clinical research and is available through the Children’s Hospital of Philadelphia’s Research Institute.

In this paper we propose the method that detects moving objects in autonomous navigation vehicle using LRF sensor data. Object detection and tracking methods are widely used in research area like safe-driving, safe-navigation of the autonomous vehicle. The proposed method consists of three steps: data segmentation, mobility classification and object tracking. In order to make the raw LRF sensor data to be useful, Occupancy grid is generated and the raw data is segmented according to its appearance. For classifying whether the object is moving or static, trajectory patterns are analysed. As the last step, Markov chain Monte Carlo (MCMC) method is used for tracking the object. Experimental results indicate that the proposed method can accurately detect moving objects.