Smart4Health project is a European project aiming to empower citizens with electronic health(care) record exchange, personal connected health services, and the ability of data donorship to the scientific community. The Smart4Health platform will enable citizens to manage, collect, store, access and share own health and healthcare data, at international level, through an easy-to-use, secure, constantly accessible and portable health data and services prototype within the EU and beyond. This shall also comprise self-quantified and citizen-generated data through IoT and wearables (e.g. smart watches, smart devices/textiles/shoes). Therefore, the citizen will not only be able to access data produced in the context of health systems, but become important contributor of health data more generally speaking. The information to be collected will feed the Smart4Health platform (4HealthPlatform – 4HP), enabling the Smart4Health user portal (4HealthNavigator – 4HN) services and applications to provide advanced personalised health services accessible whenever and wherever. In this paper we explore the work being developed for data integration coming from different smart devices aiming at enriching the citizen health and personal data as well as providing insight about citizen behaviour and support on how to modify/adapt postures and habits that may contribute for better health and wellbeing.

Free-fall absolute gravimeters are important classical high precision absolute gravimeters in many branches of scientific research. But its performance is always troubled by the ground vibration. Vibration correction method is used to correct the result by detecting the ground vibration with sensors. A Kalman filter based fusion method is proposed to obtain more accurate ground vibration signal by fusing the outputs of the seismometer and the accelerometer. Experiment is conducted with the homemade T-1 absolute gravimeter, the standard deviation of the corrected results using seismometer data and fused data are 586.32 μGal (1 μGal = 10 −8 m/s 2 ) and 508.59 μGal respectively, much better than the uncorrected result’s 6548.96 μGal. The results prove the superiority of fused data over data measured from single sensor. It is believed that the application scene of the vibration correction will be broadened and the performance of the vibration correction will also be improved by using the proposed fusion method in the future.

Heating, ventilation and cooling (HVAC) is the largest source of residential energy consumption in United States, encompassing about 25% of total residential energy usage. A significant portion of energy is wasted by unnecessary operation, such as overheating/overcooling or operation without occupants. Wasteful behaviors will consume twice the amount of energy compared to energy conscious behaviors. Many market programmable thermostats exist to address this problem, however, difficulties in persistent programming of such products and lack of understanding of underlying physics prevent users from achieving tangible impact. Hence, fully autonomous energy control system is desirable to engage as many people into energy conscious behaviors as possible. Occupancy measurement is necessary components to enable fully autonomous control. Occupancy information can save energy by automatically turn off the HVAC system when the building is not occupied, or floats to a more energy-efficient setback temperature when the activity level is low. A number of existing sensor solutions available on the market include Passive Infrared (PIR), ultrasonic, Bluetooth/GPS, and CO 2 sensors, but these are either too expensive, not user-friendly, or limited in detection scope. These sensors are also incapable of detecting whether or not the occupant is an animal or a human. The work in this paper proposes an economical, reliable, non-invasive package to both detect human presence in a residence of a wide variety of geometries at the time and predict future occupancy pattern, by utilizing temperature sensors. To accomplish this, thermal sensors will be attached to both ends of door handles to collect the temperature data. This data will allow us to create a schedule to identify human activity leaving and exiting the space. At the same time, we will be collecting the skin temperature to determine the human activity level for better identification of the thermal comfort zone for occupants. The prediction model for occupancy pattern will be developed from previous data by using machine learning algorithm. For verification, experimental setup was built to verify our model by comparing actual human presence data from a house with the measured and predicted occupancy pattern from the temperature sensors. Future steps include implementing a data fusion scheme into the model to combine information from multiple types of sensors.

Inertial and magnetic sensors are commonly used to determine orientation as they do not rely on a line of sight [1, 2]. There are many different techniques to fuse inertial measurement unit (IMU) data and obtain useful rotational data [1–3]. This study uses two separate data fusion techniques; a direction cosine matrix-based (DCM) technique and a quaternion-based Extended Kalman Filter (EKF) technique [1–3]. These techniques were altered based on performance metrics to weight sensor data when certain sensors proved not as reliable as others [2]. IMU sensors were tested on a hand mannequin and filters were developed using MATLAB software. Simulation results displayed a root-mean-squared error of less than .06° for each rotation angle. Experimental results maintained errors of less than 8° in each rotation angle.

Up-down and rifle aiming maneuvers are common tasks employed by soldiers and athletes. The movements underlying these tasks largely determine performance success, which motivates the need for a noninvasive and portable means for movement quantification. We answer this need by exploiting body-worn and rifle-mounted miniature inertial measurement units (IMUs) for measuring torso and rifle motions during up-down and aiming tasks. The IMUs incorporate MEMS accelerometers and angular rate gyros that measure translational acceleration and angular velocity, respectively. Both sensors enable independent estimates of the orientation of the IMU and thus, the orientation of a subject’s torso and rifle. Herein, we establish the accuracy of a complementary filter which fuses these estimates for tracking torso and rifle orientation by comparing IMU-derived and motion capture-derived (MOCAP) torso pitch angles and rifle elevation and azimuthal angles during four up-down and rifle aiming trials for each of 16 subjects (64 trials total). The up-down trials consist of five maximal effort get-down-get-up cycles (from standing to lying prone back to standing), while the rifle aiming trials consist of rapidly aiming five times at two targets 15 feet from the subject and 180 degrees apart. Results reveal that this filtering technique yields warfighter torso pitch angles that remain within 0.55 degrees of MOCAP estimates and rifle elevation and azimuthal angles that remain within 0.44 and 1.26 degrees on average, respectively, for the 64 trials analyzed. We further examine potential remaining error sources and limitations of this filtering approach. These promising results point to the future use of this technology for quantifying motion in naturalistic environments. Their use may be extended to other applications (e.g., sports training and remote health monitoring) where noninvasive, inexpensive, and accurate methods for reliable orientation estimation are similarly desired.

As a type of clean and renewable energy source, wind power is growing fast as more and more countries lay emphasis on it. At the end of 2011, the global wind energy capacity reached 238 GW, with a cumulative growth of more than 20% per year, which is certainly a respectable figure for any industry. There is an exigent need to reduce the costs of operating and maintaining wind turbines while they became one of the fastest growing sources of power production in the world today. Gearbox is a critical component in the transmission system of wind turbine generator. Wind turbine gearbox operates in the extreme conditions of heavy duty, low speed and non-stationary load and speed, etc., which makes it one of the components that have high failure rate. To detect the fault of gearbox, many methods have been developed, including vibration analysis, acoustic emission, oil analysis, temperature monitoring, and performance monitoring and so on. Vibration analysis is widely used in fault diagnosis process and many efforts have been made in this area. However, there are many challenging problems in detecting the failure of wind turbine gearbox. The gearbox transforms low-speed revolutions from the rotor to high-speed revolutions, for example, from 20 rpm to 1500 rpm or higher. Usually one or more planetary gear stages are adopted in a gearbox design because the load can be shared by several planet gears and the transmission ratio can get higher. One disadvantage with the planetary gear stage is that a more complex design makes the detection and specification of gearbox failure difficult. The existing fault diagnosis theory and technology for fixed-shaft gearbox cannot solve the issues in the fault diagnosis of planetary gearbox. The planetary stage of wind turbine gearbox consists of sun gear, ring gear and several planet gears. The planet gears not only rotate around their own centers but also revolve around the sun gear center, and the distance between each planet gear to the sensor varies all the time. This adds complexity to vibration signals and results in difficulty in finding the fault-related features. The paths through which the vibration propagates from its origin to the sensors are complex, and the gears of other stage vibrate at the same time. This makes fault features be buried in noises. Further, the extreme conditions of heavy duty, low speed, and non-stationary workload lead to evidently non-stationary phenomena in the collected vibration. Methods to assess fault severity of a gearbox should be developed so as to realize fault prognosis and estimate of the remaining useful life of gearbox. Finally, other issues like signal analysis based on multi-sensor data fusion are also considered. This paper gives a comprehensive investigation on the state-of-the-art development in the wind turbine gearbox condition monitoring and health evaluation. The general situation of wind energy industry is discussed, and the research progresses in each aspects of wind turbine gearbox are reviewed. The existing problems in the current research are summarized in the end.

The integration of multi-view data in the acquisition of complex surface is researched by using the CMM (coordinate measuring machine) and the laser tracking scanning system. A multi-view data integration method based on the auxiliary reference plane is presented. With the precise auxiliary reference plane, the data measured by the different measuring devices and measuring views are aligned. The problems of low-efficient and low-accuracy in the data fusion of complex surface during the reverse engineering are solved. Meanwhile, the advantages of the different measuring devices and measuring methods are developed. This method has been applied to the reverse engineering development for a motorcycle cover model. The practical result shows that this new method is feasible and efficient.

Given the highly nonlinear attribute of the underlying dynamics associated with the time evolution of multibody systems, an open question in mechanical system simulation is how one can reliably replace a model whose simulation is time consuming with a more expeditious one. Pushing this idea to the limit one can all together eliminate the dynamics of the problem using a set of simulations that train a predictor that is later used to provide the time evolution of the dynamic system. This paper investigates a Gaussian Random Function (GRF) based approach that attempts to address these questions. It relies on a framework recently proposed in the Statistical Analysis community that largely deals with the issues of model validation, calibration, and data integration. The approach investigated has several steps that are illustrated with a slider-crank mechanical systems whose time evolution is governed by a nonlinear set of index 3 Differential Algebraic Equations (DAEs). The paper concludes with a set of remarks on the potential of GRFs in the context of time domain analysis of mechanical systems.

In this paper, a framework to schedule and filter distorted multi-sensor outputs is developed. The distortion caused by sensor nonlinearity is considered. As the distorted signal is not a true representation of the original signal, it is often necessary to develop and incorporate signal recovery schemes. Once such a scheme is developed, implementation may be straightforward if the system is employed with a single sensor. When the bandwidth of a signal of interest is very high, the use of a single sensor may not be feasible. High cost and accuracy are major concerns worth noting. It is proposed that a good practical solution to this problem is to employ an array of low bandwidth sensors. Practical Implementation of recovery schemes is very challenging and difficult in this case due to a possible overlapping of multi-source data. This sensor scheduling problem is investigated in detail and a data fusion scheme based on the optimization of weighted error function is initiated and developed for the two-sensor case. Simulation results are presented to validate the fusion procedure developed.

This paper presents a fuzzy data fusion scheme for bearing defect severity classification. Instead of using time-domain features (e.g. peak value, root mean square, kurtosis) of the original signal as the inputs, wavelet transform was applied to the original signals as a data pre-processor. Defect-related features were filtered out from the corresponding wavelet coefficients. Next, fuzzifications on individual input features were performed to obtain individual decisions. Finally, data fusion was peformed to achieve an overall bearing defect severity classification. Three data fusion methods: voting, weighted decision, and fuzzy inference, were investigated. Experimental results have shown that the presented fuzzy data fusion scheme is more effective than traditional approaches using raw signals as the input features. In particular, the presented scheme has resolved conflicting decisions arising from single feature-based classification, thus improving the reliability of defect severity assessment.

Multi-scale measurements, i.e. measurements of strain, strain gradient and integrated strain data, throughout a structural volume have demonstrated a great potential for improved damage identification. However, the large number of data and their different forms make fusion of the data difficult. To overcome this problem, a neural network data fusion approach is proposed. A simulation of damage identification in an isotropic cracked plate is presented. The crack position, angle and crack length are used as test parameters to be determined. A back-propagation neural network is trained to reproduce the crack angle and length as a function of all sensor responses. The improvement gained by using both multi-scale sensing and neural network data fusion for this specific case is significant. Testing of the sensitivity of the method to measurement errors or missing data demonstrated the robustness of the neural network to errors.

Proposed in this paper is a new method to implement the high operating bandwidth sensor arrays. In certain control applications, it is necessary that a high bandwidth sensor be used to improve the efficiency of feedback. The design of a single sensor with the desired high bandwidth may not be easy and economically feasible. It is shown that the idea of sensor arrays can be utilized to obtain a cost effective and efficient solution to the problem posed. It is discussed that an effective data fusion scheme is necessary in order to implement the proposed sensor array that consists of low bandwidth pass-band sensors with possible overlapping operating regions. Moreover, we point out that obtaining accurate sensor models may not be always easy in practice and this may make the proposed sensor arrays inapplicable for certain applications. To address this issue, a new implementation scheme that utilizes feedback mechanisms to combine multi-sensor data is developed. The proposed framework is validated using simulation examples.

This paper presents a generic model for an integrated smart health monitoring system for infrastructures using multisensor fusion and condition assessment sheets. Though various techniques for health monitoring have been discussed extensively in the literature, little attention has been given to obtain high quality data from the measurement and sensing system by using an intrinsic knowledge base. The method proposed in this paper uses measurement data from different types of sensors with different resolutions and fuses it together based on the confidence in them derived from information not typically used in traditional data fusion methods. Examples of such information are operating temperature, frequency range, fatigue cycles, etc. These are fed as additional inputs to a fuzzy inference system (FIS) that has predefined membership functions for each of these variables. The outputs of the FIS are weights that are assigned to the different sensor measurement data that reflect the confidence in the sensor’s behavior and performance. A modular approach is adopted for the data fusion system. It allows adding or deleting a sensor, along with its fuzzy logic controller (FLC), anytime without affecting the entire data fusion system. The time history of problems and solutions taken to correct them are stored as a condition assessment sheet (CAS) that shows the health of each sensor and the entire measurement system at a glance. This work finds applications in the health management of civil infrastructures, power plants, airplanes and rocket/shuttle test facilities.

The long-term goal of this project is the development of embedded, optimally distributed, multi-scale sensing methodologies that can be integrated into material systems for failure identification in structural systems. The coupling of sensor data fusion with a three-dimensional predictive framework will provide insight and understanding of events that are difficult, if not impossible, in any experimental study, such as subsurface damage and crack nucleation in structural systems. The current work presents an experimental study of the survivability and degradation behavior of an optical fiber Bragg grating sensor, surface mounted on a woven fiber composite material system during multiple low velocity impacts. The results reveal that as sensor degradation occurs, additional coupling phenomena other than Bragg reflection are observed in the grating sensor. From these additional modes, information on the sensor/host bond and fiber degradation is obtained.

In the power industry, many companies are trying to improve Equipment Reliability by focusing on Performance Monitoring and the application of diagnostic technologies. Software tools are focused on performing detailed technical analysis and trending of diagnostic parameters. The diagnostic data is often stored in archival databases that can track this data for many years. However, the analysis results generated by engineering and maintenance personnel are often stored outside the diagnostic tools in various databases, spreadsheets, and word documents. This valuable information is then difficult to track, trend, and then recall when a similar event occurs in the future. The paper will focus on developed and implemented web-based tools that facilitate tracking, trending, and sharing these analysis results across sites and enterprises. This paper further discusses implementation details and demonstrates the value of incorporating this information seamlessly into existing as well as developing condition monitoring programs.

The major thrust of this paper is to develop a sensor model based on a probabilistic approach that could accurately provide information about individual sensor’s uncertainties and limitations. The sensor model aims to provide a most informative likelihood function that can be used to obtain a statistical and probabilistic estimate of uncertainties and errors due to some environmental parameters or parameters of any feature extraction algorithm used in estimation based on sensor’s outputs. This paper makes use of a neural network that has been trained with the help of a novel technique that obtains training signal from a maximum likelihood estimator. The proposed technique was applied to model stereo-vision sensors and Infra-Red (IR) proximity sensor, and information from these sensors were fused in a Bayesian framework to obtain a three-dimensional occupancy profile of objects in robotic workspace. The capability of the proposed technique in accurately obtaining three-dimensional occupancy profile and efficiently removing individual sensor uncertainties was demonstrated and validated via experiments carried out in the Robotics and Manufacturing Automation (RAMA) Laboratory at Duke University.

To analyze the steady state response of structural dynamical systems with multi-field response (example, Timoshenko shearable rod) given complex-valued databases (finite element simulations of complexified equations of motion), we have developed a Complex Proper Orthogonal Decomposition (CPOD) transform. Like the regular multi-field POD, the development of the C-POD is based on the primitive space and frequency auto-correlation operations. These data fusion operations give rise to complex Hermitian operators whose solution determines the C-POD transform. The eigen-values of the complex Hermitian operators are strictly positive and it is shown that they represent the energy fractions of the auto-correlation energy contained in the POD modes. The POD modes have both amplitudes and shapes that are complex-valued scalar functions. The C-POD transform is verified by applying it to characterize the finite element simulations of the steady state dynamics of planar beams and arches. It turns out that the real part of the shape of a POD mode coincides with the shape of the linear POD; whereas its amplitude is a localized function of frequency at a critical frequency which is identical to a natural frequency.

Current research in wireless sensor networks has chiefly focused on environmental monitoring applications. Wireless sensors are emerging as viable instrumentation techniques for industrial applications because of their flexibility, non-intrusive operation, safety and their low cost, low power characteristics. We describe a prototype gear condition monitoring system incorporating wireless sensors. Measurements of strain on gear teeth, vibration and temperature were undertaken using strain gage, accelerometer, and thermistors, respectively. The sensors interface to a sensor board that is connected to a microprocessor and a radio. Gear faults diagnosis using conventional classification techniques such as principle component analysis (PCA), Fisher linear discriminant analysis (LDA) and Nearest-Neighbor Rule (NNR) is studied in this paper. Two sets of vibration data, one set of strain data, and three sets of temperature data are used to classify a running gear under normal condition and a running gear with simulated crack teeth. Feature level data fusion is used to test the classification performance of simple but less effective features to study the fusion effects. The results show high performance of strain features, high quality of the classifier and obvious fusion effect which increases the classification performance.