Raw Acceleration Signals Research Articles

Monitoring framework development for a network of multiple laboratory structures

The stadium structures have unique structural features increasing the significance of structural monitoring systems specifically designed for them. Aside from vibrations serviceability concerns and human-induced excitations, the development and propagation of structural damage under all possible atmospheric and seismic conditions need to be closely monitored for structural resiliency and integrity of the stadia. As such, Structural Health Monitoring (SHM) methods combined with effective data evaluation methodologies need to be deployed to monitor the structural performance of stadiums. Even though stadia monitoring has been performed at multiple locations in the world, a web based and real-time SHM network of stadia is not known to authors. As a preliminary study for the network implementation of stadia monitoring with acceleration measurements, the presented work focuses on the fundamental steps to accomplish this goal, with a collaborative research effort between Qatar University, the University of Central Florida, and University of Alberta. The authors performed analytical investigations and experimental testing on stadium-type structures built in laboratory environments for the development of the SHM framework. Specialized signal processing algorithms, sensing suites and approaches considering multi-scale monitoring were used on collected acceleration measurements. The novelty of the work presented in this manuscript are the following items which exist simultaneously in the developed SHM framework. The developed framework is a web-based monitoring application where structural damage is detected in real-time. The proposed methodology operates directly on raw acceleration signals and runs at a network level. With that, the damage detection, damage localization, and damage quantification tasks are performed simultaneously, while the feature extraction and classification stages are combined in one learning body.

General spectral characteristics of human activity and its inherent scale-free fluctuations

The scale-free nature of daily human activity has been observed in different aspects; however, the description of its spectral characteristics is incomplete. General findings are complicated by the fact that—although actigraphy is commonly used in many research areas—the activity calculation methods are not standardized; therefore, activity signals can be different. The presence of 1/f noise in activity or acceleration signals was mostly analysed for short time windows, and the complete spectral characteristic has only been examined in the case of certain types of them. To explore the general spectral nature of human activity in greater detail, we have performed Power Spectral Density (PSD) based examination and Detrended Fluctuation Analysis (DFA) on several-day-long, triaxial actigraphic acceleration signals of 42 healthy, free-living individuals. We generated different types of activity signals from these, using different acceleration preprocessing techniques and activity metrics. We revealed that the spectra of different types of activity signals generally follow a universal characteristic including 1/f noise over frequencies above the circadian rhythmicity. Moreover, we discovered that the PSD of the raw acceleration signal has the same characteristic. Our findings prove that the spectral scale-free nature is generally inherent to the motor activity of healthy, free-living humans, and is not limited to any particular activity calculation method.

Inter-Brand, -Dynamic Range, and -Sampling Rate Comparability of Raw Accelerometer Data as Used in Physical Behavior Research

Objective: Previous studies that looked at comparability of accelerometer data focused on epoch or recording level comparability. Our study aims to provide insight into the comparability at raw data level. Methods: We performed five experiments with accelerometers attached to a mechanical shaker machine applying movement along a single axis in the horizontal plane. In each experiment, a 1-min no-movement condition was followed by nineteen 2-min shaker frequency conditions (30–250 rpm). We analyzed accelerometer data from Axivity, ActiGraph, GENEActiv, MOX, and activPAL devices. Comparability between commonly used brands and dynamic ranges was assessed in the frequency domain with power spectra and in the time domain with maximum lagged cross-correlation analyses. The influence of sampling rate on magnitude of acceleration across brands was explored visually. All data were published open access. Results: Magnitude of noise in rest was highest in MOX and lowest in ActiGraph. The signal mean power spectral density was equal between brands at low shaker frequency conditions (<3.13 Hz) and between dynamic ranges within the Axivity brand at all shaker frequency conditions. In contrast, the cross-correlation coefficients between time series across brands and dynamic ranges were higher at higher shaking frequencies. Sampling rate affected the magnitude of acceleration most in Axivity and least in GENEActiv. Conclusions: The comparability of raw acceleration signals between brands and/or sampling rates depends on the type of movement. These findings aid a more fundamental understanding and anticipation of differences in behavior estimates between different implementations of raw accelerometry.

Health state identification of circulating seawater pump-unit in nuclear power plant based on multi-virtual vibration source fusion in the presence of strong data imbalance

The circulating seawater pump-unit (CRF pump-unit) is a crucial equipment that provides cooling water for the secondary circuit of a nuclear power plant (NPP). Its long-term, fault-free operation ensures nuclear power safety and economy. In this study, considering the limited fault-state samples in the historical database and the random fault locations in NPP equipment, a health state identification framework of the CRF pump-unit is proposed, which combines multiple sources of network input and a feature fusion network. The proposed kernelled independent component analysis with reference (KICA-R) is utilized with reference signals to separate virtual sources from multiple raw acceleration signals. It enhances the fault characteristic frequencies related to several subsystems of a complex mechatronic system in the training samples of the network. Based on the newly designed multi-source channel attention module (MS-CAM) that can adaptively allocate fault-sensitive weight coefficients, a health state identification network incorporates the enhanced input features through the virtual sources obtained from KICA-R. By conducting the benchmark analysis on two public datasets and various types of fault simulation tests on the CRF pump-unit simulator, training sets with different levels of data imbalance are constructed, and the optimal network parameters are determined. The comparisons against cICA validate the ability of KICA-R in virtual source separation, and the strong performance of the MS-CAM-based network in identifying different health states of the CRF pump-unit is further demonstrated through detailed comparative analysis.

Kernel Regression Residual Decomposition-Based Polynomial Frequency Modulation Integral Algorithm to Identify Physical Parameters of Time-Varying Systems under Random Excitation

The physical parameters (stiffness, damping) of time-varying (TV) systems under random excitation provide valuable information for their working condition but they are often overwhelmed by noise interference. To overcome this problem, this paper presents a novel multi-level kernel regression residual decomposition method, which can not only effectively separate each modal component from the raw vibration acceleration signal, but also eliminate noise interference. Additionally, the multiple degree-of-freedom (DOF) parameter identification problem is transformed into a single DOF parameter identification problem. Combined with the derived polynomial frequency modulation integral algorithm and the cross-correlation theory based on the fractional Fourier ambiguity function, a physical parameter identification method is proposed. The method provides a new idea in modeling TV systems and identifying physical parameters under random excitation. To demonstrate the effectiveness of the proposed method, numerical simulations are conducted with three different cases of variation (variation, quadratic variation, and periodic variation) in time. Moreover, its robustness is evaluated by adding different signal-to-noise ratio levels of noise (20 dB, 50 dB, 100 dB) to the input vibration acceleration signal. The analysis results confirm the performance of the proposed method for the parameter identification of TV systems under random excitation.

Rolling bearing faults identification based on multiscale singular value

To solve the vulnerability of singular value denoising to the effective number of singular value orders and influence of Hankel matrix structure, the paper has proposed a multiscale singular value denoising algorithm (M-SVD). Meanwhile, in concerning that when a fault of bearing occurs, the fault feature information contained in vibration signal tends to be weak and complex, the M-SVD method is blended with variational mode decomposition (VMD) algorithm in identification. Firstly, when bearing faults occur, raw vibration acceleration signals are decomposed using the superiority of VMD algorithm for non-stationary signal; in addition, considering that with the larger difference between singular values of signal, feature information will be more obvious, the method takes advantage of sensibility of variance to signal intensity and proposes to constitute Hankel matrix based on variance feature of signal. During the construction of Hankel matrix, the method to determine effective length of variance sequence (optimal scale) has been brought forward by sample entropy which is able to precisely measure the regularity of signal and detect the weak change of signal. Furthermore, component signals are denoised in line with reconstructed Hankel matrix and the spectrum of signals denoised are selected to describe fault characteristic information and judge the failure types. For the sake of verification of the accuracy and effectiveness of proposed method, the paper has analyzed and verified the data of rolling bearing in 3 different failure types and 2 different installation positions of accelerometers. The comparative analysis with classical singular value difference spectrum (SVDS) and singular value median decomposition (SVMD) denoising has further proved that presented VMD combined M-SVD method can effectively restrain noise while retaining fault characteristic information of signal. Therefore, the identification for rolling bearing fault types can be implemented, accurately.

Performance of Different Accelerometry-Based Metrics to Estimate Oxygen Consumption during Track and Treadmill Locomotion over a Wide Intensity Range.

Accelerometer data can be used to estimate incident oxygen consumption (VO2) during physical activity. Relationships between the accelerometer metrics and VO2 are typically determined using specific walking or running protocols on a track or treadmill. In this study, we compared the predictive performance of three different metrics based on the mean amplitude deviation (MAD) of the raw three-dimensional acceleration signal during maximal tests performed on a track or treadmill. A total of 53 healthy adult volunteers participated in the study, 29 performed the track test and 24 the treadmill test. During the tests, the data were collected using hip-worn triaxial accelerometers and metabolic gas analyzers. Data from both tests were pooled for primary statistical analysis. For typical walking speeds at VO2 less than 25 mL/kg/min, accelerometer metrics accounted for 71-86% of the variation in VO2. For typical running speeds starting from VO2 of 25 mL/kg/min up to over 60 mL/kg/min, 32-69% of the variation in VO2 could be explained, while the test type had an independent effect on the results, except for the conventional MAD metrics. The MAD metric is the best predictor of VO2 during walking, but the poorest during running. Depending on the intensity of locomotion, the choice of proper accelerometer metrics and test type may affect the validity of the prediction of incident VO2.

Flexible sensor-based biomechanical evaluation of low-back exoskeleton use in lifting

This study aimed to establish an ambulatory field-friendly system based on miniaturised wireless flexible sensors for studying the biomechanics of human-exoskeleton interactions. Twelve healthy adults performed symmetric lifting with and without a passive low-back exoskeleton, while their movements were tracked using both a flexible sensor system and a conventional motion capture (MoCap) system synchronously. Novel algorithms were developed to convert the raw acceleration, gyroscope, and biopotential signals from the flexible sensors into kinematic and dynamic measures. Results showed that these measures were highly correlated with those obtained from the MoCap system and discerned the effects of the exoskeleton, including increased peak lumbar flexion, decreased peak hip flexion, and decreased lumbar flexion moment and back muscle activities. The study demonstrated the promise of an integrated flexible sensor-based system for biomechanics and ergonomics field studies as well as the efficacy of exoskeleton in relieving the low-back stress associated with manual lifting.

Novel Deep Learning Network for Gait Recognition Using Multimodal Inertial Sensors.

Some recent studies use a convolutional neural network (CNN) or long short-term memory (LSTM) to extract gait features, but the methods based on the CNN and LSTM have a high loss rate of time-series and spatial information, respectively. Since gait has obvious time-series characteristics, while CNN only collects waveform characteristics, and only uses CNN for gait recognition, this leads to a certain lack of time-series characteristics. LSTM can collect time-series characteristics, but LSTM results in performance degradation when processing long sequences. However, using CNN can compress the length of feature vectors. In this paper, a sequential convolution LSTM network for gait recognition using multimodal wearable inertial sensors is proposed, which is called SConvLSTM. Based on 1D-CNN and a bidirectional LSTM network, the method can automatically extract features from the raw acceleration and gyroscope signals without a manual feature design. 1D-CNN is first used to extract the high-dimensional features of the inertial sensor signals. While retaining the time-series features of the data, the dimension of the features is expanded, and the length of the feature vectors is compressed. Then, the bidirectional LSTM network is used to extract the time-series features of the data. The proposed method uses fixed-length data frames as the input and does not require gait cycle detection, which avoids the impact of cycle detection errors on the recognition accuracy. We performed experiments on three public benchmark datasets: UCI-HAR, HuGaDB, and WISDM. The results show that SConvLSTM performs better than most of those reporting the best performance methods, at present, on the three datasets.

Integrating flexibility-based curvature with quasi-static features induced by traffic loads for high-resolution damage localization in bridges

Curvature is one of the most popular damage-sensitive features in vibration-based structural health monitoring applications, typically calculated from identified modal features. While the relevant strategic or historical importance of bridges may justify dense sensor networks, a limited budget is generally assigned to monitor “minor” viaducts, thus involving inexpensive devices or extremely sparse sensing solutions. Modal parameters can only be obtained at instrumented locations. Thereby, damage assessment methods based on identified features typically have a low spatial resolution, especially when using low-cost monitoring setups with a modest number of sensing devices. This paper proposes an original identification method for the curvature of bridges based on sparse acceleration measurements that can be collected using standard accelerometers. The raw acceleration signal is processed through a particular filter bank that extracts dynamic and quasi-static signal components. The first components are employed to identify modal parameters, from which sparse yet robust estimates of the structural curvature are retrieved. On the other hand, the quasi-static acceleration generated by the structural deflection induced by traffic load is used to identify the curvature influence lines of the bridge, which are fused with modal estimates using a Kalman filter. The state variable of the analyzed system, representing a dense curvature profile of the structure subjected to concentrated loads, can be used as a damage-sensitive feature for high-resolution damage localization. The method is applied to a steel truss bridge subject to different damage configurations.

A study of the influence of the sensor sampling frequency on the performance of wearable fall detectors

Last decade has witnessed a major research interest on wearable fall detection systems. Sampling rate in these detectors strongly affects the power consumption and required complexity of the employed wearables. This study investigates the effect of the sampling frequency on the efficacy of the detection process. For this purpose, we train a convolutional neural network to directly discriminate falls from conventional activities based on the raw acceleration signals captured by a transportable sensor. Then, we analyze the changes in the performance of this classifier when the sampling rate is progressively reduced. In contrast with previous studies, the detector is tested against a wide set of public repositories of benchmarking traces. The quality metrics achieved for the different frequencies and the analysis of the spectrum of the signals reveal that a sampling rate of 20 Hz can be enough to maximize the effectiveness of a fall detector.

A Transparent Method for Step Detection using an Acceleration Threshold.

Step-based metrics provide simple measures of ambulatory activity, yet device software either includes undisclosed proprietary step detection algorithms or simply do not compute step-based metrics. We aimed to develop and validate a simple algorithm to accurately detect steps across various ambulatory and non-ambulatory activities. Seventy-five adults (21-39 years) completed seven simulated activities of daily living (e.g., sitting, vacuuming, folding laundry) and an incremental treadmill protocol from 0.22-2.2ms-1. Directly observed steps were hand-tallied. Participants wore GENEActiv and ActiGraph accelerometers, one of each on their waist and on their non-dominant wrist. Raw acceleration (g) signals from the anterior-posterior, medial-lateral, vertical, and vector magnitude (VM) directions were assessed separately for each device. Signals were demeaned across all activities and bandpass filtered [0.25, 2.5Hz]. Steps were detected via peak picking, with optimal thresholds (i.e., minimized absolute error from accumulated hand counted) determined by iterating minimum acceleration values to detect steps. Step counts were converted into cadence (steps/minute), and k-fold cross-validation quantified error (root mean squared error [RMSE]). We report optimal thresholds for use of either device on the waist (threshold=0.0267g) and wrist (threshold=0.0359g) using the VM signal. These thresholds yielded low error for the waist (RMSE<173 steps, ≤2.28 steps/minute) and wrist (RMSE<481 steps, ≤6.47 steps/minute) across all activities, and outperformed ActiLife's proprietary algorithm (RMSE=1312 and 2913 steps, 17.29 and 38.06 steps/minute for the waist and wrist, respectively). The thresholds reported herein provide a simple, transparent framework for step detection using accelerometers during treadmill ambulation and activities of daily living for waist- and wrist-worn locations.

Spectral Summation With Machine Learning Analysis of Tri-Axial Acceleration From Multiple Wearable Points on Human Body for Better Cough Detection

In this letter, coughing is detected using a multiband spectral summation of acceleration signal with machine learning (ML) from various body points where smartphones are commonly worn. A known challenge in this letter is discerning low and mid intensity cough events from noise introduced by walking from the chest, stomach, shirt pocket, upper hand, and ear where smartphones are commonly worn from among seven test subjects of varying heights. Previous studies have shown that coughing during walking can be accurately detected with only 92, 73, 62, and 82% accuracy at the chest, stomach, shirt pocket, and upper hand, respectively, just from raw acceleration signals in the time domain and ML. Newer spectrum analysis show that acceleration measured at these body points along the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">y</i> -axis are more spectrally rich during low to medium coughing and walking activity than just walking alone in seven test subjects of varying heights. In this letter, we investigate the use multiband spectral summation features of acceleration measured at these same body points on the torso with ML to improve the accuracy of low/mid intensity cough detection to between 95.2 and 98.2%. At just the chest and upper hand, the spectral sum of acceleration in the 0–5 Hz band shows a 46–142% and 58–136% increase during coughing and walking than just walking alone for different cough intensities and subject heights. This letter is useful in developing future cough-detecting apps on smartphones commonly worn on the torso.

Generative Adversarial Network for Damage Identification in Civil Structures

In recent years, many efforts have been made to develop efficient deep‐learning‐based structural health monitoring (SHM) methods. Most of the proposed methods employ supervised algorithms that require data from different damaged states of a structure in order to monitor its health conditions. As such data are not usually available for real civil structures, using supervised algorithms for the health monitoring of these structures might be impracticable. This paper presents a novel two‐stage technique based on generative adversarial networks (GANs) for unsupervised SHM and damage identification. In the first stage, a deep convolutional GAN (DCGAN) is used to detect and quantify structural damages; the detected damages are then localized in the second stage using a conditional GAN (CGAN). Raw acceleration signals from a monitored structure are used for this purpose, and the networks are trained by only the intact state data of the structure. The proposed method is validated through applications on the numerical model of a bridge health monitoring (BHM) benchmark structure, an experimental steel structure located at Qatar University, and the full‐scale Tianjin Yonghe Bridge.

Development and Validation of Open-Source Activity Intensity Count and Activity Intensity Classification Algorithms from Raw Acceleration Signals of Wearable Sensors.

Background: A popular outcome in rehabilitation studies is the activity intensity count, which is typically measured from commercially available accelerometers. However, the algorithms are not openly available, which impairs long-term follow-ups and restricts the potential to adapt the algorithms for pathological populations. The objectives of this research are to design and validate open-source algorithms for activity intensity quantification and classification. Methods: Two versions of a quantification algorithm are proposed (fixed [FB] and modifiable bandwidth [MB]) along with two versions of a classification algorithm (discrete [DM] vs. continuous methods [CM]). The results of these algorithms were compared to those of a commercial activity intensity count solution (ActiLife) with datasets from four activities (n = 24 participants). Results: The FB and MB algorithms gave similar results as ActiLife (r > 0.96). The DM algorithm is similar to a ActiLife (r ≥ 0.99). The CM algorithm differs (r ≥ 0.89) but is more precise. Conclusion: The combination of the FB algorithm with the DM results is a solution close to that of ActiLife. However, the MB version remains valid while being more adaptable, and the CM is more precise. This paper proposes an open-source alternative for rehabilitation that is compatible with several wearable devices and not dependent on manufacturer commercial decisions.

Intelligent fault diagnosis of rotating machinery based on a novel lightweight convolutional neural network

For most existing fault diagnosis methods, feature extraction is always based on a complex artificial design and the complete feature extraction from an original signal. With the gradual complication of modern industrial machinery and equipment, it has become more difficult for traditional feature extractors to achieve the desired results. Deep convolutional neural networks (DCNNs) have been developed as effective techniques for fault classification but require large-scale high-intensity computing and prohibitive hardware resource requirements. This paper proposes a lightweight CNN that can be easily used for the fault diagnosis of rotating machinery by adjusting the network structure and optimizing the network. First, the raw vibration acceleration signal is transformed into a two-dimensional gray image. Second, two mature and commonly used modules named LeNet and NIN are combined to form a new model with a simple structure. Then, through parameter adjustment and optimization, an improved and optimized CNN with a lightweight structure and fewer parameters is constructed. The experimental verification has shown that this method has high accuracy and stability in fault diagnosis. Finally, the application of this new network for the fault diagnosis of rolling bearings with different damage levels but similar fault types shows high diagnostic accuracy and good generalization ability. In addition, we attempt to explain how the feature filters of a CNN work by visualizing the convolutional layer of the network.

Accelerometer-Based Key Generation and Distribution Method for Wearable IoT Devices

With the fast development of wearable IoT devices, their applications are becoming more and more pervasive, ranging from social networking, payment, and navigation to health and activity monitoring. The security of the communication between these devices is essential to protect the transmitted sensitive information from tampering and eavesdropping. With the integration of accelerometers into wearable IoT devices, the gait-based biometric cryptography technology has emerged as a data securing tool for wearables. This article proposes a lightweight noise-based group key generation method, which utilizes the noise signals imposed on the raw acceleration signals to generate an M-bit key with high randomness and bit generation rate. Moreover, a signed sliding window coding (SSWC)-based common feature extraction method was designed to extract the common feature for sharing the generated M-bit key among devices worn on different body parts. Finally, a fuzzy vault-based group key distribution system was implemented and evaluated using a public data set. The performed comprehensive analysis of the proposed key generation and distribution method proved that the binary keys generated via the introduced noise-based procedure have high entropy and can pass both the NIST and Dieharder statistical tests with high efficiency. The experimental results obtained prove the robustness of the proposed SSWC-based common feature extraction method in terms of the similarity and discriminability of intra- and inter-class features, respectively.

Machine learning to quantify habitual physical activity in children with cerebral palsy.

To investigate whether activity-monitors and machine learning models could provide accurate information about physical activity performed by children and adolescents with cerebral palsy (CP) who use mobility aids for ambulation. Eleven participants (mean age 11y [SD 3y]; six females, five males) classified in Gross Motor Function Classification System (GMFCS) levels III and IV, completed six physical activity trials wearing a tri-axial accelerometer on the wrist, hip, and thigh. Trials included supine rest, upper-limb task, walking, wheelchair propulsion, and cycling. Three supervised learning algorithms (decision tree, support vector machine [SVM], random forest) were trained on features in the raw-acceleration signal. Model-performance was evaluated using leave-one-subject-out cross-validation accuracy. Cross-validation accuracy for the single-placement models ranged from 59% to 79%, with the best performance achieved by the random forest wrist model (79%). Combining features from two or more accelerometer placements significantly improved classification accuracy. The random forest wrist and hip model achieved an overall accuracy of 92%, while the SVM wrist, hip, and thigh model achieved an overall accuracy of 90%. Models trained on features in the raw-acceleration signal may provide accurate recognition of clinically relevant physical activity behaviours in children and adolescents with CP who use mobility aids for ambulation in a controlled setting. Machine learning may assist clinicians in evaluating the efficacy of surgical and therapy-based interventions. Machine learning may help researchers better understand the short- and long-term benefits of physical activity for children with more severe motor impairments.

The difference between first and second harmonic amplitudes correlates between glottal airflow and neck-surface accelerometer signals during phonation

Miniature high-bandwidth accelerometers on the anterior neck surface are used in laboratory and ambulatory settings to obtain vocal function measures. This study compared the widely applied L1-L2 measure (historically, H1-H2)-the difference between the log-magnitude of the first and second harmonics-computed from the glottal airflow waveform with L1-L2 derived from the raw neck-surface acceleration signal in 79 vocally healthy female speakers. Results showed a significant correlation (r = 0.72) between L1-L2 values estimated from both airflow and accelerometer signals, suggesting that raw accelerometer-based estimates of L1-L2 may be interpreted as reflecting glottal physiological parameters and voice quality attributes during phonation.

Applying singular value decomposition on accelerometer data for 1D convolutional neural network based fall detection

The usefulness of applying singular value decomposition (SVD) on triaxial accelerometer data for one-dimensional (1D) convolutional neural network (CNN) based fall and activity recognition is investigated. Three-dimensional reduction methods, namely, SVD, sparse principal component analysis, and kernel principal component analysis, are compared for their effectiveness in extracting useful features for fall and activity recognition. Experiments conducted on three public falls and activities of daily living datasets show that SVD applied to acceleration data coupled with raw acceleration data or acceleration signal magnitude vector exhibited better 1D CNN fall and activity recognition accuracy than those using other principal component analysis based acceleration features.