Smart sensors with extreme-edge data processing capabilities allow to optimize the energy budget of current Structural Health Monitoring (SHM) systems. Moving inference and Deep Neural Networks (DNNs) directly to the sensing unit can bypass the constraints due to data transmission, but it has to be properly balanced against the expenditure due to data sampling, which might easily become the main source of energy consumption when the duty cycling is not chosen appropriately. Finding the optimal sampling time and frequency is a non trivial task because it depends on the peculiarities of the target facility, the surrounding environment, and the highly non-linear behavior of DNNs. This paper presents a fully decentralized scheme for vibration damage detection where each sensor can predict the health status of the structure; a tiny Convolutional Neural Network (CNN) hosted on the node is entitled to autonomously process the measurements and to complete the inference. As a further contribution, an empirical study proves the major role played by the sampling parameters on the performance of the networks. Experiments on the Z24 bridge benchmark show that the small-size CNN can achieve classification scores comparable with state-of-the-art models while i) avoiding data transmission and ii) improving the energy budget if the optimal sensing parameters are chosen. These findings are obtained after deploying the sought CNN model on a wireless accelerometer sensor based on an STM32L496 processor. Results show that, by properly tuning the sampling parameters (length of the acquisition window and sampling frequency), one can reach the best trade-off between classification performances (accuracy above 96%) and energy consumption (less than 40 mJ/h).

Internet-of-Things (IoT) is a key enabler for the transition to the Automatic Structural Health Monitoring (ASHM) of technical facilities, thanks to the seamless flow of data from a multitude of always connected devices. Current IoT-ASHM installations, however, face the double challenge to ensure high accuracy while meeting the requirement of minimal energy consumption. The paper tackles these issues from a deep-learning perspective, and describes an IoT-enabled monitoring approach based on a distributed end-to-end deep neural network (DNN). The architecture supports both data compression and damage detection. A low-end microcontroller hosts a specific local DNN; a hardware-aware neural-architecture search strategy rules network optimization, in order to satisfy the resource constraints set by low-end computing devices. The features extracted from data feed an aggregating unit, which includes a stacked global classification layer for full-scale damage detection. After proper quantization, the designed models are eventually deployed on a wireless accelerometer sensor. Finally, a cost-benefit analysis evaluates the system’s impact on the sensor energy autonomy. Experiments on a well-known dataset proved that the proposed solution could achieve state-of-the-art classification scores (all metrics above 98.4%) with a minimal transmission cost (less than 53 B on average); as compared with conventional approaches, the described strategy yielded a reduction of three orders of magnitude in energy consumption.

Assessing the integrity of industrial and civil appliances has become a priority worldwide. Noteworthy, this goal requires a strong synergy between multiple tools, disciplines, and approaches to be attained via a joint hardware-software co-design of the different Structural Health Monitoring (SHM) system components. This work proposes the$\sf{MAC4PRO}$architecture, a sensor-to-cloud monitoring platform that seamlessly integrates sensing and software technologies for accurate data measurement, transmission, and analysis. The developed solution stands out for its interoperability and versatility, making it a promising candidate for integration in the next generation of smart structures. Our platform was validated during extensive experimental campaigns targeted at various industrial scenarios. The results show that the$\sf{MAC4PRO}$architecture can identify subtle changes, such as 1mm size leakage events in pipeline circuits, or less than 1% frequency drifts in civil buildings after seismic excitation, while ensuring more than 90% reduction in the edge-to-cloud data transfer process.

Compressed sensing (CS) for sensor-near vibration diagnostics represents a suitable approach for the design of network-efficient structural health monitoring systems. This article presents a solution for vibration analysis based on deep neural networks (DNNs) trained on compressed data. The envisioned maintenance system consists of a network of sensing nodes orchestrated by a very constrained centralizing unit. The latter is equipped with a microcontroller unit (MCU) that predicts the health state using the aggregated information. As a major contribution, the DNN architectures are generated automatically from the data through a procedure inspired by hardware-aware (HW) neural architecture search (NAS), called as HW-NAS-CS, which is uniquely refined with additional constraints that consider both the peculiarities of CS parameters and the limitation of embedded devices. The proposed approach has been validated using two real-world SHM datasets for vibration damage identification and eventually deployed on a low-end computing platform (the STM32L5 MCU). Results demonstrate that DNNs combined with adapted CS schemes can attain classification scores always above 90% even in case of very huge compression levels (higher than 64x): these performances significantly improve the ones attained by state-of-the-art approaches in the field, with the utmost advantage of being portable on embedded devices.

Mechanical complexity, wide dimensions, and big data volume may hamper the implementation of Internet of Things (IoT)-enabled structural health monitoring (SHM) systems. In particular, one of the most important challenges is the reduction of the data payload to be transmitted over the monitoring network. Addressing the problem in the context of vibration-based SHM, this work explores system identification (SysId) as an innovative strategy for data compression at the extreme edge. Indeed, SysId is a signal processing technique aiming at finding a very reduced (i.e., less then one tenth of the total signal length) set of meaningful parameters, which can provide an alternative, but yet completely equivalent, frequency characterization of the structure. In the proposed approach, an embedded system-oriented adaptation of the sequential tall-skinny QR decomposition (eS-TSQR) from the dense linear algebra domain has been exploited to tackle both the memory and computational complexity of the involved algorithms. This yielded to the embodiment of input–output and output-only SysId models into a resource constrained device (i.e., an STM32L5 microcontoller unit), targeted on low-power and low-cost SHM applications, proving high effectiveness for the structural assessment of civil and industrial plants. Besides, a cost-benefit analysis is also presented, in which the energy saving brought by SysId running in a sensor-near manner is comprehensively measured against the power consumption due to data transmission, as implied by state-of-the-art communication protocols for IoT. Results demonstrate that SysId is$1.19\times $and$2.78\times $less energy demanding (with a payload reduction of$9\times $and$45\times $) w.r.t. compressed sensing-driven and compression-free solutions, respectively.

Recent Structural Health Monitoring (SHM) systems might take advantage of Internet of Things (IoT) technologies for fine-grained and autonomic sensors data management and processing. Moreover, current SHM deployments often demand for the installation of multi-type and heterogeneous sensor devices capable to perform long-term measurements; from here, the need for dedicated software platforms allowing for scalability and interoperability requirements arises. In this paper, we jointly address the two issues above by proposing MODRON, which is a SHM-dedicated IoT platform with sensor-to-cloud support. The software architecture leverages the W3C Web of Things (WoT) standard for multi-source sensors data acquisition and fusion. The platform includes an edge component, implementing the communication with the monitoring layer and the data exposition through WoT Web Things (WTs), and a cloud component, embedding sensor/WT management capabilities, which is in charge of distributed data storage, aggregation, visualization and analytics. We illustrate the abstract MODRON architecture and its current implementation that supports two different SHM sensor types (MEMS accelerometers and piezoelectric devices). In addition, we describe the system operations on a real-world SHM system, i.e. the monitoring of a metallic structure instrumented with multiple sensor networks.

Condition Monitoring is one of the most critical applications of the Internet of Things (IoT) within the context of Industry 4.0. Current deployments typically present interoperability and management issues, requiring human intervention along the engineering process of the systems; in addition, the fragmentation of the IoT landscape, and the adoption of poor architectural solutions often make it difficult to integrate third-party devices in a seamless way. In this paper, we tackle these issues by proposing a tool-driven architecture that supports heterogeneous sensor management through well-established interoperability solutions for the IoT domain, i.e. the Eclipse Arrowhead framework and the recent Web of Things (WoT) standard released by the W3C working group. We deploy the architecture in a real Structural Health Monitoring (SHM) scenario, which validates each developed tool and demonstrates the increased automation derived from their combined usage.

The main challenge in the implementation of long-lasting vibration monitoring systems is to tackle the constantly evolving complexity of modern “mesoscale” structures. Thus, the design of energy-aware solutions is promoted for the joint optimization of data sampling rates, onboard storage requirements, and communication data payloads. In this context, the present work explores the feasibility of the rakeness-based compressed sensing (Rak-CS) approach to tune the sensing mechanism on the second-order statistics of measured data. In particular, a novel model-assisted variant (MRak-CS) is proposed, which is built on a synthetic derivation of the spectral profile of the structure by pivoting on numerical priors. Moreover, a signal-adapted sparsity basis relying on the wavelet packet transform operator is conceived, which aims at maximizing the signal sparsity while allowing for a precise time-frequency localization. The adopted solutions were tested with experiments performed on a sensorized pinned-pinned steel beam. Results prove that the rakeness-based compression strategies are superior to conventional eigenvalue approaches and to standard CS methods. The achieved compression ratio is equal to seven and the quality of the reconstructed structural parameters is preserved even in presence of defective configurations.

This paper presents a low-cost system that permits to provide structural modal analysis from a number of synchronized MEMS accelerometers distributed along the structure. The communication and computational effort is distributed among the nodes and data are provided real-time. Synchronization between sensors and accuracy of the measures permit to elaborate detailed modal analysis of the buildings. Experiments on real testbeds and structures demonstrate that the accuracy of the modal analysis is similar to the simulations executed using the models of the structures, and confirm that complex structural health evaluations are possible from a set of low-cost sensors.

Structural health monitoring is an important task in several industrial applications. For example, passive-only networks of piezoelectric sensors can be exploited to detect and localize impacts or anomalous acoustic emissions for the early identification of crack initiation and growth from the analysis of the emitted ultrasonic guided waves. In this work, a shaped piezoelectric sensor and the dedicated electronics for guided wave detection on laminate composite and metallic structures is presented. This device is meant as a basic building block for the development of passive sensor networks in smart structure applications. Elastic waves propagating on the structure are recorded by the transducer composed by multiple, irregularly shaped electrodes and then processed by a microcontroller embedded on the sensor node. The irregular shaping of the transducer electrodes is exploited to simplify the signal processing procedures that are necessary to detect the wave direction of arrival.

This paper proposes the use of independent component analysis (ICA) method for learning features from radio frequency (RF) ultrasonic signals. Conventional feature extractors usually suffer from limitations caused by some of their assumptions about the structure of imaged organ and the interaction between ultrasonic signal and tissue. ICA, on the other hand, is a data-driven approach which learns efficient representation of data by maximizing independence of some basis functions that describe the important structures of the data. It has less restrictive considerations about formation of data and as a consequence, could adapt itself to the characteristics of data with different natures. These features of ICA make it a proper candidate for dealing with the problem of feature extraction from medical ultrasound signals where variations in tissue structures and data acquisition conditions are considerable. Experimental results on raw backscattered RF signals of prostate gland show favorable performance of ICA as compared with the conventional methods designed for extracting feature from RF signals.

One of the popular structural health monitoring (SHM) applications of both automotive and aeronautic fields is devoted to the non-destructive localization of impacts in plate-like structures. The aim of this paper is to develop a miniaturized, self-contained and low power device for automated impact detection that can be used in a distributed fashion without central coordination. The proposed device uses an array of four piezoelectric transducers, bonded to the plate, capable to detect the guided waves generated by an impact, to a STM32F4 board equipped with an ARM Cortex-M4 microcontroller and a IEEE802.15.4 wireless transceiver. The waves processing and the localization algorithm are implemented on-board and optimized for speed and power consumption. In particular, the localization of the impact point is obtained by cross-correlating the signals related to the same event acquired by the different sensors in the warped frequency domain. Finally the performance of the whole system is analysed in terms of localization accuracy and power consumption, showing the effectiveness of the proposed implementation.

In this work we introduce a new frequency warping operator for non-smooth warping function allowing perfect reconstruction. The transformation is based on a previously introduced aliasing compensated frequency warping operator having a residual error because of finite output dimension. By adding some redundancy, the effect of an infinite output dimension can be taken into account in a compressed way, based on an analytical factorization. In the reconstruction process, the additional redundant samples are expanded, making the inverse transform differ from the direct one, but guaranteeing perfect reconstruction.

Computer-aided detection (CAD) schemes are decision making support tools, useful to overcome limitations of problematic clinical procedures. Trans-rectal ultrasound image based CAD would be extremely important to support prostate cancer diagnosis. An effective approach to realize a CAD scheme for this purpose is described in this work, employing a multi-feature kernel classification model based on generalized discriminant analysis. The mutual information of feature value and tissue pathological state is used to select features essential for tissue characterization. System-dependent effects are reduced through predictive deconvolution of the acquired radio-frequency signals. A clinical study, performed on ground truth images from biopsy findings, provides a comparison of the classification model applied before and after deconvolution, showing in the latter case a significant gain in accuracy and area under the receiver operating characteristic curve.

In this work we provide an accurate analytical estimation of the frame bounds for frequency warping operators of arbitrary shaped non-smooth warping maps. We deal with both the nonuniform Fourier transform approximation and the aliasing suppressed form of the frequency warping operator. The estimation procedure is obtained by using an analytical model of the aliasing operators which has been previously introduced. The provided estimations can be used as design formulas for the parameters, such as degree of smoothness and redundancy, involved in the definition of the frequency warping operators.

In this work we introduce an analytical characterization of the frequency warping operator of arbitrary shaped non-smooth warping maps. The transformation matrix is decomposed in two additive terms: the first term represents its nonuniform Fourier transform approximation while the second term is imposed for aliasing suppression. The first transformation is known to be analytically characterized and fast computable by an interpolation approach. For the second transformation an analytical representation is introduced which allows a fast computation and a simple design. Finally, an example of a potential application is shown.

Biomedical ultrasound (US) image quality is limited due to the blurring of tissue reflectivity introduced by the transducer point spread function (PSF). Deconvolution techniques can be used to eliminate this effect and to obtain the pure tissue response. In this paper we propose a new enhancement method based on recursive least squares (RLS) adaptive filtering. The method handles the spatial variability of the PSF by means of blind homomorphic characterization of the transducer. The results show that the RLS adaptation increases contrast and resolution by eliminating the spectral shaping of the backscattered echo and thus enhancing the diagnostic capability of US images.

In this work we describe a mathematical operator for the fast calculation of frequency warping. Such invertible transformation can be used to reshape the tiling of time- frequency analysis in a flexible way. In our approach the warping is computed by splitting the transformation operator in two additive terms: the first term represents its Nonuniform Fourier Transform approximation while the second term is imposed for aliasing suppression. The first transformation is known to be analytically characterized and rapidly computable by interpolation approaches. An analytical characterization of the second transformation operator is proposed for arbitrary shaped non-smooth warping maps commonly used for signal processing applications.

In this paper we present a deconvolution technique for ultrasound images based on a Maximum Likelihood (ML) estimation procedure. In our approach the ultrasonic radio-frequency (RF) signal is considered as a sequence affected by intersymbol interference (ISI) and AWGN noise. In order to reduce the computational cost, the estimation is performed with a reduced-state Viterbi algorithm. The channel effect is estimated in two different ways: either measuring the transducer response with an experimental setting or with blind homomorphic techniques. We verify an image quality enhancement with respect to different metrics. Extensive tests are made to estimate the quantization alphabet that gives the best performances.

In this work we present a fast and accurate algorithm to compute frequency warping of arbitrary shaped maps. In contrast to the common Laguerre approach, frequency warping is represented by a matrix of truncated finite dimensions. The transformation matrix is decomposed in two additive terms: the first term represents its nonuniform Fourier transform approximation while the second term is imposed for aliasing suppression. Both matrices are approximated with a least square approach according to a suitable set of vectors. The cardinality of this set is shown to be nearly proportional to the logarithm of the matrix dimension. Finally, trade-off aspects between algorithm complexity and performance are discussed

A new technique has been implemented with the aim of providing an adaptive tool for the generation of computational meshes for 3-D semiconductor device simulation. The core of the proposed wavelet-based adaptive method (WAM) is a refinement algorithm based on the wavelet transform, which gives an estimation of solution regularity and progressively adapts the grid by increasing the resolution only in regions where the important physical phenomena take place. WAM is inserted into a validation tool that provides the interfacing filters with both the solver and a meshing engine. Additional features assure the quality of the generated meshes by increasing the selectivity properties of the refinement tool, preventing numerical instabilities or artifacts, such as abrupt variations of I-V curves between successive bias points during quasi-stationary simulations. Simulation results are provided, which show the effectiveness of the proposed approach as a means to guide the automatic refinement of the discretization grid, preserving accuracy with negligible computational overhead and no skilled control from the user.