The structural health monitoring (SHM) of civil structures and infrastructures is becoming a crucial issue in our smart and hyper-connected age. Due to structural aging and to unexpected loading conditions, partially linked to extreme events caused by the climate change, reliable and real-time SHM schemes are currently facing a burst in development and applications. In this work, we propose a procedure that relies upon a surrogate modeling scheme based on a multi-fidelity (MF) deep neural network (DNN), which has been conceived to sense and identify a structural damage under operational (and possibly environmental) variability. By exploiting the sensor recordings from a densely deployed network within a fully stochastic framework, the MF-DNN model is adopted to feed a Markov chain Monte Carlo sampling procedure and update the probability distribution of the structural state, conditioned on noisy observations. As information regarding the health of real structures is usually rather limited, the datasets to train the MF-DNN are generated with physical (e.g. finite element) models: high-fidelity (HF) and low-fidelity (LF) models are adopted to simulate the structural response under the mentioned varying conditions, respectively in the presence or absence of a structural damage. As far as the architecture of the DNN is concerned, the MF approach is obtained by merging a fully connected LF-DNN and a long short-term memory HF-DNN. The LF-DNN mimics the output of the sensor network in the undamaged condition, while the HF-DNN is exploited to improve the LF model and appropriately catch the structural response in the presence of a pre-defined set of damaged patterns. Thanks to the adaptive enrichment of the LF signals carried out by the MF-DNN, the proposed model updating strategy is reported capable of locating (and possibly quantifying) a damage event.

The Polar H10 is a low-cost wearable with a heart rate monitor and tri-axial accelerometer with potential for many applications. While the device’s heart rate monitor has been widely studied, there is no research validating the accelerometer specifically. The purpose of this study was to conduct a validation of the Polar H10 accelerometer to establish static and dynamic validity during a sports-based task. Static validity was determined by computing the relative error when using a level guide to hold each axis of the Polar H10 against gravity. Fifteen healthy adults (8F/7F) participated in sports-based tasks while wearing the Polar H10 (Polar Electro, Poland) and a comparison device, the MetaMotionR inertial measurement unit (MbientLab Inc., USA). Dynamic validity was characterized using Pearson’s correlation coefficient and root mean square error (RMSE). Additionally, common features in human activity recognition (mean magnitude, root mean square, power, and signal magnitude area) were computed in 2s windows and compared via RMSE and Wilcoxon rank sum tests. When held against gravity, the Polar H10 had relative errors ranging from 2.620% to 4.288%, suggesting high static validity. During sports-based tasks, the accelerometers had correlations between 0.888 and 0.954, indicating sufficient concurrent validity for all axes, as well as acceleration magnitude. The differences in acceleration features were minimal (RMSE for mean, root mean square, power, and signal magnitude were 0.003G, 0.004G, 0.112, and 0.017G, respectively), but all reached significance (P<.001). These results provide evidence for the use of the Polar H10 accelerometer to measure movement during sport-like activities.

Electronic nose (EN) is a technology for fast and adequate identification of complex gaseous mixtures in many vitally important areas. The idea is to use a small set of low-selective sensors with overlapping selectivity profiles instead of a huge number of specific sensors. The response of such an array of low-selective sensors is a chemical image (CI) of a gaseous mixture. The ability of an EN system to distinguish different mixtures is defined by its ability to produce unique CI. The latter is defined, first of all, by the sensors adsorption properties.

We propose an approach to increase the versatility of low-selective sensor arrays by using the virtual sensors approach. By “virtual sensor” in this case we mean a sensor able to change its adsorption properties in conditions of illumination. This allows multiple responses of the same physical sensor for an analyte to be obtained - a set of these responses for different lighting conditions is a data set for forming a unique chemical image of the analyte.

Within this scenario, we were able to successfully distinguish between homologous alcohols (methyl, ethyl, and isopropyl alcohols) using organic-inorganic nanostructured sensitive layers based on ZnO nanoparticles and phthalocyanines. It must be emphasized that this is a rather difficult analytical problem, since alcohols have both similar molecular weights and physicochemical properties.

Thus, we have demonstrated the possibility of reconfiguring multisensor arrays by controlling the adsorption properties of sensors by illuminating them in different regions of the spectrum (usually UV-VIS-IR). This is a step forward in the development of multi-purpose universal EN systems, the selectivity profile of which can be controlled by external influences without changing the composition of the base sensor array.

Nucleic acids and corresponding mutations are crucial in the diagnosis of a broad range of genetic diseases such as cystic fibrosis.¹ This is the most common and fatal autosomal recessive genetic disease in EU countries and the USA.²

Over the years, different electrochemical sensors have been developed for the detection of nucleic acids to meet the demand for point-of-care diagnostics. However, these technologies have different drawbacks such as: i) low limit of detection ii) need of a well-defined orientation of DNA strands on the electrode surface iii) need of a trained person and iv) time-consuming sample preparation.³

This work contributes to the diagnosis of cystic fibrosis via the development of a novel photoelectrochemical biosensor for the detection of its most common DNA mutation (i.e. F508, accounting for approximately 70% of all mutations) in the gene cystic fibrosis transmembrane conductance regulator.

This groundbreaking platform exploits a sandwich assay combining i) photosensitizers, that produce singlet oxygen (¹O₂), as a label in the detection strategy, ii) a redox reporter (i.e. hydroquinone) and iii) magnetic beads, used to attract the synthetic DNA sequences close to the electrode surface, enhancing the sensitivity.⁴ Since the signal is only triggered by light, a main advantage of our sensor is the clear distinction between signal and background by turning on/off the light source.

Using this platform, we explore the effect of different buffers on the resulting photocurrent and we demonstrate the specific detection of the desired target (F508) while avoiding unwanted interactions with random sequences.

References: : ¹Wei, F. et al., Pediatr. Res., 2010, 67, 5, 458–468. ²Scotet, V. et al., Genes, 2020, 11, 589. ³Rashid, J. I. A. et al., Sens. Bio-Sens. Res., 2017, 16, 19–31. ⁴Shanmugam, S. T. et al., Biosens. Bioelectron., 2022, 195.

Flexible tactile sensors are foreseen to be extensively used soon in wearable devices. Various materials in flexible sensor fabrication offer sensing properties with multiple capabilities. The materials, including nanocomposites, have a crucial research area for flexible tactile sensors. While the nanocomposites' electrical properties mainly depend on the nanofillers, the mechanical properties are determined by the polymer component. Carbon nanotubes are one of the most promising materials among nanofillers due to their high electrical conductivity, thermal stability, and durability. However, carbon nanotubes should be processed to increase the binding capacity with the polymer structure. In this study, the nanocomposite used for the sensor manufacturing consists of acid-functionalized carbon nanotubes and sodium alginate as the nanofiller and the polymer material, respectively. The sensors were cross-linked using calcium chloride, and glycerin was involved in the sensor fabrication to check the effect on the sensing and flexibility. Also, it is critical to note that sodium alginate and glycerin are biocompatible and biodegradable substances. On the scope of the study, the impedance changes of the fabricated tactile sensors were examined in the 100 Hz – 10 MHz frequency range and the equivalent circuits of the sensors were created. Besides, the impedance changes were obtained when the alternating forces were applied to the sensors. The results show that the frequency responses of the sensors differ from each other in different frequency ranges. Also, each sensor has different sensing mechanisms in specific frequency ranges, and the sensor, including glycerin, has higher flexibility, but less sensitivity.

Dangerous air pollutants, like ozone (O3) and nitrogen dioxide (NO2), pose major health risks. Technological advancements in low-cost gas sensors equipped with deep learning algorithms enable monitoring of such gases on a large scale. In our application, O3 and NO2 concentrations are predicted using a Gated Recurrent Unit (GRU) neural network. However, when using neural networks, there comes a challenge to make particular predictions interpretable for humans. Our research aims to address this difficulty by adopting two explainable artificial intelligence (XAI) methodologies for gas sensors to understand the reasoning behind a model’s predictions while also facilitating the characterization of sensor behavior.

The first technique quantifies the contribution of each input feature to the predictions using the Shapley Additive Explanations (SHAP) method. The features with the highest scores are considered the most important and vice-versa. Our analysis aided in dropping off the features with low scores from the model, resulting in less memory and computation resources, thus, making the model more energy efficient. It also helped enhance the quality of our sensor material. i.e., one of our core features initially scored low, indicating that the corresponding sensor’s material was under scrutiny; finally, after material improvements, that feature became one of the most impactful features.

The second approach, network dissection, tries to explain the inner-working of the network by examining the hidden state activations of each GRU unit to comprehend certain (unexpected) predictions. Our analysis demonstrated which GRU units respond to O3 and which to NO2. It also showed that for higher O3 concentrations, NO2 is masked by O3, which is also consistent with the underlying physics of the sensing material. Understanding the behavior of the dissected blocks of a neural network also helps choose an optimal number of hyperparameters for a leaner and more robust model.

Data-driven formulations are currently developed and can result extremely helpful to deal with the complexity of the multi-physics governing the response of micro-electro-mechanical systems (MEMS) to the external stimuli. Such devices are in fact characterized by a hierarchy of length- and time-scales, which are difficult to fully account for in a purely model-based approach [1]. In this work, we specifically refer to a (single-axis) Lorentz force micro-magnetometer designed for navigation purposes. Due to an alternating current flowing in a slender mechanical part (beam) and featuring an ad-hoc set frequency, the micro-system is driven into resonance so that its sensitivity to the magnetic field gets improved. A reduced-order physical model was formerly developed for the aforementioned movable part of the device; this model was then used to feed and speed up a multi-physics and multi-objective topology optimization procedure, aiming to design a robust and performing magnetometer. The stochastic effects, which are responsible for the scattering in the experimental data at the microscale [2], were not accounted for in such a model-based approach. A recently proposed formulation, see [3], is here discussed and further extended to allow for such stochastic effects. The proposed multi-scale deep learning approach features: at the material scale, a deep neural network adopted to learn the scattering in the mechanical properties of polysilicon induced by its morphology; at the device scale, a multi-input deep neural network adopted to learn the imperfection-sensitive geometric features of the movable part of the magnetometer. The two data-driven models adopted at the material and device length scales are linked through the physical model proposed in [1] to provide a kind of hybrid solution to the problem. Results relevant to different neural network architectures are discussed, along with a proposal to frame the approach as a multi-fidelity, uncertainty quantification procedure.

[1] S. Mariani, A. Ghisi, A. Corigliano, R. Martini, B. Simoni. Two-scale simulation of drop-induced failure of polysilicon MEMS sensors. Sensors, 11, pp. 4972-4989, 2011.

[2] M. Bagherinia, S. Mariani. Stochastic effects on the dynamics of the resonant structure of a Lorentz force MEMS magnetometer. Actuators,8, 36, 2019.

[3] S. Mariani, J.P. Quesada Molina. A two-scale multi-physics deep learning model for smart MEMS sensors. Journal of Materials Science and Chemical Engineering, 9, pp. 41-52, 2021.

In recent years model-based fault techniques become really popular due to reducing calculation cost. Bayesian Network and Two Stage Kalman Filter based methods have recently become quite popular due to their robustness. In this paper, model-based fault diagnosis method is presented that uses Bayesian Network and Two Stage Kalman Filter(TSKF) together to determine the sensor faults robustly in the Unmanned Aerial Vehicle (UAV) system. In the fault detection algorithm, six residual values are calculated. The threshold values of all the calculated residuals are determined using synthetic dataset. Depending on whether the residuals exceed the threshold value or not, the fault generation coefficients in the Bayesian Network are also dynamically updated to provide precise information regarding which sensor has a fault. By using these two approaches together, the robustness of the detection of the fault in the sensor improved. For demonstrating the behavior of the proposed method, numerical simulations are performed in MATLAB/Simulink^TM environment. The results show that the proposed method is capable of detecting the faults more robustly.

With the invention of GPS and related technologies outdoor positional system is possible with great accuracy. However, there is still a need for efficient, reliable and less expensive technology for indoor navigation. There are lots of techniques which are used for indoor navigation such as acoustic, wifi-based, proximity-based,infraded systems and SLAM algorithms. In this study, it was tried to obtain an accurate position estimation by combining the acceleration and gyroscope data obtained from the Marvelmind Beacon sensor, one of the ultrasonic sensors, and the raw distance data with the help of the Extended Kalman Filter (EKF). Initially, a position estimation is obtained using the Recursive Least Square(RLS) method with a trilateration algorithm. This solution is used as a starting point for RLS. Here, the first solution point is updated as the initial solution for each distance data, and the result calculated by the RLS method is updated as the next solution.This approach enables the distance measurement and position estimation to be executed simultaneously and it avoids the unnecessary waiting time and speeds up the positioning estimation. After that, this position estimation is fused with the acceleration and gyroscope data which collected from the Marvelmind sensor. In this respect, this designed algorithm is similar to the tightly coupled EKF structure that produces its own GPS solution. In order to test the designed algorithm, data was collected from the Marvelmind sensor. While collecting these data, four stable and one moving Marvelmind sensor were used. Data collection was carried out via the serial port on the sensors with the help of the ROS platform. The collected data was first obtained in txt format and then converted to MATLAB format. The tests of the designed algorithm were carried out with these data. As a result of the tests, it has been observed that, this tightly coupled indoor EKF structure created for indoor navigation gives more accurate results.

Automated vibration-based structural health monitoring (SHM) strategies have been recently proven promising in the presence of aging and material deterioration threatening the structural safety of civil structures. Within such a framework, ensuring high quality and informative data is a critical aspect, highly dependent on the deployment of the sensors in the network and on their capability to provide damage-sensitive features to be exploited. This paper presents a novel data-driven approach to the optimal sensor placement, devised to identify sensor locations that maximize the information effectiveness for SHM purposes. The optimization of the sensor network is addressed by means of a deep neural network (DNN) equipped with an attention mechanism, a state-of-the-art technique in natural language processing useful to focus on a limited number of important components in the information stream. The trained attention mechanism eventually allows to quantify the relevance of each sensor in terms of the so-called attention scores, and therefore enables to identify the most useful input channels to solve the relevant downstream SHM task. With reference to the damage localization task, framed here as a classification problem handling a set of predefined damage scenarios, the DNN is trained to locate damage on labeled data, that have been formerly simulated to emulate the effects of damage under different operational conditions. The capabilities of the proposed method are demonstrated by referring to an eight-story shear building, characterized by damage states of unknown severity and possibly located at any story.