Leader: Alessio Tamburrano (Sapienza); Other collaborator(s): Maggi, Noale (CNR), Ferrarese, Bellelli, Franchi (UNIMIB), Pilotto, Brattico, De Luca, Gesualdo, Solfrizzi (UNIBA), RTD-A UNIBA
Investigation of new cost-effective solutions based on novel non-invasive highly sensitive physical sensors integrated on textiles and porous structures by the deposition of graphene nanoplatelets based inks through screen printing techniques or infiltration methods. Guaranteeing perspiration, washability and comfort the piezoresistive and conducting properties of nanostructured coatings will be exploited, using electromechanical tests and electrical resistance tomography techniques with dedicated machine learning algorithms for strain/pressure patterns acquisition and reconstruction. The idea is to realise a sensor system capable of managing the strain and pressure measurement over large areas, and to be included for example in mattress pads or bedspreads. This system is meant for elderly adults and bedridden people, keeping track of the bed occupancy and improving the fall prevention, the nocturnal sleep-wake rhythms and avoiding bedsores by identifying the bodyweight distribution.
Brief description of the activities and of the intermediate results
The partners of Task 5.4 focused on developing and utilizing new piezoresistive coatings based on graphene-modified paints/inks as pressure sensors, with a particular emphasis on their application in healthcare. These systems aim to measure deformation and pressure on large surfaces, such as mattress covers or blankets, with the ultimate goal of monitoring the elderly, especially during bed rest, to mitigate issues like bedsores and improve fall prevention and nighttime sleep-wake patterns. Three different configuration have been identified.
The most promising configuration will serve as the basis for developing the sensorized mat/bedspread prototype. The partners started by developing a smaller system to explore its potential, and successfully fabricated a small piezoresistive foam mat. This prototype underwent morphological, electrical, and piezoresistive characterization through various experimental investigations, including scanning electron microscopy, voltamperometric measurements, and electromechanical tests. ERT was implemented to identify different deformation scenarios caused by fingertip pressure on the soft mat sensor. Data processing was performed using MATLAB scripts, with algorithms (Support Vector Machine, Artificial Neural Network, K-Nearest Neighbors, and Decision Tree) trained and validated on experimental measurements.
Please see the next reporting period.
Building on the experience gained, researchers have nearly completed the development of a larger sensorized surface (80 cm x 50 cm), covered with a screen-printed fitted sheet using piezoresistive graphene-based ink. This continuous piezoresistive coating has been successfully tested, enabling detailed pressure mapping through Electrical Resistance Tomography (ERT) and machine learning (ML) algorithms. The system has demonstrated its capability to monitor pressure and recognize sleeping postures with high accuracy, achieving 96.3% accuracy in detecting compressed regions and 94.4% in posture classification.The next phase will focus on studying the robustness of the data collected from this prototype, ensuring its reliability. To obtain more realistic data, partners are evaluating the use of a neonatal dummy, improving upon the previous tests that used wooden blocks as surrogates for body pressure. Upon completion of this phase, the system will be scaled to a larger configuration, such as a standard bed, to extend its application for monitoring bedridden elderly patients.In parallel, the unit is working on assessing the piezoresistive coefficients of the coatings, which will be used to build a mathematical model for Finite Element Method (FEM) simulations. Moreover, an alternative measurement system based on impedance tomography rather than resistance tomography is also under development. This ongoing work will help optimize the sensor's design by accurately simulating its mechanical and electrical behaviour under different conditions, ultimately improving the performance of future, larger-scale implementations.
The first prototype of a sensorized mattress, measuring 80 x 50 cm, was successfully completed and tested. This work was presented at the IEEE Sensors 2024 international conference held in Kobe, Japan, from October 20–23. Following these results, piezoresistive coefficients were experimentally obtained, enabling the development of a finite element method (FEM) simulation model to replicate the piezoresistive characteristics of a sensorized crib-size mattress. This simulation model also incorporated the application of the electrical resistance tomography (ERT)-inspired methodology. The data generated through these simulations were used as training inputs for a machine learning (ML) algorithm, with the primary objective of achieving precise pressure detection. To simulate real-world conditions, a neonatal dummy weighing approximately 5 kg was used to evaluate the mattress's ability to differentiate between various sleeping positions. Special attention was given to detecting potentially dangerous positions for a baby. The tests were conducted both with and without an additional blanket on the mattress to verify system functionality under different conditions. The ML classification algorithm achieved an accuracy of 91% without a blanket and 86% with the blanket. Additionally, system robustness was assessed by simulating electrode malfunctions, which yielded strong performance with an accuracy of 89%. Furthermore, an alternative measurement system based on impedance tomography, rather than resistance tomography, is under development. This system has been installed and configured on the existing prototype and will be incorporated into the final one. Currently, work is underway to realize the final prototype. Upon completion, the sensing system will be scaled to larger configurations, such as standard-sized beds, to extend its applications for monitoring bedridden elderly patients.
Building on the progress described in the December 31, 2024 update, the first quarter of 2025 (January–March) focused on consolidating and extending the sensing and classification capabilities of the system. The previously developed approach for sleep position recognition using a sensorized 80 × 50 cm mattress was further refined, with algorithmic improvements informed by new experimental data. As a result, classification accuracy was significantly enhanced, reaching 99.4% without a blanket, 94.8% with a blanket, and maintaining strong performance even in the presence of electrode faults. A hierarchical classification strategy was also introduced, improving generalization and enabling reliable recognition of 30 distinct sleep postures, including potentially hazardous ones. On the hardware side, new 32-channel measurements were conducted on the crib-sized prototype using the recently developed Electrical Impedance Tomography (EIT) acquisition system. This marked the first complete validation of the platform with EIT, which provides improved sensitivity and spatial resolution compared to the previous ERT-inspired configuration. The system was then successfully scaled to a 64-channel setup, in preparation for integration into the full-scale prototype. Regarding the final prototype, the screen-printing process was adapted to produce a sensorized sheet for a full-size mattress (190 × 80 cm), marking a major step toward deployment in elderly patient monitoring. The same coating formulation used in the crib-sized prototype—selected for its optimized piezoresistive response and compatibility with scalable fabrication—was confirmed for the full-scale version, ensuring consistent sensing performance across applications.
During the second quarter of 2025, the project advanced on both dissemination and development fronts. A new publication was finalized and released, presenting the results of the previously developed crib-sized prototype (80 × 50 cm). In parallel, two full-scale piezoresistive sensing blankets (200 × 90 cm) were fabricated. The first prototype is a screen-printed textile using graphene nanoplatelet ink, segmented into three macro-areas and tailored to standard bed dimensions. This version adopts an ERT-inspired acquisition scheme and is equipped with 64 electrodes distributed along its perimeter, interfaced with a 64-channel impedance meter. The new hardware setup enables data acquisition via Electrical Impedance Tomography (EIT) through AC current injection, providing higher spatial resolution and more robust signal quality. These enhancements support the integration of advanced AI-based classification pipelines, with potential for hierarchical or multi-layer architectures to improve performance under real-world conditions. The second prototype features a grid configuration, with orthogonally arranged piezoresistive strips covering the sensing area. Although it has a lower density of sensorized zones, this layout offers several advantages: simplified fabrication, fewer electrical connections, and improved algorithmic interpretability—since signals from individual strips can be analyzed independently to reconstruct pressure distributions. Both prototypes are currently undergoing experimental evaluation and will serve as the basis for scalable, non-invasive pressure monitoring systems aimed at adult posture recognition in bed.
The work mainly focused on consolidating the large-area sensing blankets into a reproducible setup and on aligning the acquisition strategy with the posture-recognition objectives defined in the previous quarter. The current-injection scheme was revised to make it consistent with the segmentation already implemented on the full-scale blanket (macro-areas along the length of the bed) and to keep the acquisition sequence compatible with the 64-electrode perimeter arrangement. The updated sequence was ported to the existing impedance-measurement platform and tested at different frequencies on the 200 × 90 cm screen-printed prototype, confirming that the electrodes and the cabling layout remain stable over the full bed size and that the system can sustain prolonged measurements without critical artefacts. In parallel, the ML algorithms were trained on data from multiple subjects, and the data-collection protocol was broadened to include not only canonical supine and lateral postures, but also intermediate and slightly “imperfect” conditions (rotated lateral positions, non-centred subject, pillow use), with the aim of approaching realistic in-bed behaviour while preserving compatibility with the AI pipelines developed in the previous phase. Experimental tests in this period were completed on the macro-zoned, coating-based blanket, in order to verify signal uniformity, electrode-contact stability, and adherence to the planned injection sequence on a full-scale textile surface. The same measurement procedure is now being prepared for the strip-based/grid prototype, so that the two sensing layouts can be compared on the basis of acquisitions carried out under comparable conditions and processed through the same pipeline. Regarding dissemination, an additional journal paper is currently in preparation/submission, focusing on the system configuration and on the results obtained so far.