This thesis work is inserted in a bigger research project proposed by the Italian Space Agency named MARS-PRE, which is intended to identify chemical and functional biomarkers able to characterize the physiological adaptations of different systems to the microgravity condition experienced in Space. In particular, microgravity leads to the unloading of bones and muscles, responsible for bones demineralization and loss of muscle mass and strength. To counteract the musculoskeletal deconditioning, astronauts follow a specific exercise program pre, during and post-flight. On board the International Space Station (ISS) there are different devices that allow astronauts to train: a treadmill and a cycloergometer are used for aerobic exercises, while resistive exercises to maintain muscle mass and strength are performed with the Advanced Resistive Exercise Device (ARED). It simulates the use of free weights in microgravity by generating a constant load, which can be changed from 0 to 272.5 Kg, and permits to perform exercises such as normal squat, wide stance squat and deadlift. ARED works by adding an external load through a bar positioned on the astronaut’s shoulders, which can cause discomfort and back pain, especially if exercises are wrongly executed. Currently the training sessions are not supervised, therefore if astronauts do not perform the exercises correctly, in the long run it could lead to serious repercussions on their health. This brings to the current work, whose aim is to design a wearable system able to monitor astronauts training performed with ARED. The system should be composed of a minimum set of inertial sensors easy to set up, cheap, and suitable in microgravity conditions. The sensors should provide acceleration and angular velocity signals representative of the exercise executions, used to compute a set of relevant features to give as input to a classifier based on machine learning algorithms. The classifier should be able to distinguish between correct and wrong executions of the target exercises and provide real-time corrective feedback. The first operational step consisted in building a system of six wearable sensors able to simulate microgravity on ground. Each sensory unit is composed of an IMU, a microcontroller, a battery, and a Bluetooth module, electrically connected though a Printed Circuit Board and inserted in a 3D-printed case. Microcontrollers have been programmed to interface with sensors and stream data toward a Graphical User Interface (GUI) that is able to visualize and record them. Microgravity simulation, used to obtain data on-ground that are representative of the ones that would be acquired on the ISS, has been accomplished in the GUI: these data are necessary to build the classifier for the astronauts training. The system has been validated using inertial sensors from Xsens as ground truth: high values of correlation and coherence between signals coming from the two sensors have been accomplished. For what concerns the classifier, data coming from more than 20 subjects have been used to train and test different machine learning algorithms and find the most suitable for the application. The only kind of exercise analysed was the normal squat performed with loads: the Multi-Layer Perceptron model was selected as it was able to classify squat executions into six classes of error, with an accuracy higher than 90% on both the train and test set. Then, to achieve a real-time classification, an algorithm for repetition extraction has been implemented in the GUI during data recording, together with a features computation step: the chosen classifier has been added, and the classification results have been visually presented to the user as an indication to guide its training. A future development, and the purpose of the MARS-PRE project, would be to create individual classifiers for the astronauts that could be trained directly in-flight, or from data acquired with ARED in the pre-flight phase.
Questo lavoro di tesi è inserito in un più ampio progetto di ricerca proposto dall'Agenzia Spaziale Italiana denominato MARS-PRE, che ha lo scopo di identificare biomarcatori chimici e funzionali in grado di caratterizzare gli adattamenti fisiologici di diversi sistemi alla condizione di microgravità sperimentata nello Spazio. In particolare, la microgravità comporta una riduzione del carico gravante su ossa e muscoli, causando una perdita di minerali nelle ossa e di massa e forza nei muscoli. Per contrastare il decondizionamento muscolo-scheletrico, gli astronauti seguono uno specifico programma di esercizi prima, durante e dopo il volo. A bordo della Stazione Spaziale Internazionale (ISS) sono presenti diversi dispositivi che consentono agli astronauti di allenarsi: un treadmill e un cicloergometro vengono utilizzati per gli esercizi aerobici, mentre gli esercizi resistivi per mantenere la massa muscolare vengono eseguiti con l'Advanced Resistive Exercise Device (ARED). Esso simula l'utilizzo di pesi liberi in microgravità generando un carico costante, che può essere modificato da 0 a 272,5 Kg, e permette di eseguire esercizi come squat normale, wide-stance squat e deadlift. ARED funziona aggiungendo un carico esterno attraverso una barra posizionata sulle spalle dell'astronauta; ciò può risultare fastidioso e causare mal di schiena, soprattutto se gli esercizi vengono eseguiti in modo errato. Attualmente le sessioni di allenamento non sono supervisionate, quindi in caso gli astronauti non eseguano correttamente gli esercizi, a lungo andare questo potrebbe portare a gravi ripercussioni sulla loro salute. Per tale motivo il presente lavoro ha l’obiettivo di progettare un sistema indossabile in grado di monitorare l'addestramento degli astronauti effettuato con ARED. Il sistema dovrebbe essere composto da un set minimo di sensori inerziali facili da configurare, economici e adatti a condizioni di microgravità. I sensori dovrebbero fornire segnali di accelerazione e velocità angolare rappresentativi delle esecuzioni degli esercizi, utilizzati per calcolare una serie di caratteristiche rilevanti da fornire in ingresso a un classificatore basato su algoritmi di machine learning. Il classificatore dovrebbe essere in grado di distinguere tra esecuzioni corrette ed errate degli esercizi target e fornire feedback correttivi in tempo reale. Il primo obiettivo del progetto ha riguardato la realizzazione hardware di un sistema di sensori indossabili in grado di simulare la microgravità a terra. Ogni unità sensoriale è composta da un'IMU, un microcontrollore, una batteria e un modulo Bluetooth, collegati elettricamente tramite un circuito stampato ed inseriti in un contenitore stampato in 3D. I microcontrollori sono stati programmati per interfacciarsi con i sensori e trasmettere i dati verso un’interfaccia grafica (GUI) in grado di visualizzarli e registrarli. Nella GUI è stata realizzata la simulazione della microgravità, utilizzata per ottenere dati a terra rappresentativi di quelli che si acquisirebbero sulla ISS: questi dati sono necessari per costruire il classificatore per allenare gli astronauti. Il sistema è stato validato utilizzando come riferimento dei sensori inerziali dell’azienda Xsens: sono stati raggiunti alti valori di correlazione e coerenza tra i segnali provenienti dai due sensori. Per quanto riguarda il classificatore, i dati provenienti da più di 20 soggetti sono stati utilizzati per addestrare e testare diversi algoritmi di machine learning e trovare quello più adatto all'applicazione. L'unico tipo di esercizio analizzato è stato lo squat normale eseguito con i pesi: è stato scelto il modello Multi-Layer Perceptron in quanto in grado di classificare le esecuzioni di squat in sei classi di errore, con una precisione superiore al 90% sia sui dati di allenamento che di test. Quindi, per ottenere una classificazione in tempo reale, è stato implementato un algoritmo per l'estrazione delle ripetizioni nella GUI, insieme ad una parte di calcolo delle caratteristiche: è stato aggiunto il classificatore scelto, e i risultati della classificazione sono stati presentati visivamente all'utente come indicazioni per guidarne l’allenamento. Uno sviluppo futuro, e lo scopo del progetto MARS-PRE, sarebbe quello di creare classificatori individuali per gli astronauti che possano essere allenati direttamente in volo, o da dati acquisiti con ARED nella fase di pre-volo.
Development of a wearable, IMU based, system for real-time monitoring of astronauts training during long term space missions
Giani, Laura
2019/2020
Abstract
This thesis work is inserted in a bigger research project proposed by the Italian Space Agency named MARS-PRE, which is intended to identify chemical and functional biomarkers able to characterize the physiological adaptations of different systems to the microgravity condition experienced in Space. In particular, microgravity leads to the unloading of bones and muscles, responsible for bones demineralization and loss of muscle mass and strength. To counteract the musculoskeletal deconditioning, astronauts follow a specific exercise program pre, during and post-flight. On board the International Space Station (ISS) there are different devices that allow astronauts to train: a treadmill and a cycloergometer are used for aerobic exercises, while resistive exercises to maintain muscle mass and strength are performed with the Advanced Resistive Exercise Device (ARED). It simulates the use of free weights in microgravity by generating a constant load, which can be changed from 0 to 272.5 Kg, and permits to perform exercises such as normal squat, wide stance squat and deadlift. ARED works by adding an external load through a bar positioned on the astronaut’s shoulders, which can cause discomfort and back pain, especially if exercises are wrongly executed. Currently the training sessions are not supervised, therefore if astronauts do not perform the exercises correctly, in the long run it could lead to serious repercussions on their health. This brings to the current work, whose aim is to design a wearable system able to monitor astronauts training performed with ARED. The system should be composed of a minimum set of inertial sensors easy to set up, cheap, and suitable in microgravity conditions. The sensors should provide acceleration and angular velocity signals representative of the exercise executions, used to compute a set of relevant features to give as input to a classifier based on machine learning algorithms. The classifier should be able to distinguish between correct and wrong executions of the target exercises and provide real-time corrective feedback. The first operational step consisted in building a system of six wearable sensors able to simulate microgravity on ground. Each sensory unit is composed of an IMU, a microcontroller, a battery, and a Bluetooth module, electrically connected though a Printed Circuit Board and inserted in a 3D-printed case. Microcontrollers have been programmed to interface with sensors and stream data toward a Graphical User Interface (GUI) that is able to visualize and record them. Microgravity simulation, used to obtain data on-ground that are representative of the ones that would be acquired on the ISS, has been accomplished in the GUI: these data are necessary to build the classifier for the astronauts training. The system has been validated using inertial sensors from Xsens as ground truth: high values of correlation and coherence between signals coming from the two sensors have been accomplished. For what concerns the classifier, data coming from more than 20 subjects have been used to train and test different machine learning algorithms and find the most suitable for the application. The only kind of exercise analysed was the normal squat performed with loads: the Multi-Layer Perceptron model was selected as it was able to classify squat executions into six classes of error, with an accuracy higher than 90% on both the train and test set. Then, to achieve a real-time classification, an algorithm for repetition extraction has been implemented in the GUI during data recording, together with a features computation step: the chosen classifier has been added, and the classification results have been visually presented to the user as an indication to guide its training. A future development, and the purpose of the MARS-PRE project, would be to create individual classifiers for the astronauts that could be trained directly in-flight, or from data acquired with ARED in the pre-flight phase.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Magistrale_Giani_Laura.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
35.81 MB
Formato
Adobe PDF
|
35.81 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/174150