The venture of autonomous driving technologies represents one of the most important accomplishments to develop the future age transportation. In this regard, legislative frameworks defined in United States and in Europe clearly explicit the primary role of safety, nonetheless regulations concerning the management of unpredictable events are still unclear. Along public roads, many factors can generate unexpected crash instances, which sometimes can be inevitable. As a result, the implementation of quantitative methods devoted to enable automated driving systems to manage hazardous conditions or potential 'dilemma scenarios' is essential. In this view, a full-scale crash database based upon guidelines defined by American Association of State Highway and Transportation Offcials (AASHTO) is set up to develop a global understanding about run-o_ road impacts characteristics. This range of established collisions is simulated through finite element (FE) models of vehicles and roadside hardware, supplied by Center for Collision Safety and Analysis (CCSA) and properly tuned to perform in a consistent way all the subcases using LS-DYNA. Afterwards, results are postprocessed to investigate both critical issues and the success of the simulations. Besides, time histories of linear accelerations and angular rates are used to determine the effectiveness of the safety features by the use of Test Risk Assessment Program (TRAP). Furthermore, in order to highlight the sensitivity of the crash outcomes with respect to the inputs, speed up the storage procedure and eventually expanding the database with additional cases, a standalone software is set up within Matlab's environment. Finally, with the aim of estimating the occupant risk of lesion, identifying trends in the effects of the impact parameters and then managing dilemma situations, Injury Indices are examined to develop a supervised machine learning algorithm to produce Impact Severity Envelopes for each crash scenario. Moreover, a classification algorithm is developed to generalize the safety performances of the roadside hardware and to prove the goodness of the envelopes in estimating the impact consequences.
L'avvento delle tecnologie per la guida autonoma rappresenta uno dei più importanti traguardi per lo sviluppo dei trasporti futuri. In tale contesto, il ruolo chiave della sicurezza è chiaramente espresso dalle regolamentazioni americane ed europee, tuttavia i criteri riguardanti la gestione di eventi imprevisti devono essere ancora affinati. Lungo le strade pubbliche numerose istanze possono porre i presupposti per incidenti inaspettati, i quali talvolta risultano inevitabili. Di conseguenza, per i veicoli autonomi risulta essenziale l'implementazione di metodi quantitativi volti alla gestione di scenari di pericolo o situazioni di dilemma. Con questa prospettiva, seguendo le linee guida date dal American of State Highway and Transportation Offcials (AASHTO), viene imbastita una banca dati contenente simulazioni ad elementi finiti di impatti stradali tra diversi modelli di veicoli e barriere. Questi ultimi, forniti dal Center for Collision Safety and Analysis (CCSA), sono tarati appositamente per rappresentare in modo consistente i vari casi tramite LS-DYNA. Successivamente, i risultati vengono analizzati per verificare la buona resa delle simulazioni ed eventuali aspetti critici. Parallelamente, le storie temporali di accelerazioni e ratei angolari vengono processate per determinare l'efficacia delle infrastrutture tramite il Test Risk Assessment Program (TRAP). Inoltre, all'interno dell'ambiente Matlab viene implementato un software indipendente volto ad evidenziare la sensitività dei risultati rispetto ai parametri degli impatti, a velocizzare il salvataggio dei dati e ad espandere eventualmente la banca dati con casi aggiuntivi. Infine, per stimare il rischio di lesione per gli occupanti, identificare possibili andamenti negli effetti dei parametri di collisione e quindi gestire situazioni di incertezza, partendo dallo studio degli Indici di Lesione viene sviluppato un algoritmo di apprendimento automatico dedicato a generare Inviluppi di Severità di Impatto per ogni scenario. In aggiunta, viene proposto un metodo di classificazione per generalizzare le prestazioni di sicurezza delle barriere e verificare le stime date dagli inviluppi.
Evaluation of run-off road impacts for autonomous vehicles
LA PORTA, ANDREA
2018/2019
Abstract
The venture of autonomous driving technologies represents one of the most important accomplishments to develop the future age transportation. In this regard, legislative frameworks defined in United States and in Europe clearly explicit the primary role of safety, nonetheless regulations concerning the management of unpredictable events are still unclear. Along public roads, many factors can generate unexpected crash instances, which sometimes can be inevitable. As a result, the implementation of quantitative methods devoted to enable automated driving systems to manage hazardous conditions or potential 'dilemma scenarios' is essential. In this view, a full-scale crash database based upon guidelines defined by American Association of State Highway and Transportation Offcials (AASHTO) is set up to develop a global understanding about run-o_ road impacts characteristics. This range of established collisions is simulated through finite element (FE) models of vehicles and roadside hardware, supplied by Center for Collision Safety and Analysis (CCSA) and properly tuned to perform in a consistent way all the subcases using LS-DYNA. Afterwards, results are postprocessed to investigate both critical issues and the success of the simulations. Besides, time histories of linear accelerations and angular rates are used to determine the effectiveness of the safety features by the use of Test Risk Assessment Program (TRAP). Furthermore, in order to highlight the sensitivity of the crash outcomes with respect to the inputs, speed up the storage procedure and eventually expanding the database with additional cases, a standalone software is set up within Matlab's environment. Finally, with the aim of estimating the occupant risk of lesion, identifying trends in the effects of the impact parameters and then managing dilemma situations, Injury Indices are examined to develop a supervised machine learning algorithm to produce Impact Severity Envelopes for each crash scenario. Moreover, a classification algorithm is developed to generalize the safety performances of the roadside hardware and to prove the goodness of the envelopes in estimating the impact consequences.| File | Dimensione | Formato | |
|---|---|---|---|
|
2019_07_La_Porta.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
67.11 MB
Formato
Adobe PDF
|
67.11 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/10589/149413