The anti-lock braking system (ABS) for bicycles is designed to prevent front wheel lockups and nose-over motions (stoppie) due to excessively aggressive braking, which is one of the most frequent causes of bicycle accidents. Recently the device has been implemented on electric bicycles (e-bikes), a vehicle whose spread is rapidly changing the world of personal transport and which has brought much attention to the problem of cyclist safety. The complexity of the ABS device makes it vulnerable to failures and malfunctions that could prevent proper operation and lead to very serious consequences for the cyclist, such as falls and collisions with other vehicles. For these reasons, the ABS device is called safety-critical and therefore requires a method for identifying malfunctions. The objective of the thesis is to develop a fault detection system that recognizes all the faults possible in the ABS device; the detection must be at the same time robust (low probability of false alarm and missing detection) and guarantee satisfactory fault detection times. The report presents the analysis and development of algorithms dedicated to automatic detection of faults, proposing robust methods for the calibration, validation and implementation of each one of them. Appropriate cost functions have been adopted to quantify the performance of each algorithm. Since ABS systems for bicycles are in continuous development, the proposed procedures have the characteristic of being easily reproducible and repeatable, making the analysis usable also in the future. The algorithms implemented are of three categories: knowledge based, model based and data driven. The former are based on the qualitative knowledge of the functioning of the system, making use of simple logical rules for the identification of faults; the model based algorithms use mathematical models to detect behaviors of the system that deviate from the nominal conditions; the data driven algorithms analyze the characteristics of the signals and their relations with each other in order to detect anomalies. Each algorithm has been verified on the real system or in simulation, both under nominal operating conditions (verifying the absence of false alarms), and issuing fault injection for checking the correct detection of faults. Therefore, simulators have been developed based on identified models of the system in order to reproduce as faithfully as possible typologies of faults that can not be tested on the real device.
Il sistema di assistenza alla frenata (ABS) per biciclette ha come scopo evitare i bloccaggi della ruota anteriore e i ribaltamenti dovuti a frenate troppo aggressive, cio`e due delle cause piu` frequenti di incidenti in bicicletta. Di recente il dispositivo `e stato implementato su biciclette elettriche, un veicolo la cui diffusione sta rapidamente cambiando il mondo del trasporto personale e che ha portato molta attenzione sul problema della sicurezza del ciclista. La complessita` del dispositivo ABS lo rende vulnerabile a guasti e malfunzionamenti che potrebbero impedire il funzionamento corretto e portare a conseguenze molto gravi per il ciclista, come cadute e scontri con altri veicoli. Per questi motivi il dispositivo ABS viene definito safety-critical e necessita dunque di un metodo per l’identificazione dei malfunzionamenti. L’obiettivo che si pone l’elaborato `e quello di ottenere un sistema di fault detection che comprenda tutti i guasti possibili nel dispositivo di assistenza alla frenata, abbia buone caratteristiche di robustezza (bassa probabilita` di falsi allarmi o mancate identificazioni) e garantisca tempi di rilevamento dei guasti soddisfacenti. L’elaborato presenta l’analisi e lo sviluppo di algoritmi dedicati al rilevamento automatico dei guasti, proponendo robuste metodologie per la calibrazione e validazione di ciascuno di questi. Opportune cifre di costo sono state adottate per quantificare le prestazioni di ciascun algoritmo. Dal momento che il sistema ABS per biciclette `e in continuo sviluppo, le procedure proposte hanno la caratteristica di essere facilmente ripetibili, rendendo l’analisi utilizzabile anche in futuro. Gli algoritmi implementati sono di tre categorie: knowledge based, model based e data driven. I primi si basano sulla conoscenza qualitativa del funzionamento del sistema, facendo uso di semplici regole logiche per l’identificazione dei guasti; gli algoritmi model based utilizzano i modelli matematici per rilevare comportamenti del sistema che si allontanano dalle condizioni nominali; gli algoritmi data driven analizzano le caratteristiche dei segnali e le loro relazioni di modo da rilevare anomalie. Ogni algoritmo `e stato testato sul sistema in esame o in simulazione sia in condizioni di funzionamento nominale (verificando l’assenza di falsi allarmi), sia effettuando fault injection per la verifica della corretta identificazione dei guasti. Sono stati dunque sviluppati dei simulatori basati su modelli identificati del sistema al fine di riprodurre il piu` fedelmente possibile tipologie di fault che non possono essere testate sul sistema reale.
Analisi e sviluppo di algoritmi di fault detection per un sistema di assistenza alla frenata su bicicletta
PASSARIN, FEDERICO
2017/2018
Abstract
The anti-lock braking system (ABS) for bicycles is designed to prevent front wheel lockups and nose-over motions (stoppie) due to excessively aggressive braking, which is one of the most frequent causes of bicycle accidents. Recently the device has been implemented on electric bicycles (e-bikes), a vehicle whose spread is rapidly changing the world of personal transport and which has brought much attention to the problem of cyclist safety. The complexity of the ABS device makes it vulnerable to failures and malfunctions that could prevent proper operation and lead to very serious consequences for the cyclist, such as falls and collisions with other vehicles. For these reasons, the ABS device is called safety-critical and therefore requires a method for identifying malfunctions. The objective of the thesis is to develop a fault detection system that recognizes all the faults possible in the ABS device; the detection must be at the same time robust (low probability of false alarm and missing detection) and guarantee satisfactory fault detection times. The report presents the analysis and development of algorithms dedicated to automatic detection of faults, proposing robust methods for the calibration, validation and implementation of each one of them. Appropriate cost functions have been adopted to quantify the performance of each algorithm. Since ABS systems for bicycles are in continuous development, the proposed procedures have the characteristic of being easily reproducible and repeatable, making the analysis usable also in the future. The algorithms implemented are of three categories: knowledge based, model based and data driven. The former are based on the qualitative knowledge of the functioning of the system, making use of simple logical rules for the identification of faults; the model based algorithms use mathematical models to detect behaviors of the system that deviate from the nominal conditions; the data driven algorithms analyze the characteristics of the signals and their relations with each other in order to detect anomalies. Each algorithm has been verified on the real system or in simulation, both under nominal operating conditions (verifying the absence of false alarms), and issuing fault injection for checking the correct detection of faults. Therefore, simulators have been developed based on identified models of the system in order to reproduce as faithfully as possible typologies of faults that can not be tested on the real device.File | Dimensione | Formato | |
---|---|---|---|
thesis_main.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
16.21 MB
Formato
Adobe PDF
|
16.21 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/10589/144884