“Industry 4.0” is the name of an ongoing transformation process of the industrial world based on innovative frameworks to coordinate new digital technology, data analysis and big datasets, objects sensorization, and artificial intelligence with the aim of creating value for companies. Industries have entered the movement to obtain long term strategic advantages in the global market, by increasing productivity, capability, and flexibility. In this new framework, traditional sensors have evolved into “smart sensors”, enabling innovative high-performance industrial applications. Smart sensors can manage and process runtime data in an embedded system, unlocking “edge computing” possibilities. They can acquire and process data as close as possible to the data source, for this reason increasing system security and safety, response latency and resilience. STMicroelectronics, a European leader in the semiconductor industry and holding a big portion of the MEMS market, launched an evaluation board, STEVAL-STWINBXCS1, which consists of a multi-sensor board with a micro-controller unit (STM32 family). Accelerometers, microphones, and ambient sensors can be used to collect data or to run Artificial Intelligence-based algorithms on the same board where the sensors are fixed. An evaluation board allows for fast prototyping and to test the sensor network before any deployment.The proposed work starts from the modification of a basic firmware provided by ST-Microelectronics: then it proposes, deploys, and tests a demonstration applications of “Embedded Artificial Intelligence (A.I.)” for anomaly detection. The presented applications have been developed by using the sensor board STEVAL-STWINBXCS1 and an industrial grade MEMS microphone. The same board is used for data acquisition, to implement the neural network processing of the input. The final prediction is made available on a virtual COM port. Parallel to the experiments, the fundamentals concepts of Machine Learning are reported; these are now the state of the art: an effort is being performed to shrink and adapt them to the limited microcontroller memory, together with a study about a general methodology to develop successful applications and standard model prediction testing to assess a system performance.
Il termine inglese “Industry 4.0” indica il processo di trasformazione digitale in ambito industrale. Partecipare a questa innovazione permette alle aziende di ottenere una posizione strategica dominante nel lungo periodo, aumentando l’efficienza produttiva e rimanendo competitive sul mercato globale. L’evoluzione dei sensori in “sensori intelligenti” (“smart sensor node”) offre numerose possibilità di applicazioni di “Industry 4.0”. Questi tipi di sensori sono capaci di gestire e valutare dati a bordo del sensore stesso e in diverse posizioni all’interno di una rete di sensori. I sensori intelligenti permettono di acquisire dati ed elaborarli direttamente a bordo del sensore o eventualmente nel gateway, il sistema che raccoglie i dati per il loro invio ai luoghi di memorizzazione e/o di analisi globale. La minimizzazione della distanza tra sorgente di dati, strumento di acquisizione e strumento di elaborazione offre numerosi vantaggi in termini di sicurezza informatica, velocità di risposta e resilienza del sistema. STMicroelectronics, leader nel settore dei semiconduttori e sensori di tipo MEMS ha introdotto nel mercato l’STEVAL-STWINBXCS1 che consiste in una evaluation board di 50 x 50 mm composta da più sensori e un microprocessore. Gli accelerometri, microfoni o sensori ambientali possono essere utilizzati per raccogliere dati e utilizzarli per lanciare un algoritmo di intelligenza artificiale installato a bordo della scheda stessa.L’uso di evaluation board permette di prototipare reti di sensori prima di effettuare un’installazione definitiva. Il presente lavoro di tesi, propone, implementa e verifica applicazioni di “Embedded Artificial Intelligence (A.I)” per la rilevazione di anomalie di sistemi meccanici, tramite l’impiego di segnali acustici. La tesi si pone pertanto come obiettivo sia lo sviluppo di firmware dedicati per la gestione di sensoristica mems, sia lo studio di una metodologia per l’implementazione di strategie di AI in sistemi dalla ridotta capactià computazionale quali i microcontrollori. La STEVAL-STWINBXCS1 è stata utilizzata per acquisire dati nei differenti test-case analizzati. a stessa board è stata poi impiegata come target di verifica delle performance delle strategie di AI sviluppate, dimostrando la possiblità di poter essere impiegata in applicazione di IIoT (Industrial Internet of Things) in ambito "Industry 4.0"
Embedded artificial intelligence for IIoT applications on mechanical systems
SILVA, GIOVANNI
2021/2022
Abstract
“Industry 4.0” is the name of an ongoing transformation process of the industrial world based on innovative frameworks to coordinate new digital technology, data analysis and big datasets, objects sensorization, and artificial intelligence with the aim of creating value for companies. Industries have entered the movement to obtain long term strategic advantages in the global market, by increasing productivity, capability, and flexibility. In this new framework, traditional sensors have evolved into “smart sensors”, enabling innovative high-performance industrial applications. Smart sensors can manage and process runtime data in an embedded system, unlocking “edge computing” possibilities. They can acquire and process data as close as possible to the data source, for this reason increasing system security and safety, response latency and resilience. STMicroelectronics, a European leader in the semiconductor industry and holding a big portion of the MEMS market, launched an evaluation board, STEVAL-STWINBXCS1, which consists of a multi-sensor board with a micro-controller unit (STM32 family). Accelerometers, microphones, and ambient sensors can be used to collect data or to run Artificial Intelligence-based algorithms on the same board where the sensors are fixed. An evaluation board allows for fast prototyping and to test the sensor network before any deployment.The proposed work starts from the modification of a basic firmware provided by ST-Microelectronics: then it proposes, deploys, and tests a demonstration applications of “Embedded Artificial Intelligence (A.I.)” for anomaly detection. The presented applications have been developed by using the sensor board STEVAL-STWINBXCS1 and an industrial grade MEMS microphone. The same board is used for data acquisition, to implement the neural network processing of the input. The final prediction is made available on a virtual COM port. Parallel to the experiments, the fundamentals concepts of Machine Learning are reported; these are now the state of the art: an effort is being performed to shrink and adapt them to the limited microcontroller memory, together with a study about a general methodology to develop successful applications and standard model prediction testing to assess a system performance.| File | Dimensione | Formato | |
|---|---|---|---|
|
Thesis_final_v2.1.pdf
non accessibile
Dimensione
16.72 MB
Formato
Adobe PDF
|
16.72 MB | Adobe PDF | Visualizza/Apri |
|
ExecutiveSummary_v12.pdf
non accessibile
Dimensione
494.16 kB
Formato
Adobe PDF
|
494.16 kB | 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/10589/201818