The use of computer vision-based quality control systems is increasing in industries due to their ability to perform fast and reliable inspections, which are essential to ensure product quality and optimize production efficiency. These inspections are applied to a wide range of manufacturing products and vary in time and efficiency depending on their characteristics. In particular, the inspection of highly reflective surfaces, such as chrome, is more complex and requires specialized and effective inspection processes. In this thesis, an innovative vision system based on tradition neural networks is presented, which allows the quality control of the chrome-plated surface of a motorbike exhaust pipe to be analyzed at the end of its production phase. In the study, several neural networks were trained to perform binary classification, to check the compliance of the samples, and multiclass to identify the defect type. The results obtained were compared to assess accuracy and prediction times. As the surface is highly reflective, image acquisition was challenging and therefore special focus was given to illumination techniques. In particular, two solutions are presented in this study that allowed defects to be highlighted and identified. To analyze the entire surface of the exhaust pipe from all perspectives, a customized setup with a rotary table was designed to allow a complete and accurate scan of the sample. Then, the images captured using the different techniques were processed and, subsequently, combined to create a single multi-channel image, providing greater detail to highlight any defects. This work presents an efficient solution to be applied in an industrial environment for the quality control of highly reflective surfaces. Furthermore, as the vision system is designed to provide a high level of flexibility, the solution can easily be applied to other samples with similar characteristics.
L’adozione di sistemi di visione per il controllo di qualità è in forte crescita nell’industria, grazie alla loro capacità di effettuare ispezioni rapide e affidabili, necessarie a garantire la qualità dei prodotti e ottimizzare l’efficienza produttiva. Questi controlli vengono applicati a una vasta gamma di prodotti manifatturieri e la loro rapidità ed efficacia variano in base alle loro caratteristiche. Nello specifico, l’ispezione delle superfici altamente riflettenti, come quelle cromate, è complessa e richiede processi di controllo mirati ed efficaci. In questa tesi viene presentato un sistema di visione innovativo, basato su reti neurali tradizionali, che permette di analizzare la qualità della superficie cromata di un silenziatore di una moto al termine della sua fase di produzione. Nello studio sono state addestrate diverse reti neurali al fine di eseguire una classificazione binaria, per verificare la conformità dei campioni, e multiclasse per identificare la tipologia del difetto. I risultati ottenuti sono stati confrontati per valutare l’accuratezza e i tempi di predizione. Essendo la superficie altamente riflettente, l’acquisizione delle immagini è stata impegnativa e dunque è stata posta particolare attenzione alle tecniche di illuminazione. In particolare, in questo studio vengono presentate due soluzioni che hanno permesso di evidenziare e identificare i difetti. Per analizzare l’intera superficie del silenziatore da diverse angolazioni, è stata progettata su misura una struttura con una tavola rotante per consentire un’analisi completa e accurata del campione. Le immagini acquisite con le diverse tecniche sono state quindi elaborate e, in seguito, combinate per creare un’unica immagine multicanale, che permettesse di evidenziare i difetti contenendo più informazioni. Questo lavoro propone una soluzione efficace da applicare in ambito industriale per il controllo della qualità delle superfici altamente riflettenti. Inoltre, essendo il sistema di visione progettato per garantire un elevato livello di flessibilità, la soluzione può essere facilmente applicata ad altri campioni con caratteristiche simili.
Vision system for quality inspection of motorcycle components
TIMMER, VALERIA
2023/2024
Abstract
The use of computer vision-based quality control systems is increasing in industries due to their ability to perform fast and reliable inspections, which are essential to ensure product quality and optimize production efficiency. These inspections are applied to a wide range of manufacturing products and vary in time and efficiency depending on their characteristics. In particular, the inspection of highly reflective surfaces, such as chrome, is more complex and requires specialized and effective inspection processes. In this thesis, an innovative vision system based on tradition neural networks is presented, which allows the quality control of the chrome-plated surface of a motorbike exhaust pipe to be analyzed at the end of its production phase. In the study, several neural networks were trained to perform binary classification, to check the compliance of the samples, and multiclass to identify the defect type. The results obtained were compared to assess accuracy and prediction times. As the surface is highly reflective, image acquisition was challenging and therefore special focus was given to illumination techniques. In particular, two solutions are presented in this study that allowed defects to be highlighted and identified. To analyze the entire surface of the exhaust pipe from all perspectives, a customized setup with a rotary table was designed to allow a complete and accurate scan of the sample. Then, the images captured using the different techniques were processed and, subsequently, combined to create a single multi-channel image, providing greater detail to highlight any defects. This work presents an efficient solution to be applied in an industrial environment for the quality control of highly reflective surfaces. Furthermore, as the vision system is designed to provide a high level of flexibility, the solution can easily be applied to other samples with similar characteristics.File | Dimensione | Formato | |
---|---|---|---|
2024_12_Timmer_Tesi.pdf
non accessibile
Descrizione: testo tesi
Dimensione
11.93 MB
Formato
Adobe PDF
|
11.93 MB | Adobe PDF | Visualizza/Apri |
2024_12_Timmer_Executive Summary.pdf
non accessibile
Descrizione: executive summary
Dimensione
4.64 MB
Formato
Adobe PDF
|
4.64 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/10589/231030