Carbon nanotube (CNT)-reinforced nanocomposite is a new type of materials represented by a matrix reinforced by a tubular nanoparticle, which can provide improved mechanical properties as well as interesting multifunctional applications, and thus attracts lots of effort on related investigations. In this study, we focused on analysing the mechanical behaviours of CNT-reinforced nanocomposite that exploit the epoxy polymer as matrix. First goal of this work is to develop automatic modelling procedure for a large number of reliable numerical models capable to investigate the impact of CNT mass fraction and matrix adjacent length on the mechanical response of CNT-reinforced nanocomposite under tensile loading in the numerical framework. The other goal of this study is to apply the Deep Learning for evaluating suitable CNT mass fraction and matrix adjacent length for the model with desired stress-strain curve of tension. The approaches of possible automation of all process involved was investigated. An algorithm of automatic modelling based on software was developed, all steps of which were introduced with its limitations of algorithm discussed as well. Forty numerical models, with physical parameters defined based on experimental results, was successfully developed and processed. Obtained results were automatically post-processed and the impact of CNT mass fraction and matrix adjacent length on the mechanical responses of CNT-reinforced nanocomposite was discussed. The application of Deep Learning was applied in a way of creating, training, validating and testing of shallow artificial neural networks. The data for training was obtained from numerical models with the current automatic modelling strategy. Training optimization was used to define the best lay-out of neural network for this application. Neural network targeting suitable CNT mass fraction with desired stress-strain curve presented a poor performance, while the possible reasons were discussed in this thesis. Neural network for evaluating the adjacent matrix length in numerical model was developed and showed an good performance. Using this artificial neural network, it is possible to assess the value of adjacent matrix length for the model with desired stress-strain curve in tension.
Il nanocomposito rinforzato con nanotubi di carbonio (CNT) è un nuovo tipo di materiale rappresentato da una matrice rinforzata da una nanoparticella tubolare, che può fornire proprietà meccaniche migliorate e interessanti applicazioni multifunzionali, e quindi attrae molti sforzi nelle relative indagini. In questo studio, ci siamo concentrati sull'analisi dei comportamenti meccanici dei nanocompositi rinforzati con CNT che sfruttano il polimero epossidico come matrice. Il primo obiettivo di questo lavoro è quello di sviluppare una procedura di modellazione automatica per un gran numero di modelli numerici affidabili in grado di studiare l'impatto della frazione di massa del CNT e della lunghezza adiacente della matrice sulla risposta meccanica del nanocomposito rinforzato con CNT sotto carico di trazione nel quadro numerico. L'altro obiettivo di questo studio è applicare il Deep Learning per valutare la frazione di massa CNT adatta e la lunghezza adiacente della matrice per il modello con la curva tensione-deformazione desiderata. Sono stati studiati gli approcci di possibile automazione di tutti i processi coinvolti. È stato sviluppato un algoritmo di modellazione automatica basato su software, di cui sono stati introdotti tutti i passaggi e discussi anche i limiti dell'algoritmo. Quaranta modelli numerici, con parametri fisici definiti sulla base di risultati sperimentali, sono stati sviluppati ed elaborati con successo. I risultati ottenuti sono stati automaticamente post-elaborati ed è stato discusso l'impatto della frazione di massa di CNT e della lunghezza adiacente della matrice sulle risposte meccaniche del nanocomposito rinforzato con CNT. L'applicazione del Deep Learning è stata applicata in modo da creare, addestrare, convalidare e testare reti neurali artificiali superficiali. I dati per l'addestramento sono stati ottenuti da modelli numerici con l'attuale strategia di modellazione automatica. L'ottimizzazione dell'addestramento è stata utilizzata per definire il miglior layout della rete neurale per questa applicazione. La rete neurale che mira a una frazione di massa CNT adatta con la curva sforzo-deformazione desiderata ha presentato una prestazione scadente, mentre le possibili ragioni sono state discusse in questa tesi. La rete neurale per valutare la lunghezza della matrice adiacente nel modello numerico è stata sviluppata e ha mostrato buone prestazioni. Utilizzando questa rete neurale artificiale, è possibile valutare il valore della lunghezza della matrice adiacente per il modello con la curva sforzo-deformazione desiderata in tensione.
Finite element analysis and deep learning application on evaluating the impact of CNT mass fraction and matrix adjacent length on mechanical modelling of CNT-reinforced nanocomposite
Kalagaev, Iurii
2020/2021
Abstract
Carbon nanotube (CNT)-reinforced nanocomposite is a new type of materials represented by a matrix reinforced by a tubular nanoparticle, which can provide improved mechanical properties as well as interesting multifunctional applications, and thus attracts lots of effort on related investigations. In this study, we focused on analysing the mechanical behaviours of CNT-reinforced nanocomposite that exploit the epoxy polymer as matrix. First goal of this work is to develop automatic modelling procedure for a large number of reliable numerical models capable to investigate the impact of CNT mass fraction and matrix adjacent length on the mechanical response of CNT-reinforced nanocomposite under tensile loading in the numerical framework. The other goal of this study is to apply the Deep Learning for evaluating suitable CNT mass fraction and matrix adjacent length for the model with desired stress-strain curve of tension. The approaches of possible automation of all process involved was investigated. An algorithm of automatic modelling based on software was developed, all steps of which were introduced with its limitations of algorithm discussed as well. Forty numerical models, with physical parameters defined based on experimental results, was successfully developed and processed. Obtained results were automatically post-processed and the impact of CNT mass fraction and matrix adjacent length on the mechanical responses of CNT-reinforced nanocomposite was discussed. The application of Deep Learning was applied in a way of creating, training, validating and testing of shallow artificial neural networks. The data for training was obtained from numerical models with the current automatic modelling strategy. Training optimization was used to define the best lay-out of neural network for this application. Neural network targeting suitable CNT mass fraction with desired stress-strain curve presented a poor performance, while the possible reasons were discussed in this thesis. Neural network for evaluating the adjacent matrix length in numerical model was developed and showed an good performance. Using this artificial neural network, it is possible to assess the value of adjacent matrix length for the model with desired stress-strain curve in tension.File | Dimensione | Formato | |
---|---|---|---|
2021_10_Kalagaev.pdf
accessibile in internet per tutti
Descrizione: Thesis text file
Dimensione
5.88 MB
Formato
Adobe PDF
|
5.88 MB | Adobe PDF | Visualizza/Apri |
Automation_folder_Kalagaev.zip
accessibile in internet solo dagli utenti autorizzati
Descrizione: Folder with codes of automatic modelling procedure described in text
Dimensione
15.22 MB
Formato
Unknown
|
15.22 MB | Unknown | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/179952