Details

Inverse prediction of design parameters for forming soft morphing structures using deep generative models
ID Brzin, Tomaž (Author), ID Brojan, Miha (Mentor) More about this mentor... This link opens in a new window, ID Jawed, M. Khalid (Comentor)

.pdfPDF - Presentation file, Download (33,47 MB)
MD5: 366BBC9B635C7CCD1435276B2BDD4355

Abstract
In this doctoral dissertation, we present a data-driven approach for the inverse design of morphing soft kirigami composites that utilize the principles of kirigami and strain mismatch to achieve various target shapes. At the center of our methodology is the generative adversarial network, a neural network framework designed to train the generative model tasked to generate the necessary design parameters. By using a pre-trained simulator network, we condition the generative model to generate not only feasible but also accurate design parameters that are used to produce composites that morph into the target shapes. Our findings demonstrate that the generative model effectively predicts the required design parameters, enabling the realization of complex target shapes from planar designs, specifically the composite structures that are able to self-deploy into various 3D shapes; and composite beams that can demonstrate complex motions between the prescribed target positions. We verify our results with the ones found in the literature, by performing numerical simulations and conducting accurate desktop experiments – we have fabricated the composites according to the generated design parameters and found excellent agreement between the target and fabricated shapes. We also compare the method to the competing approaches and demonstrate its superiority.

Language:English
Keywords:inverse design, design parameters, generative adversarial networks, morphing structures, kirigami composites, experiments
Work type:Doctoral dissertation
Typology:2.08 - Doctoral Dissertation
Organization:FS - Faculty of Mechanical Engineering
Publisher:[T. Brzin]
Year:2025
Number of pages:XXXIV, 158 str.
PID:20.500.12556/RUL-170537 This link opens in a new window
UDC:539.3:681.5:004.8(043.3)
COBISS.SI-ID:244759811 This link opens in a new window
Publication date in RUL:09.07.2025
Views:224
Downloads:50
Metadata:XML DC-XML DC-RDF
:
Copy citation
Share:Bookmark and Share

Secondary language

Language:Slovenian
Title:Inverzno napovedovanje konstrukcijskih parametrov za tvorjenje mehkih preobraznih struktur z uporabo globokih generativnih modelov
Abstract:
V doktorski disertaciji predstavljamo podatkovno-podprt pristop za inverzno snovanje preobrazljivih mehkih kirigami kompozitov, ki za doseganje različnih ciljnih oblik uporabljajo načela kirigami razreza in razliko v deformacijah. V središču naše metodologije je generativna kontradiktorna mreža, ogrodje nevronske mreže, ki je zasnovano za učenje generativnega modela, katerega naloga je generirati potrebne konstrukcijske parametre. Z uporabo vnaprej naučene simulacijske mreže pogojujemo generativni model, da generira ne le izvedljive, temveč tudi natančne konstrukcijske parametre, ki se uporabljajo za izdelavo kompozitov za preobrazenje v ciljne oblike. Naše ugotovitve kažejo, da generativni model učinkovito napoveduje potrebne konstrukcijske parametre, kar omogoča realizacijo kompleksnih ciljnih oblik iz ravninskih konstrukcij, zlasti kompozitnih struktur, ki se lahko same preobrazijo v različne 3D oblike; in kompozitnih nosilcev, ki lahko izkazujejo kompleksna gibanja med predpisanimi ciljnimi položaji. Rezultate preverjamo s tistimi iz literature, z izvedbo numeričnih simulacij in natančnimi namiznimi eksperimenti - kompozite smo izdelali v skladu z generiranimi konstrukcijskimi parametri in ugotovili odlično ujemanje med ciljnimi in izdelanimi oblikami. Metodo primerjamo tudi s konkurenčnimi pristopi in dokažemo njeno superiornost.

Keywords:inverzno snovanje, konstrukcijski parametri, generativne kontradiktorne mreže, preobrazne strukture, kirigami kompoziti, eksperimenti

Similar documents

Similar works from RUL:
Similar works from other Slovenian collections:

Back