|
| |
|
| |
|
|
The influence of first CNN layer initialization on training convergence
Krejsa, Jiří ; Věchet, Stanislav ; Chen, K.S.
During evaluation of convolution neural networks on the task of sign language single hand alphabet classification we have discovered that in small but not negligible number of cases the training of the network does not converge at all. This paper investigates the problem that we believe is independent of the application. While the true cause of training divergence was not discovered, we can offer the reader an easy solution from practical point of view – initialization of the first CNN layer using pretrained networks parameters.
|
|
|
Engineering Mechanics 2023
Radolf, Vojtěch ; Zolotarev, Igor
The proceedings contains papers presented at the 29th International Conference on Engineering Mechanics, which has been held in OREA Resort Devět Skal in Czech Republic under auspices of the Czech Society of Mechanics and being a part of IFTOMM (The International Federation for the Promotion of Mechanism and Machine Science) activities. As it corresponds with character of the conference, this proceedings consists of several topic oriented parts: Biomechanics, Fluid mechanics, Dynamics, Fracture mechanics, Mechatronics, Reliability of structures, Mechanics of solids, Technological processes, Thermodynamics. The volume represents a well-balanced overview of theoretical, numerical and experimental work on fundamental and applied studies.
|
|
|
Finite element modal analysis of a silicone vocal fold filled with fluid
Hájek, P. ; Radolf, Vojtěch ; Horáček, Jaromír ; Švec, J. G.
A three dimensional (3D) finite element (FE) model of a silicone vocal fold (VF) filled with fluid is presented here. The silicone part of the model is based on partial differential equations of the continuum mechanics and consider large deformations. The fluid domain encapsulated in the silicone VF is defined semianalytically as a lumped-element model describing the fluid in hydrostatic conditions. The elongated and pressurized silicone VF was subjected to perturbed modal analysis. Results showed that the choice of the fluid inside the VF substantially influences the natural frequencies. Namely, the water-filling lowers the natural frequencies approximately by half over the air-filling. Besides, the procedure of reverse engineering for obtaining the geometry of the VF from already 3D-printed mold is introduced.
|
|
|
Development of autonomous experimental system to analyse yield surfaces distortion due to multiaxial ratcheting
Svárovský, Jiří ; Parma, Slavomír ; Štefan, Jan ; Ciocanel, C. ; Feigenbaum, H. P. ; Marek, René ; Klepač, Vilém ; Plešek, Jiří
Multiaxial ratcheting is a failure mode of structures characterized by the accumulation of plastic strain due to cyclic loading. Despite several models having been developed to predict multiaxial ratcheting, they often fail when validated with experimental data collected under a wide array of loading conditions. In this study, an experimental setup was developed and an autonomous testing procedure was used to experimentally analyze the evolution of the yield surface shape due to cyclic biaxial loading. Thin-walled tubular test specimens were made of 304L steel with a diameter of 40mm and underwent axial-torsional testing using the Instron 8852 system. The total axial strain was increased from 0 to 1% while the total shear strain underwent 5 cycles with the strain amplitude of 0.5% and the mean strain of 0.5%. Three yield surfaces were measured after the straining sequence was completed. Results showed strong directional distortional hardening and good agreement between the flow vectors and the normals to the yield surface, lending support to the associative flow rule.
|
|
|
Deep learning methods for the acoustic emission methods to evaluate an onset of plastic straining
Parma, Slavomír ; Kovanda, Martin ; Chlada, Milan ; Štefan, Jan ; Kober, Jan ; Feigenbaum, H. P. ; Plešek, Jiří
Development of phenomenological plasticity models, hardening rules, and plasticity theories relies on experimental data of plastic straining. The experimental data are usually measured as the stress–strain response of the material being loaded and do not provide any clues or information about the local response of\nmaterial. In this paper, we analyze the plastic deformation of the material using the acoustic emission method and current state-of-the-art neural network models such as the InceptionTime architecture.
|
|
|
Calibrated finite volume method-based simulation framework for laser shock peening
Isoz, Martin ; Gruber, Pavel ; Schmidt, Jaroslav ; Kubíčková, Lucie ; Štefan, Jan ; Kaufman, Jan ; Brajer, Jan ; Gabriel, Dušan
Modern and highly competitive industry seeks components with high strength and fatigue resistance. Both of these properties may be improved by peening of the component surface and the standard peening processes, such as the shot peening, are widely used in both automotive and aerospace industries. The laser shock\npeening (LSP), i.e. hardening of the material surface by a laser-induced shock wave, is a modern alternative to the standard peening. Concurrently, the industrial applications of LSP are promoted by recently emerged affordable high power-density lasers. However, the nascent LSP applications are still mostly a trial-and-error\nprocesses based on an extensive experimental testing. Consequently, we focused on a highly application-driven development of a framework for LSP modeling, and the internal workings and results of which are the focus of the present contribution.
|
|
|
Thermal and creep analysis of VVER-1000 reactor pressure vessel at high temperatures caused by fuel melting during severe accident
Gabriel, Dušan ; Gál, P. ; Kotouč, M. ; Dymáček, Petr ; Masák, Jan ; Kopačka, Ján
Thermal and creep analysis of the VVER-1000 reactor pressure vessel (RPV) was performed at high temperatures caused by fuel melting during severe accident. First, the integral code ASTEC was applied simulating severe accident evolution since an initiating event up to a hypothetical radioactive release into the environment. The ASTEC outputs including the remaining RPV wall thickness, the heat flux achieved and the temperature profile in the ablated vessel wall served as boundary conditions for the consequent assessment of RPV integrity carried out with the aid of finite element method (FEM). The FEM analysis was performed including the creep behaviour of RPV material using a complex creep probabilistic exponential model with damage. The objective of the analysis was to computationally assess emergency condition and, on this basis, to propose a general methodology for evaluating the integrity of RPV at high temperatures due to fuel melting during severe accident.
|
|
|
Dynamic tensile experiments of hard rubber on instron test machine under finite deformations
Šulc, Petr ; Pešek, Luděk ; Bula, Vítězslav ; Šnábl, Pavel
This paper deals with the first experimental tests of hard rubbers under tensile harmonic loading on the axial-torsional servohydraulic Instron 8852 test machine. The objective of these experiments on cylindrical rubber specimens on this machine is to obtain stress-strain curves of rubber specimens, first considering uniaxial finite longitudinal strains for a single load cycle. Next, the hysteresis loops and their skeleton curves are compared for two selected strain measures as a function of their conjugate stress. It is shown how to obtain the dependence between elongation in the axial direction and constriction in the radial direction, which is used to determine the Cauchy stress in this configuration.
|
host ::
RSS.