Search

The paper deals with the isogeometric analysis (IGA) of active composite laminates with piezoelectric layers. IGA is a special formulation of the finite element method (FEM) that aims at seamless integration of geometric and finite element modelling. NURBS basis functions are employed to develop isogeometric shell formulation based on the Reissner-Mindlin kinematics. Piezolayers characterized by electro-mechanical coupled field effects enable active behavior of the considered structures. The electric field acts across the thickness of the piezolayers and is coupled to the in-plane strains. In addition to a number of advantages that NURBS modelling provides, defining the surface normal vector at the points of the control polygon, which are generally not located on the surface, creates certain difficulties. A method of determining the surface normal vectors at the points of the control polygon based on the Greville's points is discussed. In order to demonstrate the applicability of the developed formulation, a benchmark case is computed and the results are compared with those obtained by means of classical FEM formulation, which are available in the literature.

This paper provides an overview of the characteristics of different neural networks trained on the same theoretical database of n-type silicon photoacoustic signals. By adding different levels of random Gaussian noise to the training input signals, two important goals were achieved. First, the optimal level of noise was found which significantly shortens the training networks with minimal loss of accuracy of its predictions. Second, the termination criteria of networks training were activated to avoid overtraining, i.e., networks generalization was performed. A networks efficiency analysis was performed on both theoretical and experimental photoacoustic signals, resulting in a selection of one neural network that is optimal to the performance requirements of the real experiment. It is indicated that the application of such trained networks is more reliable on thicker semiconductors, whose thickness is greater than the value of the carrier diffusion length in the investigated sample. © 2020, Springer Nature B.V.

We developed a method of inverse problem solving in semiconductor photoacoustics based on neural networks application. Simple structured neural networks, trained on a large set of data obtained by the well–known theoretical models in the 20 Hz–20 kHz modulation frequency range, are applied to determine thermal diffusivity, coefficient of linear expansion and thickness of n–type silicon samples, using undistorted experimental photoacoustic signals. The efficiency of the neural networks was tested depending on the type of input data, showing the best performances in the case when signal amplitudes and phases are simultaneously presented to the network. Real–time parameter prediction is achieved together with high accuracy and reliability allowing one to perform the full characterization of a sample in the frequency domain.

In this paper, a simple multilayer perceptron neural network with forward signal propagation was designed and used to simultaneously determine the main physical parameters, such as: the thermal diffusivity, thermal expansion coefficient and thickness, from the transmission, frequency-modulated photoacoustic response of the sample. The amplitude and phase responses of the transmission open-cell photoacoustic signals were calculated in n-type silicon plates using a theoretical model and were used to train and test a neural network. The simulation was done in the modulation frequency range from 20 Hz to 20 kHz and using a wide range of expected values of thermal diffusivity and the thermal coefficient of expansion for semiconductor samples as well as their thickness. The advantages and disadvantages of neural networks utilization as an appropriate mathematical tool designated for semiconductor measurement-oriented purposes are analyzed. Network reliability, precision, and the possibility of operation in real time have been verified on an independent set of signals, establishing photoacoustics as a competitive and powerful technique assigned for material characterization. © 2019, Springer Nature B.V.

An analysis of the application of neural networks as a reliable, precise, and fast tool in open-cell photoacoustics setups for the recognition of microphone effects in the frequency domain from 10 Hz to 100 × 104 Hz is presented. The network is trained to achieve simultaneous recognition of microphone characteristics, which are the most important parameters leading to the distortion of photoacoustic signals in both amplitude and phase. The training is carried out using a theoretically obtained database of amplitudes and phases as the input and five microphone characteristics as the output, based on transmission measurements obtained using an open photoacoustic cell setup. The results show that the network can precisely and reliably interpolate the output to recognize microphone characteristics including electronic effects in the low and acoustic effects in the high frequency domain. The simulations reveal that the network is not capable of interpolating an input including modulation frequencies. Consequently, in real applications, the network training must be adapted to the experimental frequencies, or vice versa. The total number of frequencies used in the experiment must also be in accordance with the total number of frequencies used in the network training.

This paper introduces the possibility of the determination of optical absorption and reflexivity coefficient of silicon samples using neural networks and reverse-back procedure based on the photoacoustics response in the frequency domain. Differences between neural network predictions and parameters obtained with standard photoacoustic signal correction procedures are used to adjust our experimental set-up due to the instability of the optical excitation source and the state (contamination) of the illuminated surface. It has been shown that the changes of the optical absorption values correspond to the light source wavelength fluctuations, while changes in the reflexivity coefficient, obtained in this way, correspond to the small effect of the ultrathin layer formation of SiO2 due to the natural process of surface oxidation.

This thesis discusses the use of machine learning to design guidance, navigation, and control algorithms as an alternative to traditional algorithms for a missile system. The machine learning algorithm used in this thesis is the neural network. It is trained using the Neuro Evolution of Augmenting Topologies algorithm. Furthermore, the missile system and its environment have been modeled in order to simulate and compare the missile performances. The terminal guidance neural network will be compared to the proportional navigation algorithm. In addition, the neural network GPS/INS integration will be compared to the Kalman filter GPS/INS integration. Moreover, the neural network roll, pitch, and yaw autopilots will be compared to the traditional PID roll, pitch, and yaw autopilots. The goal of this thesis is to design neural network guidance, navigation, and control solutions which is expected to perform similar or better than their traditional counterparts. Thereby, the viability of the neural network designs as a guidance, navigation, or control solution will be verified., У овој докторској тези се разматра употреба машинског учења у синтези алгоритама навигације, управљања и вођења ракете, као алтернативи традиционалним алгоритмима. Алгоритам машинског учења који се користи у овој докторској тези је заснован на примени неуронских мрежа. Неуронска мрежа се обучава Неуро еволуционим алгоритмом са приширеном топологијом. Осим тога, извршено је математичко моделовање вођеног пројектила и његовог окружења како би се извршиле нумеричке симулације и упоредиле његове перформансе. Извршено је поређење неуронске мреже алгоритма вођења терминалне фазе са алгоритмом пропорционалне навигације. Осим тога, интеграција GPS/INS-а на бази неуронских мрежа је упоређена са Калмановим филтром. На крају је дато поређење аутопилота по каналима ваљања, пропињања и скретања реализованих неуронским мрежама насупрот традиционалним аутопилотима са ПИД управљачким алгоритмима. Циљ ове докторске тезе је синтеза алгоритма вођења и управљања пројектила применом неуронских мрежа које треба да покаже слично или боље понашање од традиционалних решења. Притом, верификује се одрживост решења примене неуронских мрежа у синтези алгоритама управљања и вођења.