Modeling of complex dynamic systems using differential neural networks with the incorporation of a priori knowledge

Bellamine, F.H.; Almansoori, Ali; Elkamel, Ali

doi:10.1016/j.amc.2015.05.122

Cited by 12 publications

(4 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, in the constrained backpropagation training (Di Muro & Ferrari, 2008;Ferrari & Jensenius, 2008;He, Reif, & Unbehauen, 2000;Rudd, Muro, & Ferrari, 2014), the prior knowledge of boundary conditions was explicitly embedded as equality constraints and imposed on the weights of neural networks. Moreover, the prior knowledge of model forms and boundary conditions can be embedded as regularization terms in the objective function of a neural network to solve ODEs (Bellamine, Almansoori, & Elkamel, 2015;Malek & Shekari Beidokhti, 2006). The prior knowledge of model forms and boundary conditions can also be embedded as regularization terms in the objective function of a neural network by transforming the original PDEs into their weighted residual forms (Dissanayake & Phan-Thien, 1994).…”

Section: Physics-constrained Machine Learningmentioning

confidence: 99%

A Dual-Dimer method for training physics-constrained neural networks with minimax architecture

Liu

Wang

2021

Neural Networks

View full text Add to dashboard Cite

Section: Physics-constrained Machine Learningmentioning

confidence: 99%

A Dual-Dimer method for training physics-constrained neural networks with minimax architecture

Liu

Wang

2021

Neural Networks

View full text Add to dashboard Cite

“…The neural network is designed to perform the monitoring tasks defined by input sequences and sequences of desired values corresponding to the output neurons (reference model) [22][23][24][25]. The NARX dynamic neural network structure proposed in this work is shown in Figure 6.…”

Section: Vibrations Modeling Using Dynamic Neural Networkmentioning

confidence: 99%

Detection and Modeling Vibrational Behavior of a Gas Turbine Based on Dynamic Neural Networks Approach

Benrahmoune

Hafaifa

Guemana

et al. 2018

Strojnícky Časopis - Journal of Mechanical Engineering

View full text Add to dashboard Cite

During the gas turbine exploitation the presence of small defects can cause very high vibration amplifications, localized on the components of this rotating machine. For this, a diagnostic process is necessary for decision-making during the monitoring of failures caused by vibration phenomena, which consists in observing the system by comparing its current data with the data coming from a normal operation. These indicators help engineer to determine the symptoms for the failing components of the system. This work deals with problems related to these vibrations, with the aim of developing a system of detection of failures using dynamic neural networks approach. The originality of this contribution is to calculate the various alarms based on this system which used the determined vibration models in order to ensure a reliable and safe operation of the gas compression installation using the examined gas turbine.

show abstract

“…However, the quality of the prediction can be greatly improved by using customized multi-objective loss functions during the network training phase. Although at the early stages of NN analysis custom loss functions were not commonly used, the idea of a multi-objective loss function has attracted considerable interest particularly in physics-based NN modeling [38][39][40]. This aspect will be discussed in the following sections.…”

Section: Introductionmentioning

confidence: 99%

Finite Element Network Analysis: A Machine Learning based Computational Framework for the Simulation of Physical Systems

Jokar,

Semperlotti

2020

Preprint

View full text Add to dashboard Cite

This study introduces the concept of finite element network analysis (FENA) which is a physics-informed, machine-learning-based, computational framework for the simulation of complex physical systems. The framework leverages the extreme computational speed of trained neural networks and the unique transfer knowledge property of bidirectional recurrent neural networks (BRNN) to provide a uniquely powerful and flexible computing platform. One of the most remarkable properties of this framework consists in its ability to simulate the response of complex systems, made of multiple interconnected components, by combining individually pre-trained network models that do not require any further training following the assembly phase. This remarkable result is achieved via the use of key concepts such as transfer knowledge and network concatenation. Although the computational framework is illustrated and numerically validated for the case of a mechanical system under static loading, the conceptual structure of the framework has broad applicability and could be extended to the most diverse field of computational science. The framework is numerically validated against the solution provided by traditional finite element analysis and the results highlight the outstanding performance of this new concept of computational platform.

show abstract

Modeling of complex dynamic systems using differential neural networks with the incorporation of a priori knowledge

Cited by 12 publications

References 37 publications

A Dual-Dimer method for training physics-constrained neural networks with minimax architecture

A Dual-Dimer method for training physics-constrained neural networks with minimax architecture

Detection and Modeling Vibrational Behavior of a Gas Turbine Based on Dynamic Neural Networks Approach

Finite Element Network Analysis: A Machine Learning based Computational Framework for the Simulation of Physical Systems

Contact Info

Product

Resources

About