Physics informed neural networks(PINNs)are a deep learning approach designed to solve partial differential equations(PDEs).Accurately learning the initial conditions is crucial when employing PINNs to solve PDEs.Howev...Physics informed neural networks(PINNs)are a deep learning approach designed to solve partial differential equations(PDEs).Accurately learning the initial conditions is crucial when employing PINNs to solve PDEs.However,simply adjusting weights and imposing hard constraints may not always lead to better learning of the initial conditions;sometimes it even makes it difficult for the neural networks to converge.To enhance the accuracy of PINNs in learning the initial conditions,this paper proposes a novel strategy named causally enhanced initial conditions(CEICs).This strategy works by embedding a new loss in the loss function:the loss is constructed by the derivative of the initial condition and the derivative of the neural network at the initial condition.Furthermore,to respect the causality in learning the derivative,a novel causality coefficient is introduced for the training when selecting multiple derivatives.Additionally,because CEICs can provide more accurate pseudo-labels in the first subdomain,they are compatible with the temporal-marching strategy.Experimental results demonstrate that CEICs outperform hard constraints and improve the overall accuracy of pre-training PINNs.For the 1D-Korteweg–de Vries,reaction and convection equations,the CEIC method proposed in this paper reduces the relative error by at least 60%compared to the previous methods.展开更多
The single event effects(SEEs)evaluations caused by atmospheric neutrons were conducted on three different convolutional neural network(CNN)models(Yolov3,MNIST,and ResNet50)in the atmospheric neutron irradiation spect...The single event effects(SEEs)evaluations caused by atmospheric neutrons were conducted on three different convolutional neural network(CNN)models(Yolov3,MNIST,and ResNet50)in the atmospheric neutron irradiation spectrometer(ANIS)at the China Spallation Neutron Source(CSNS).The Yolov3 and MNIST models were implemented on the XILINX28-nm system-on-chip(So C).Meanwhile,the Yolov3 and ResNet50 models were deployed on the XILINX 16-nm Fin FET Ultra Scale+MPSoC.The atmospheric neutron SEEs on the tested CNN systems were comprehensively evaluated from six aspects,including chip type,network architecture,deployment methods,inference time,datasets,and the position of the anchor boxes.The various types of SEE soft errors,SEE cross-sections,and their distribution were analyzed to explore the radiation sensitivities and rules of 28-nm and 16-nm SoC.The current research can provide the technology support of radiation-resistant design of CNN system for developing and applying high-reliability,long-lifespan domestic artificial intelligence chips.展开更多
Optical neural networks have significant advantages in terms of power consumption,parallelism,and high computing speed,which has intrigued extensive attention in both academic and engineering communities.It has been c...Optical neural networks have significant advantages in terms of power consumption,parallelism,and high computing speed,which has intrigued extensive attention in both academic and engineering communities.It has been considered as one of the powerful tools in promoting the fields of imaging processing and object recognition.However,the existing optical system architecture cannot be reconstructed to the realization of multi-functional artificial intelligence systems simultaneously.To push the development of this issue,we propose the pluggable diffractive neural networks(P-DNN),a general paradigm resorting to the cascaded metasurfaces,which can be applied to recognize various tasks by switching internal plug-ins.As the proof-of-principle,the recognition functions of six types of handwritten digits and six types of fashions are numerical simulated and experimental demonstrated at near-infrared regimes.Encouragingly,the proposed paradigm not only improves the flexibility of the optical neural networks but paves the new route for achieving high-speed,low-power and versatile artificial intelligence systems.展开更多
High-resolution flow field data has important applications in meteorology,aerospace engineering,high-energy physics and other fields.Experiments and numerical simulations are two main ways to obtain high-resolution fl...High-resolution flow field data has important applications in meteorology,aerospace engineering,high-energy physics and other fields.Experiments and numerical simulations are two main ways to obtain high-resolution flow field data,while the high experiment cost and computing resources for simulation hinder the specificanalysis of flow field evolution.With the development of deep learning technology,convolutional neural networks areused to achieve high-resolution reconstruction of the flow field.In this paper,an ordinary convolutional neuralnetwork and a multi-time-path convolutional neural network are established for the ablative Rayleigh-Taylorinstability.These two methods can reconstruct the high-resolution flow field in just a few seconds,and further greatlyenrich the application of high-resolution reconstruction technology in fluid instability.Compared with the ordinaryconvolutional neural network,the multi-time-path convolutional neural network model has smaller error and canrestore more details of the flow field.The influence of low-resolution flow field data obtained by the two poolingmethods on the convolutional neural networks model is also discussed.展开更多
This paper addresses the issue of nonfragile state estimation for memristive recurrent neural networks with proportional delay and sensor saturations. In practical engineering, numerous unnecessary signals are transmi...This paper addresses the issue of nonfragile state estimation for memristive recurrent neural networks with proportional delay and sensor saturations. In practical engineering, numerous unnecessary signals are transmitted to the estimator through the networks, which increases the burden of communication bandwidth. A dynamic event-triggered mechanism,instead of a static event-triggered mechanism, is employed to select useful data. By constructing a meaningful Lyapunov–Krasovskii functional, a delay-dependent criterion is derived in terms of linear matrix inequalities for ensuring the global asymptotic stability of the augmented system. In the end, two numerical simulations are employed to illustrate the feasibility and validity of the proposed theoretical results.展开更多
Efficiently solving partial differential equations(PDEs)is a long-standing challenge in mathematics and physics research.In recent years,the rapid development of artificial intelligence technology has brought deep lea...Efficiently solving partial differential equations(PDEs)is a long-standing challenge in mathematics and physics research.In recent years,the rapid development of artificial intelligence technology has brought deep learning-based methods to the forefront of research on numerical methods for partial differential equations.Among them,physics-informed neural networks(PINNs)are a new class of deep learning methods that show great potential in solving PDEs and predicting complex physical phenomena.In the field of nonlinear science,solitary waves and rogue waves have been important research topics.In this paper,we propose an improved PINN that enhances the physical constraints of the neural network model by adding gradient information constraints.In addition,we employ meta-learning optimization to speed up the training process.We apply the improved PINNs to the numerical simulation and prediction of solitary and rogue waves.We evaluate the accuracy of the prediction results by error analysis.The experimental results show that the improved PINNs can make more accurate predictions in less time than that of the original PINNs.展开更多
Physics-informed neural networks(PINNs)have become an attractive machine learning framework for obtaining solutions to partial differential equations(PDEs).PINNs embed initial,boundary,and PDE constraints into the los...Physics-informed neural networks(PINNs)have become an attractive machine learning framework for obtaining solutions to partial differential equations(PDEs).PINNs embed initial,boundary,and PDE constraints into the loss function.The performance of PINNs is generally affected by both training and sampling.Specifically,training methods focus on how to overcome the training difficulties caused by the special PDE residual loss of PINNs,and sampling methods are concerned with the location and distribution of the sampling points upon which evaluations of PDE residual loss are accomplished.However,a common problem among these original PINNs is that they omit special temporal information utilization during the training or sampling stages when dealing with an important PDE category,namely,time-dependent PDEs,where temporal information plays a key role in the algorithms used.There is one method,called Causal PINN,that considers temporal causality at the training level but not special temporal utilization at the sampling level.Incorporating temporal knowledge into sampling remains to be studied.To fill this gap,we propose a novel temporal causality-based adaptive sampling method that dynamically determines the sampling ratio according to both PDE residual and temporal causality.By designing a sampling ratio determined by both residual loss and temporal causality to control the number and location of sampled points in each temporal sub-domain,we provide a practical solution by incorporating temporal information into sampling.Numerical experiments of several nonlinear time-dependent PDEs,including the Cahn–Hilliard,Korteweg–de Vries,Allen–Cahn and wave equations,show that our proposed sampling method can improve the performance.We demonstrate that using such a relatively simple sampling method can improve prediction performance by up to two orders of magnitude compared with the results from other methods,especially when points are limited.展开更多
This article investigates the issue of finite-time state estimation in coupled neural networks under random mixed cyberattacks,in which the Markov process is used to model the mixed cyberattacks.To optimize the utiliz...This article investigates the issue of finite-time state estimation in coupled neural networks under random mixed cyberattacks,in which the Markov process is used to model the mixed cyberattacks.To optimize the utilization of channel resources,a decentralized event-triggered mechanism is adopted during the information transmission.By establishing the augmentation system and constructing the Lyapunov function,sufficient conditions are obtained for the system to be finite-time bounded and satisfy the H_(∞ )performance index.Then,under these conditions,a suitable state estimator gain is obtained.Finally,the feasibility of the method is verified by a given illustrative example.展开更多
Reliable on line cutting tool conditioning monitoring is an essential feature of automatic machine tool and flexible manufacturing system (FMS) and computer integrated manufacturing system (CIMS). Recently artificia...Reliable on line cutting tool conditioning monitoring is an essential feature of automatic machine tool and flexible manufacturing system (FMS) and computer integrated manufacturing system (CIMS). Recently artificial neural networks (ANNs) are used for this purpose in conjunction with suitable sensory systems. The present work in Norwegian University of Science and Technology (NTNU) uses back propagation neural networks (BP) and fuzzy neural networks (FNN) to process the cutting tool state data measured with force and acoustic emission (AE) sensors, and implements a valuable on line tool condition monitoring system using the ANNs. Different ANN structures are designed and investigated to estimate the tool wear state based on the fusion of acoustic emission and force signals. Finally, four case studies are introduced for the sensing and ANN processing of the tool wear states and the failures of the tool with practical experiment examples. The results indicate that a tool wear identification system can be achieved using the sensors integration with ANNs, and that ANNs provide a very effective method of implementing sensor integration for on line monitoring of tool wear states and abnormalities.展开更多
Aim To eliminate the influences of backlash nonlinear characteristics generally existing in servo systems, a nonlinear compensation method using backpropagation neural networks(BPNN) is presented. Methods Based on s...Aim To eliminate the influences of backlash nonlinear characteristics generally existing in servo systems, a nonlinear compensation method using backpropagation neural networks(BPNN) is presented. Methods Based on some weapon tracking servo system, a three layer BPNN was used to off line identify the backlash characteristics, then a nonlinear compensator was designed according to the identification results. Results The simulation results show that the method can effectively get rid of the sustained oscillation(limit cycle) of the system caused by the backlash characteristics, and can improve the system accuracy. Conclusion The method is effective on sloving the problems produced by the backlash characteristics in servo systems, and it can be easily accomplished in engineering.展开更多
A new approach to damage classification for health monitoring of a time-varylng system is presented. The functional-series time-dependent auto regressive moving average （FS-TARMA） time series model is applied to the...A new approach to damage classification for health monitoring of a time-varylng system is presented. The functional-series time-dependent auto regressive moving average （FS-TARMA） time series model is applied to the vibration signal observed in the time-varying system for estimating the TAR/TMA parameters and the innovation variance. These parameters are the functions of the time, represented by a group of projection coefficients on the certain functional subspace with specific basis functions. The estimated TAR/TMA parameters and the innovation variance are further used to calculate the latent components （LCs） as the more informative data for health monitoring evaluation, based on an eigenvalue decomposition technique. LCs are then combined and reduced to numerical values （NVs） as feature sets, which are input to a probabilistic neural network （PNN） for the damage classification. For the evaluation of the proposed method, numerical simulations of the damage classification for a tlme-varylng system are used, in which different classes of damage are modeled by the mass or stiffness reductions. It is demonstrated that the method can identify the damages in the course of operation and the change of parameters on the time-varying background of the system.展开更多
Aim To study the identification and control of nonlinear systems using neural networks. Methods A new type of neural network in which the dynamical error feedback is used to modify the inputs of the network was empl...Aim To study the identification and control of nonlinear systems using neural networks. Methods A new type of neural network in which the dynamical error feedback is used to modify the inputs of the network was employed to reduce the inherent network approximation error. Results A new identification model constructed by the proposed network and stable filters was derived for continuous time nonlinear systems, and a stable adaptive control scheme based on the proposed networks was developed. Conclusion Theory and simulation results show that the modified neural network is feasible to control a class of nonlinear systems.展开更多
Recently, wavelet neural networks have become a popular tool for non-linear functional approximation. Wavelet neural networks, which basis functions are orthonormal scalling functions, are more suitable in approximati...Recently, wavelet neural networks have become a popular tool for non-linear functional approximation. Wavelet neural networks, which basis functions are orthonormal scalling functions, are more suitable in approximating to function. Based on it, approximating to NLAR(p) with wavelet neural networks is studied.展开更多
This paper investigates the problem of the model validation in identifying discrete-time-nonlinear dynamic systems by using neural networks with a single hidden layer.Based on the estimation theory,a synthetic error-i...This paper investigates the problem of the model validation in identifying discrete-time-nonlinear dynamic systems by using neural networks with a single hidden layer.Based on the estimation theory,a synthetic error-index(SEI)criterion for the neural network models has been developed.By using the powerful training algorithm of recursive prediction error (RPE),two simulated non-linear systems are studied,and the results show that the synthetic error-index criterion can be used to verify the dynamic neural network models.Furthermore,the proposed technique is much simple in calculation than that of the effective correlation tests.Finally,some problems required by further study are discussed.展开更多
This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that th...This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.展开更多
This paper introduced the Genetic Algorithms (GAs) and Artificial Neural Networks (ANNs), which have been widely used in optimization of allocating. The combination way of the two optimizing algorithms was used in boa...This paper introduced the Genetic Algorithms (GAs) and Artificial Neural Networks (ANNs), which have been widely used in optimization of allocating. The combination way of the two optimizing algorithms was used in board allocating of furniture production. In the experiment, the rectangular flake board of 3650 mm 1850 mm was used as raw material to allocate 100 sets of Table Bucked. The utilizing rate of the board reached 94.14 % and the calculating time was only 35 s. The experiment result proofed that the method by using the GA for optimizing the weights of the ANN can raise the utilizing rate of the board and can shorten the time of the design. At the same time, this method can simultaneously searched in many directions, thus greatly in-creasing the probability of finding a global optimum.展开更多
Accurate prediction of chemical composition of vacuum gas oil （VGO） is essential for the routine operation of refineries. In this work, a new approach for auto-design of artificial neural networks （ANN） based on a...Accurate prediction of chemical composition of vacuum gas oil （VGO） is essential for the routine operation of refineries. In this work, a new approach for auto-design of artificial neural networks （ANN） based on a genetic algorithm （GA） is developed for predicting VGO saturates. The number of neurons in the hidden layer, the momentum and the learning rates are determined by using the genetic algorithm. The inputs for the artificial neural networks model are five physical properties, namely, average boiling point, density, molecular weight, viscosity and refractive index. It is verified that the genetic algorithm could find the optimal structural parameters and training parameters of ANN. In addition, an artificial neural networks model based on a genetic algorithm was tested and the results indicated that the VGO saturates can be efficiently predicted. Compared with conventional artificial neural networks models, this approach can improve the prediction accuracy.展开更多
In this paper, Neural Networks （NNs） are used in the modeling of ship maneuvering motion. A nonlinear response model and a linear hydrodynamic model of ship maneuvering motion are also investigated. The maneuverabil...In this paper, Neural Networks （NNs） are used in the modeling of ship maneuvering motion. A nonlinear response model and a linear hydrodynamic model of ship maneuvering motion are also investigated. The maneuverability indices and linear non-dimensional hydrodynamic derivatives in the models are identified by using two-layer feed forward NNs. The stability of parametric estimation is confirmed. Then, the ship maneuvering motion is predicted based on the obtained models. A comparison between the predicted results and the model test results demonstrates the validity of the proposed modeling method.展开更多
Traditional scheduled maintenance systems are costly, labor intensive, and typically provide noncomprehensive detection and diagnosis of engine faults. The engine monitoring system (EMS) on modern aircrafts has the p...Traditional scheduled maintenance systems are costly, labor intensive, and typically provide noncomprehensive detection and diagnosis of engine faults. The engine monitoring system (EMS) on modern aircrafts has the potential to provide maintenance personnel with valuable information for detecting and diagnosing engine faults. In this paper, an RBF neural network approach is applied to aeroengine gas path fault diagnosis. It can detect multiple faults and quantify the amount of deterioration of the various engine components as a function of measured parameters. The results obtained demonstrate that the accuracy of diagnosis is consistent with practical requirements. The approach takes advantage of the nonlinear mapping feature of neural networks to capture the appropriate characteristics of an aeroengine. The methodology is generic and applicable to other similar plants having high complexity.展开更多
Uniaxial Compressive Strength (UCS) and modulus of elasticity (E) are the most important rock parameters required and determined for rock mechanical studies in most civil and mining projects. In this study, two mathem...Uniaxial Compressive Strength (UCS) and modulus of elasticity (E) are the most important rock parameters required and determined for rock mechanical studies in most civil and mining projects. In this study, two mathematical methods, regression analysis and Artificial Neural Networks (ANNs), were used to predict the uniaxial compressive strength and modulus of elasticity. The P-wave velocity, the point load index, the Schmidt hammer rebound number and porosity were used as inputs for both meth-ods. The regression equations show that the relationship between P-wave velocity, point load index, Schmidt hammer rebound number and the porosity input sets with uniaxial compressive strength and modulus of elasticity under conditions of linear rela-tions obtained coefficients of determination of (R2) of 0.64 and 0.56, respectively. ANNs were used to improve the regression re-sults. The generalized regression and feed forward neural networks with two outputs (UCS and E) improved the coefficients of determination to more acceptable levels of 0.86 and 0.92 for UCS and to 0.77 and 0.82 for E. The results show that the proposed ANN methods could be applied as a new acceptable method for the prediction of uniaxial compressive strength and modulus of elasticity of intact rocks.展开更多
基金supported by the National Natural Science Foundation of China(Grant Nos.1217211 and 12372244).
文摘Physics informed neural networks(PINNs)are a deep learning approach designed to solve partial differential equations(PDEs).Accurately learning the initial conditions is crucial when employing PINNs to solve PDEs.However,simply adjusting weights and imposing hard constraints may not always lead to better learning of the initial conditions;sometimes it even makes it difficult for the neural networks to converge.To enhance the accuracy of PINNs in learning the initial conditions,this paper proposes a novel strategy named causally enhanced initial conditions(CEICs).This strategy works by embedding a new loss in the loss function:the loss is constructed by the derivative of the initial condition and the derivative of the neural network at the initial condition.Furthermore,to respect the causality in learning the derivative,a novel causality coefficient is introduced for the training when selecting multiple derivatives.Additionally,because CEICs can provide more accurate pseudo-labels in the first subdomain,they are compatible with the temporal-marching strategy.Experimental results demonstrate that CEICs outperform hard constraints and improve the overall accuracy of pre-training PINNs.For the 1D-Korteweg–de Vries,reaction and convection equations,the CEIC method proposed in this paper reduces the relative error by at least 60%compared to the previous methods.
基金Project supported by the National Natural Science Foundation of China(Grant No.12305303)the Natural Science Foundation of Hunan Province of China(Grant Nos.2023JJ40520,2024JJ2044,and 2021JJ40444)+3 种基金the Science and Technology Innovation Program of Hunan Province,China(Grant No.2020RC3054)the Postgraduate Scientific Research Innovation Project of Hunan Province,China(Grant No.CX20240831)the Natural Science Basic Research Plan in the Shaanxi Province of China(Grant No.2023-JC-QN0015)the Doctoral Research Fund of University of South China(Grant No.200XQD033)。
文摘The single event effects(SEEs)evaluations caused by atmospheric neutrons were conducted on three different convolutional neural network(CNN)models(Yolov3,MNIST,and ResNet50)in the atmospheric neutron irradiation spectrometer(ANIS)at the China Spallation Neutron Source(CSNS).The Yolov3 and MNIST models were implemented on the XILINX28-nm system-on-chip(So C).Meanwhile,the Yolov3 and ResNet50 models were deployed on the XILINX 16-nm Fin FET Ultra Scale+MPSoC.The atmospheric neutron SEEs on the tested CNN systems were comprehensively evaluated from six aspects,including chip type,network architecture,deployment methods,inference time,datasets,and the position of the anchor boxes.The various types of SEE soft errors,SEE cross-sections,and their distribution were analyzed to explore the radiation sensitivities and rules of 28-nm and 16-nm SoC.The current research can provide the technology support of radiation-resistant design of CNN system for developing and applying high-reliability,long-lifespan domestic artificial intelligence chips.
基金The authors acknowledge the funding provided by the National Key R&D Program of China(2021YFA1401200)Beijing Outstanding Young Scientist Program(BJJWZYJH01201910007022)+2 种基金National Natural Science Foundation of China(No.U21A20140,No.92050117,No.62005017)programBeijing Municipal Science&Technology Commission,Administrative Commission of Zhongguancun Science Park(No.Z211100004821009)This work was supported by the Synergetic Extreme Condition User Facility(SECUF).
文摘Optical neural networks have significant advantages in terms of power consumption,parallelism,and high computing speed,which has intrigued extensive attention in both academic and engineering communities.It has been considered as one of the powerful tools in promoting the fields of imaging processing and object recognition.However,the existing optical system architecture cannot be reconstructed to the realization of multi-functional artificial intelligence systems simultaneously.To push the development of this issue,we propose the pluggable diffractive neural networks(P-DNN),a general paradigm resorting to the cascaded metasurfaces,which can be applied to recognize various tasks by switching internal plug-ins.As the proof-of-principle,the recognition functions of six types of handwritten digits and six types of fashions are numerical simulated and experimental demonstrated at near-infrared regimes.Encouragingly,the proposed paradigm not only improves the flexibility of the optical neural networks but paves the new route for achieving high-speed,low-power and versatile artificial intelligence systems.
基金National Natural Science Foundation of China(1180500311947102+4 种基金12004005)Natural Science Foundation of Anhui Province(2008085MA162008085QA26)University Synergy Innovation Program of Anhui Province(GXXT-2022-039)State Key Laboratory of Advanced Electromagnetic Technology(Grant No.AET 2024KF006)。
文摘High-resolution flow field data has important applications in meteorology,aerospace engineering,high-energy physics and other fields.Experiments and numerical simulations are two main ways to obtain high-resolution flow field data,while the high experiment cost and computing resources for simulation hinder the specificanalysis of flow field evolution.With the development of deep learning technology,convolutional neural networks areused to achieve high-resolution reconstruction of the flow field.In this paper,an ordinary convolutional neuralnetwork and a multi-time-path convolutional neural network are established for the ablative Rayleigh-Taylorinstability.These two methods can reconstruct the high-resolution flow field in just a few seconds,and further greatlyenrich the application of high-resolution reconstruction technology in fluid instability.Compared with the ordinaryconvolutional neural network,the multi-time-path convolutional neural network model has smaller error and canrestore more details of the flow field.The influence of low-resolution flow field data obtained by the two poolingmethods on the convolutional neural networks model is also discussed.
文摘This paper addresses the issue of nonfragile state estimation for memristive recurrent neural networks with proportional delay and sensor saturations. In practical engineering, numerous unnecessary signals are transmitted to the estimator through the networks, which increases the burden of communication bandwidth. A dynamic event-triggered mechanism,instead of a static event-triggered mechanism, is employed to select useful data. By constructing a meaningful Lyapunov–Krasovskii functional, a delay-dependent criterion is derived in terms of linear matrix inequalities for ensuring the global asymptotic stability of the augmented system. In the end, two numerical simulations are employed to illustrate the feasibility and validity of the proposed theoretical results.
基金Project supported by the National Natural Science Foundation of China(Grant Nos.42005003 and 41475094).
文摘Efficiently solving partial differential equations(PDEs)is a long-standing challenge in mathematics and physics research.In recent years,the rapid development of artificial intelligence technology has brought deep learning-based methods to the forefront of research on numerical methods for partial differential equations.Among them,physics-informed neural networks(PINNs)are a new class of deep learning methods that show great potential in solving PDEs and predicting complex physical phenomena.In the field of nonlinear science,solitary waves and rogue waves have been important research topics.In this paper,we propose an improved PINN that enhances the physical constraints of the neural network model by adding gradient information constraints.In addition,we employ meta-learning optimization to speed up the training process.We apply the improved PINNs to the numerical simulation and prediction of solitary and rogue waves.We evaluate the accuracy of the prediction results by error analysis.The experimental results show that the improved PINNs can make more accurate predictions in less time than that of the original PINNs.
基金Project supported by the Key National Natural Science Foundation of China(Grant No.62136005)the National Natural Science Foundation of China(Grant Nos.61922087,61906201,and 62006238)。
文摘Physics-informed neural networks(PINNs)have become an attractive machine learning framework for obtaining solutions to partial differential equations(PDEs).PINNs embed initial,boundary,and PDE constraints into the loss function.The performance of PINNs is generally affected by both training and sampling.Specifically,training methods focus on how to overcome the training difficulties caused by the special PDE residual loss of PINNs,and sampling methods are concerned with the location and distribution of the sampling points upon which evaluations of PDE residual loss are accomplished.However,a common problem among these original PINNs is that they omit special temporal information utilization during the training or sampling stages when dealing with an important PDE category,namely,time-dependent PDEs,where temporal information plays a key role in the algorithms used.There is one method,called Causal PINN,that considers temporal causality at the training level but not special temporal utilization at the sampling level.Incorporating temporal knowledge into sampling remains to be studied.To fill this gap,we propose a novel temporal causality-based adaptive sampling method that dynamically determines the sampling ratio according to both PDE residual and temporal causality.By designing a sampling ratio determined by both residual loss and temporal causality to control the number and location of sampled points in each temporal sub-domain,we provide a practical solution by incorporating temporal information into sampling.Numerical experiments of several nonlinear time-dependent PDEs,including the Cahn–Hilliard,Korteweg–de Vries,Allen–Cahn and wave equations,show that our proposed sampling method can improve the performance.We demonstrate that using such a relatively simple sampling method can improve prediction performance by up to two orders of magnitude compared with the results from other methods,especially when points are limited.
基金Project supported by the National Natural Science Foundation of China(Grant No.62303016)the Research and Development Project of Engineering Research Center of Biofilm Water Purification and Utilization Technology of the Ministry of Education of China(Grant No.BWPU2023ZY02)+1 种基金the University Synergy Innovation Program of Anhui Province,China(Grant No.GXXT-2023-020)the Key Project of Natural Science Research in Universities of Anhui Province,China(Grant No.2024AH050171).
文摘This article investigates the issue of finite-time state estimation in coupled neural networks under random mixed cyberattacks,in which the Markov process is used to model the mixed cyberattacks.To optimize the utilization of channel resources,a decentralized event-triggered mechanism is adopted during the information transmission.By establishing the augmentation system and constructing the Lyapunov function,sufficient conditions are obtained for the system to be finite-time bounded and satisfy the H_(∞ )performance index.Then,under these conditions,a suitable state estimator gain is obtained.Finally,the feasibility of the method is verified by a given illustrative example.
文摘Reliable on line cutting tool conditioning monitoring is an essential feature of automatic machine tool and flexible manufacturing system (FMS) and computer integrated manufacturing system (CIMS). Recently artificial neural networks (ANNs) are used for this purpose in conjunction with suitable sensory systems. The present work in Norwegian University of Science and Technology (NTNU) uses back propagation neural networks (BP) and fuzzy neural networks (FNN) to process the cutting tool state data measured with force and acoustic emission (AE) sensors, and implements a valuable on line tool condition monitoring system using the ANNs. Different ANN structures are designed and investigated to estimate the tool wear state based on the fusion of acoustic emission and force signals. Finally, four case studies are introduced for the sensing and ANN processing of the tool wear states and the failures of the tool with practical experiment examples. The results indicate that a tool wear identification system can be achieved using the sensors integration with ANNs, and that ANNs provide a very effective method of implementing sensor integration for on line monitoring of tool wear states and abnormalities.
文摘Aim To eliminate the influences of backlash nonlinear characteristics generally existing in servo systems, a nonlinear compensation method using backpropagation neural networks(BPNN) is presented. Methods Based on some weapon tracking servo system, a three layer BPNN was used to off line identify the backlash characteristics, then a nonlinear compensator was designed according to the identification results. Results The simulation results show that the method can effectively get rid of the sustained oscillation(limit cycle) of the system caused by the backlash characteristics, and can improve the system accuracy. Conclusion The method is effective on sloving the problems produced by the backlash characteristics in servo systems, and it can be easily accomplished in engineering.
文摘A new approach to damage classification for health monitoring of a time-varylng system is presented. The functional-series time-dependent auto regressive moving average （FS-TARMA） time series model is applied to the vibration signal observed in the time-varying system for estimating the TAR/TMA parameters and the innovation variance. These parameters are the functions of the time, represented by a group of projection coefficients on the certain functional subspace with specific basis functions. The estimated TAR/TMA parameters and the innovation variance are further used to calculate the latent components （LCs） as the more informative data for health monitoring evaluation, based on an eigenvalue decomposition technique. LCs are then combined and reduced to numerical values （NVs） as feature sets, which are input to a probabilistic neural network （PNN） for the damage classification. For the evaluation of the proposed method, numerical simulations of the damage classification for a tlme-varylng system are used, in which different classes of damage are modeled by the mass or stiffness reductions. It is demonstrated that the method can identify the damages in the course of operation and the change of parameters on the time-varying background of the system.
文摘Aim To study the identification and control of nonlinear systems using neural networks. Methods A new type of neural network in which the dynamical error feedback is used to modify the inputs of the network was employed to reduce the inherent network approximation error. Results A new identification model constructed by the proposed network and stable filters was derived for continuous time nonlinear systems, and a stable adaptive control scheme based on the proposed networks was developed. Conclusion Theory and simulation results show that the modified neural network is feasible to control a class of nonlinear systems.
文摘Recently, wavelet neural networks have become a popular tool for non-linear functional approximation. Wavelet neural networks, which basis functions are orthonormal scalling functions, are more suitable in approximating to function. Based on it, approximating to NLAR(p) with wavelet neural networks is studied.
文摘This paper investigates the problem of the model validation in identifying discrete-time-nonlinear dynamic systems by using neural networks with a single hidden layer.Based on the estimation theory,a synthetic error-index(SEI)criterion for the neural network models has been developed.By using the powerful training algorithm of recursive prediction error (RPE),two simulated non-linear systems are studied,and the results show that the synthetic error-index criterion can be used to verify the dynamic neural network models.Furthermore,the proposed technique is much simple in calculation than that of the effective correlation tests.Finally,some problems required by further study are discussed.
文摘This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.
基金This paper is supported by the Nature Science Foundation of Heilongjiang Province.
文摘This paper introduced the Genetic Algorithms (GAs) and Artificial Neural Networks (ANNs), which have been widely used in optimization of allocating. The combination way of the two optimizing algorithms was used in board allocating of furniture production. In the experiment, the rectangular flake board of 3650 mm 1850 mm was used as raw material to allocate 100 sets of Table Bucked. The utilizing rate of the board reached 94.14 % and the calculating time was only 35 s. The experiment result proofed that the method by using the GA for optimizing the weights of the ANN can raise the utilizing rate of the board and can shorten the time of the design. At the same time, this method can simultaneously searched in many directions, thus greatly in-creasing the probability of finding a global optimum.
文摘Accurate prediction of chemical composition of vacuum gas oil （VGO） is essential for the routine operation of refineries. In this work, a new approach for auto-design of artificial neural networks （ANN） based on a genetic algorithm （GA） is developed for predicting VGO saturates. The number of neurons in the hidden layer, the momentum and the learning rates are determined by using the genetic algorithm. The inputs for the artificial neural networks model are five physical properties, namely, average boiling point, density, molecular weight, viscosity and refractive index. It is verified that the genetic algorithm could find the optimal structural parameters and training parameters of ANN. In addition, an artificial neural networks model based on a genetic algorithm was tested and the results indicated that the VGO saturates can be efficiently predicted. Compared with conventional artificial neural networks models, this approach can improve the prediction accuracy.
基金Partially Supported by the Special Item for the Fujian Provincial Department of Ocean and Fisheries(No.MHGX-16)the Special Item for Universities in Fujian Province by the Education Department(No.JK15003)
文摘In this paper, Neural Networks （NNs） are used in the modeling of ship maneuvering motion. A nonlinear response model and a linear hydrodynamic model of ship maneuvering motion are also investigated. The maneuverability indices and linear non-dimensional hydrodynamic derivatives in the models are identified by using two-layer feed forward NNs. The stability of parametric estimation is confirmed. Then, the ship maneuvering motion is predicted based on the obtained models. A comparison between the predicted results and the model test results demonstrates the validity of the proposed modeling method.
文摘Traditional scheduled maintenance systems are costly, labor intensive, and typically provide noncomprehensive detection and diagnosis of engine faults. The engine monitoring system (EMS) on modern aircrafts has the potential to provide maintenance personnel with valuable information for detecting and diagnosing engine faults. In this paper, an RBF neural network approach is applied to aeroengine gas path fault diagnosis. It can detect multiple faults and quantify the amount of deterioration of the various engine components as a function of measured parameters. The results obtained demonstrate that the accuracy of diagnosis is consistent with practical requirements. The approach takes advantage of the nonlinear mapping feature of neural networks to capture the appropriate characteristics of an aeroengine. The methodology is generic and applicable to other similar plants having high complexity.
文摘Uniaxial Compressive Strength (UCS) and modulus of elasticity (E) are the most important rock parameters required and determined for rock mechanical studies in most civil and mining projects. In this study, two mathematical methods, regression analysis and Artificial Neural Networks (ANNs), were used to predict the uniaxial compressive strength and modulus of elasticity. The P-wave velocity, the point load index, the Schmidt hammer rebound number and porosity were used as inputs for both meth-ods. The regression equations show that the relationship between P-wave velocity, point load index, Schmidt hammer rebound number and the porosity input sets with uniaxial compressive strength and modulus of elasticity under conditions of linear rela-tions obtained coefficients of determination of (R2) of 0.64 and 0.56, respectively. ANNs were used to improve the regression re-sults. The generalized regression and feed forward neural networks with two outputs (UCS and E) improved the coefficients of determination to more acceptable levels of 0.86 and 0.92 for UCS and to 0.77 and 0.82 for E. The results show that the proposed ANN methods could be applied as a new acceptable method for the prediction of uniaxial compressive strength and modulus of elasticity of intact rocks.
