Physics-guided Neural Network Research Articles

Physics-guided neural network for predicting international roughness index on flexible pavements considering accuracy, uncertainty and stability

Physics-guided neural network for predicting international roughness index on flexible pavements considering accuracy, uncertainty and stability

Imperfect physics-guided neural networks

Imperfect physics-guided neural networks

Simultaneous Prediction of Gas Flow Rate and Concentration in the Micro-Calorimetric Sensor Using Physics-Guided Neural Networks

Simultaneous Prediction of Gas Flow Rate and Concentration in the Micro-Calorimetric Sensor Using Physics-Guided Neural Networks

Physics-guided neural network for structural health monitoring with lamb waves through boundary reflection elimination

Lamb-wave-based structural health monitoring (SHM) technology for damage location in plate-like structures relies on the postprocessing of captured signals after interacting with damage. Traditional methods typically leverage the time of flight (ToF) of scattered waves from damage. However, these methods are prone to reflected waves from structural boundaries which mix with scattered waves from damage. This is a vital problem faced by most ToF-based detection methods, which seriously narrows the inspection area. To tackle this problem, a machine learning framework, consisting of a multiscale spatiotemporal (MSST) fusion network, is proposed to facilitate the accurate extraction of the ToF of scattered waves through eliminating the influence of boundary reflections. Experiments are conducted with the time-domain Lamb wave signals recorded by a tactically designed piezoelectric sensor array on a 2-mm-thick Al-6061 plate. A pair of circle magnets is attached onto the plate as the wave reflectors. Through step-by-step moving of the magnets in the predefined grids, the corresponding Lamb wave signals are measured to construct a database. An MSST is subsequently designed to minimize the error between estimated and theoretical ToFs, with wavelet coefficients of the signals and transducer position as inputs. The model is trained with the Adam algorithm where 80% of samples in the database are used for training and the rest for evaluation. The final validations are conducted with the scatters off the predefined grids. Results demonstrate that the designed neural network architecture can effectively eliminate boundary reflections and enable precise ToF extraction of the scattered waves from damage. This allows the enlargement of the detection area and presents a promising and useful tool for enhancing the detection performance of existing SHM methods in complex structures.

Fault-Tolerant Sun-Pointing Attitude Control Based on Physics-Guided Neural Networks

Fault-Tolerant Sun-Pointing Attitude Control Based on Physics-Guided Neural Networks

A feedforward control method for turntable servo system based on physics-guided neural network

Model-based feedforward control is a critical control method for high-dynamic and high-precision requirements of turntable servo systems. However, the turntable servo system will be affected by the change of working state in actual work, so the traditional physical feedforward control makes it difficult to obtain an accurate model of the turntable servo system. In this article, a feedforward controller based on a physics-guided neural network (PGNN) is proposed to improve the tracking performance of the turntable servo system. Firstly, a PGNN model is constructed according to the characteristics of the turntable, which is divided into a physical model part and a neural network part. Then the above model is applied to the design of the feedforward controller to replace the traditional feedforward controller. In order to improve the extension of the network, a new penalty function mechanism is proposed, which integrates regularization and physical consistency on the basis of the mean square error term. The automatic tuning algorithm of the parameters of the PGNN model is also designed to facilitate its wide application in the field of turntable control. Simulation results show that the neural network model designed in this article can accurately describe the nonlinear characteristics of the turntable servo system, and the fitting effect under different forms of excitation signals is better than the traditional physical model, with strong generalization ability and robustness. In addition, the feedforward control based on the model effectively improves the tracking accuracy of the system, which has important value and significance for the intelligent development of traditional servo control systems.

Physics-Guided Neural Networks for Intraventricular Vector Flow Mapping.

Intraventricular vector flow mapping (iVFM) seeks to enhance and quantify color Doppler in cardiac imaging. In this study, we propose novel alternatives to the traditional iVFM optimization scheme by utilizing physics-informed neural networks (PINNs) and a physics-guided nnU-Net-based supervised approach. When evaluated on simulated color Doppler images derived from a patient-specific computational fluid dynamics model and in vivo Doppler acquisitions, both approaches demonstrate comparable reconstruction performance to the original iVFM algorithm. The efficiency of PINNs is boosted through dual-stage optimization and pre-optimized weights. On the other hand, the nnU-Net method excels in generalizability and real-time capabilities. Notably, nnU-Net shows superior robustness on sparse and truncated Doppler data while maintaining independence from explicit boundary conditions. Overall, our results highlight the effectiveness of these methods in reconstructing intraventricular vector blood flow. The study also suggests potential applications of PINNs in ultrafast color Doppler imaging and the incorporation of fluid dynamics equations to derive biomarkers for cardiovascular diseases based on blood flow.

Coupling physics in artificial neural network to predict the fatigue behavior of corroded steel wire

To accurately predict the fatigue life of corroded steel wire, the authors proposed an artificial neural network (ANN) model and a probabilistic physics-guided neural network (PPgNN) model. Factors including stress range, mean stress, and corrosion rate were considered as input features of these two neural networks. The ANN model exhibited the best prediction accuracy with a determination coefficient of 0.94; however, it cannot capture the dispersion of fatigue life. Through introduction of physics constraints, PPgNN model was able to predict the standard deviation of fatigue life. The results showed that 94.79% of the dataset was within the 95% confidence interval.

Development of physics-guided neural network framework for acid-base treatment prediction using carbon dioxide-based tubular reactor

Accurate acid-base treatment prediction is necessary to achieve the required yield, given the inherent complexity, high nonlinearity, and restricted availability of data samples; to address this challenge, a data-driven approach was developed. However, the technique is constrained by the need for sufficient data to construct an accurate model and lacks both process insight and physical consistency. Therefore, this study introduces a physics-guided neural network model for acid-base treatment prediction in a dynamic tubular reactor using the fundamental physical intermediate variables obtained through the derivation process of the reaction schematic. By integrating batch experimental data, which provides key intermediate variables such as residence time and hydroxide ion concentration, the model addresses the challenge of high nonlinearity and limited data availability. The result shows that the physics-guided potential of a hydrogen predictor had outstanding performance in terms of prediction accuracy (greatest coefficient of determination value of 0.9381). The proposed model demonstrated an average improvement of 24.92% in pH prediction accuracy compared to traditional models without physical guidance, with a maximum improvement of up to 64.95% under limited data conditions. Moreover, downsampling tests revealed that the proposed model maintained robust performance with minimal accuracy reduction even when data was limited without overfitting implication.

High-Efficiency Finite Element Model Updating of Bridge Structure Using a Novel Physics-Guided Neural Network

An accurate finite element model (FEM) plays a critical role in the structural damage identification. However, due to the existence of the uncertainties, such as material properties and modeling errors, it always exists some gaps between the analytical FEM and experimental structure. While an artificial neural network (ANN)-based model updating methods have been widely adopted to narrow the gap and obtain a baseline FEM, it still faces inaccurate results and fails to meet the physical law. In this regard, the study proposes a novel physics-based loss function inspired by modal sensitivity analysis and incorporates it into the residual neural network, thereby forming a novel physics-guided neural network (PGNN) method. The mapping relationship between the input of structural responses and output of model updating variables is constrained to retain its physical meaning by guiding the training process instead of pure data association, which aims to improve the accuracy of the ANN-based method and achieve accurate and high-efficiency model updating. An experimental example of a continuous rigid frame bridge is adopted to verify the feasibility of the proposed method. Additionally, other common model updating methods, including moth-flame optimization and regularization method, are used to make a comparison. The noise-robustness of the proposed method is investigated as well. Compared to the existing method, the results illustrate that the proposed PGNN method can achieve better model updating and good noise-robustness under high uncertainties, which means the introduction of the physics-based loss function significantly enhances the parameters updating ability of the neural network. The proposed method exhibits high efficiency and promising potential for large-scale bridge structure model updating.

Harnessing physics-guided neural networks for tailoring the orbital angular momentum spectrum in second harmonic generation

The management of orbital angular momentum (OAM) in frequency conversion processes is essential for numerous applications such as quantum and classical optical communications. This paper presents a wavefront modulation approach for the fundamental beam in second harmonic generation (SHG) to efficiently control the OAM spectrum. We employ an inverse design method to derive the necessary wavefront shape of the fundamental beam for achieving a desired SHG OAM spectrum. Specifically, we introduce an efficient inverse design technique based on physics-guided neural networks (PGNNs) that incorporates the coupled equations governing SHG, aimed at tailoring the OAM spectrum of SHG. Utilizing the proposed PGNN, we design the phase pattern for a spatial light modulator (SLM) to shape the wavefront of the fundamental beam. Furthermore, we present a novel loss function, to our knowledge, that effectively links the OAM of the SHG spectrum and efficiency to the SLM phase pattern and crystal temperature, independent of empirical weight coefficients. The proposed PGNN facilitates the purification of the SHG OAM spectrum, even when the fundamental beam comprises mixed Laguerre–Gaussian (LG) modes. Additionally, we demonstrate the generation of desired SHG spectra using the proposed PGNN framework. This study introduces what we believe to be a groundbreaking inverse design method for developing photonic devices with customized functionalities, addressing challenges associated with traditional data-driven deep learning techniques.

Physics Guided Neural Networks with Knowledge Graph

Over the past few decades, machine learning (ML) has demonstrated significant advancements in all areas of human existence. Machine learning and deep learning models rely heavily on data. Typically, basic machine learning (ML) and deep learning (DL) models receive input data and its matching output. Within the model, these models generate rules. In a physics-guided model, input and output rules are provided to optimize the model’s learning, hence enhancing the model’s loss optimization. The concept of the physics-guided neural network (PGNN) is becoming increasingly popular among researchers and industry professionals. It has been applied in numerous fields such as healthcare, medicine, environmental science, and control systems. This review was conducted using four specific research questions. We obtained papers from six different sources and reviewed a total of 81 papers, based on the selected keywords. In addition, we have specifically addressed the difficulties and potential advantages of the PGNN. Our intention is for this review to provide guidance for aspiring researchers seeking to obtain a deeper understanding of the PGNN.

Super-Resolving and Denoising 4D flow MRI of Neurofluids Using Physics-Guided Neural Networks.

To obtain high-resolution velocity fields of cerebrospinal fluid (CSF) and cerebral blood flow by applying a physics-guided neural network (div-mDCSRN-Flow) to 4D flow MRI. The div-mDCSRN-Flow network was developed to improve spatial resolution and denoise 4D flow MRI. The network was trained with patches of paired high-resolution and low-resolution synthetic 4D flow MRI data derived from computational fluid dynamic simulations of CSF flow within the cerebral ventricles of five healthy cases and five Alzheimer's disease cases. The loss function combined mean squared error with a binary cross-entropy term for segmentation and a divergence-based regularization term for the conservation of mass. Performance was assessed using synthetic 4D flow MRI in one healthy and one Alzheimer' disease cases, an in vitro study of healthy cerebral ventricles, and in vivo 4D flow imaging of CSF as well as flow in arterial and venous blood vessels. Comparison was performed to trilinear interpolation, divergence-free radial basis functions, divergence-free wavelets, 4DFlowNet, and our network without divergence constraints. The proposed network div-mDCSRN-Flow outperformed other methods in reconstructing high-resolution velocity fields from synthetic 4D flow MRI in healthy and AD cases. The div-mDCSRN-Flow network reduced error by 22.5% relative to linear interpolation for in vitro core voxels and by 49.5% in edge voxels. The results demonstrate generalizability of our 4D flow MRI super-resolution and denoising approach due to network training using flow patches and physics-based constraints. The mDCSRN-Flow network can facilitate MRI studies involving CSF flow measurements in cerebral ventricles and association of MRI-based flow metrics with cerebrovascular health.

Physics-guided federated learning as an enabler for digital twins

Digital twins bring the potential to increase the efficiency of assets, systems, and processes by building virtual replicas through real-time data and modeling. However, data are often confidential and distributed, high-fidelity models based on physical principles tend to be slow and require detailed knowledge about the asset design that may not be available, simplified models lack accuracy, and data-driven models are not interpretable and limited to their training space. We demonstrate the interplay of two enabling technologies, namely federated learning and hybrid analysis and modeling, to combine the strengths of physics-based and data-driven models on distributed data sets that are kept confidential by different stakeholders. Throughout the work, an ensemble of physics-guided neural networks is designed, optimized, and validated to infer the parameters of digital twin components. The physics-guided neural network is injected with output from a simplified physics-based model in intermediate layers for increased performance, and the injection layer is optimized through weight regularization. The model is then integrated into a simulation of the digital twin asset and validated there. It is shown that the model outperforms the simplified physics-based and the data-driven models both on component and on asset level. In the final step, the physics-guided neural network is trained using federated methods, which allows training on confidential data owned by different stakeholders without requiring the stakeholders to share their data. We show that federated hybrid models can – with the right training strategy – by design achieve results that are always identical and potentially superior to models trained on a central data set. The problem chosen for this demonstration is the prediction of airfoil lift and drag coefficients based on the airfoil shape and the angle of attack, which is a task crucial for simulating wind turbines. Using the physics-guided neural network, the estimation of the airfoil coefficients experiences a speedup of several orders of magnitude compared to the high-fidelity simulation, and integration into blade element momentum theory shows that the turbine thrust, torque, and mechanical power can be accurately calculated. The federated approach makes it possible to extend the training of the model to any number of distributed confidential data sets from both simulations and experiments to further increase the model’s accuracy.

Dynamic wake steering control for maximizing wind farm power based on a physics-guided neural network dynamic wake model

Wake effect is a significant factor contributing to power loss in wind farms. Studies have shown that wake steering control can mitigate this power loss. Currently, wind farm wake control strategies primarily utilize fixed yaw control due to limitations in the accuracy and efficiency of dynamic wake models. However, fixed yaw control fails to fully exploit the power improvement potential of wake steering control. Therefore, in this study, we first propose a dynamic wake model for wind farms based on the physics-guided neural network (PGNN) approach. This model can predict the dynamic wake flow field within wind farms in real time using instantaneous inflow wind speed and turbine operational states. Then, by employing the PGNN dynamic wake model as the predictive model, a wind farm dynamic wake control strategy based on the model predictive control method is proposed. To quantify the advantages of the proposed control strategy, both fixed yaw control and dynamic yaw control are tested on a wind farm with a 3 × 2 layout. Results from large eddy simulations demonstrate that the proposed dynamic wake control strategy increases the power output of the wind farm by 11.51% compared to a 6.56% increase achieved with fixed yaw control.

Physics-Guided Inverse Regression for Crop Quality Assessment

AbstractWe present an innovative approach leveraging Physics-Guided Neural Networks (PGNNs) for enhancing agricultural quality assessments. Central to our methodology is the application of physics-guided inverse regression, a technique that significantly improves the model’s ability to precisely predict quality metrics of crops. This approach directly addresses the challenges of scalability, speed, and practicality that traditional assessment methods face. By integrating physical principles, notably Fick’s second law of diffusion, into neural network architectures, our developed PGNN model achieves a notable advancement in enhancing both the interpretability and accuracy of assessments. Empirical validation conducted on cucumbers and mushrooms demonstrates the superior capability of our model in outperforming conventional computer vision techniques in postharvest quality evaluation. This underscores our contribution as a scalable and efficient solution to the pressing demands of global food supply challenges.

Prediction of oil pipeline process operating parameters based on mechanism and data mining

Abstract Precisely forecasting the operational characteristics of oil pipelines is essential for developing rational design, production, and operation strategies, as well as reducing energy consumption and saving energy. Due to significant disparities in the computation outcomes of conventional mechanism models and the inadequate performance of machine learning models when handling limited sample data, their conclusions likewise lack tangible significance. In this study, a novel physics-guided neural network (PGNN) model, which integrates mechanisms with machine learning models, is introduced. The proposed model incorporates essential physical intermediate factors that impact the temperature and pressure of oil pipelines as artificial neurons within the loss function. Additionally, an adaptive moment estimate approach is employed to optimize the parameters of the model. Through a comparative analysis of various models' predictive capabilities on an oil pipeline, it was shown that PGNN has the highest level of accuracy in forecasting pipeline temperature and pressure. Furthermore, PGNN demonstrates the ability to generate satisfactory prediction outcomes even with a limited sample size. Simultaneously, the predictive outcomes of PGNN exhibit a stronger correlation with variables that have a direct impact on temperature and pressure.

Physics-guided deep learning-based inversion for airborne electromagnetic data

SUMMARY The Earth's subsurface structure provides critical insights into sustainable resource management and geologic evolution. The airborne electromagnetic (AEM) method is an efficient data acquisition technique and can be used to image the underground resistivity structure with high spatial resolution. However, inversion of the increasingly huge volume of AEM data poses a heavy computational burden. In this study, we develop a hybrid deep learning-based approach by using the physics-guided neural network (PGNN) which incorporates the governing physical laws into the loss function to solve the AEM inverse problem. The PGNN integrates the strength of data-driven method for representation learning with electromagnetic laws and allows for the underlying physical constraints to be strictly satisfied. We validate the effectiveness of our approach using both synthetic and field datasets. Compared with the classic Gauss–Newton method, our PGNN inversion system shows strong robustness against multiple noise sources and reduces the risk of being trapped in local extrema. Moreover, the PGNN-inverted results are physically more consistent with the AEM observations compared to the purely data-driven approach. Application to the field AEM data from Northern Australia demonstrates that the PGNN-based inversion framework effectively estimates the subsurface electrical properties with considerable lateral continuity and significantly higher efficiency, completing the inversion of more than 2734000 AEM soundings taking only minutes on a common PC. Our proposed PGNN-based method shows great promise for large-scale underground resistivity imaging, and the well-identified subsurface resistivity structure can effectively improve our understanding of resource distributions and geological hazards.

Development of a semi-empirical physical model for transient NOx emissions prediction from a high-speed diesel engine.

With emissions regulations becoming increasingly restrictive and the advent of real driving emissions limits, control of engine-out NOx emissions remains an important research topic for diesel engines. Progress in experimental engine development and computational modelling has led to the generation of a large amount of high-fidelity emissions and in-cylinder data, making it attractive to use data-driven emissions prediction and control models. While pure data-driven methods have shown robustness in NOx prediction during steady-state engine operation, deficiencies are found under transient operation and at engine conditions far outside the training range. Therefore, physics-based, mean value models that capture cyclic-level changes in in-cylinder thermo-chemical properties appear as an attractive option for transient NOx emissions modelling. Previous experimental studies have highlighted the existence of a very strong correlation between peak cylinder pressure and cyclic NOx emissions. In this study, a cyclic peak pressure-based semi-empirical NOx prediction model is developed. The model is calibrated using high-speed NO and NO2 emissions measurements during transient engine operation and then tested under different transient operating conditions. The transient performance of the physical model is compared to that of a previously developed data-driven (artificial neural network) model, and is found to be superior, with a better dynamic response and low (<10%) errors. The results shown in this study are encouraging for the use of such models as virtual sensors for real-time emissions monitoring and as complimentary models for future physics-guided neural network development.

Federated Learning and Unlearning as Enablers of Wind Turbine Digital Twins

Real-time capable models are paramount for the successful adaptation of digital twin technology into industries such as wind energy, but high-fidelity physics-based models cannot achieve the required speed, while data-driven and hybrid methods require large amounts of training data which is typically confidential. In this work, the combination of federated learning with hybrid modeling is proposed to train fast and reliable models across multiple confidential data sets owned by different stakeholders. The approach is demonstrated on physics-guided neural networks to estimate the lift and drag of wind turbine airfoils. A scenario is devised where multiple confidential data subsets are confined to different client devices. It is shown that the physics-guided neural networks can be trained through federated learning across those devices and data subsets and that the resulting models can recover and even surpass the accuracy of a model that is trained conventionally by merging the data subsets on a single device. The presented approach is highly scalable and can be easily adapted to many other applications. This work also discusses federated unlearning methods, which allow data owners to remove all traces of a data subset used in training if they decide to revoke their contribution.