Abstract:This paper introduces a Physics-Informed model architecture that can be adapted to various backbone networks. The model incorporates physical information as additional input and is constrained by a Physics-Informed loss function. Experimental results demonstrate that the additional input of physical information substantially improve the model's ability with a increase in performance observed. Besides, the adoption of the Softplus activation function in the final two fully connected layers significantly enhances model performance. The incorporation of a Physics-Informed loss function has been shown to correct the model's predictions, bringing the back-projections closer to the actual inputs and reducing the errors associated with inversion algorithms. In this work, we have developed a Phantom Data Model to generate customized line integral diagnostic datasets and have also collected SXR diagnostic datasets from EAST and HL-2A. The code, models, and some datasets are publicly available at <a class="link-external link-https" href="https://github.com/calledice/onion" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to solve the problem of fast and accurate inverse reconstruction in the line integral diagnostics system in Tokamak devices. Specifically, the author has developed a Physics - Informed Deep Learning Model to improve the ability to diagnose and understand the internal state of plasma. #### Main problems include: 1. **Limitations of traditional methods**: - Traditional nuclear fusion diagnosis and inverse reconstruction require accurate physical models. These models are based on first - principles and consider factors such as spatio - temporal dynamics, multi - physical fields and system coupling. This results in high model complexity and long computation time. - Existing data - driven surrogate models lack physical interpretability and cross - device transferability, which limits their wide application. 2. **Improving model performance and physical consistency**: - Introduce physical information as an additional input and constrain the model through a physical - information loss function to improve the model's prediction accuracy and physical consistency. - Propose a new Phantom Data Model for generating customized line - integral diagnostic data sets, ensuring data quality and reducing algorithm errors. 3. **Enhancing the generalization ability and transferability of the model**: - Design a flexible model architecture that can adapt to different backbone networks, such as VGG, ResNet, etc., thereby improving the model's generalization ability and transferability between different Tokamak devices. - By introducing the Softplus activation function and optimizing the loss function, further improve the model performance, especially in terms of edge smoothness and overall prediction accuracy. 4. **Promoting the construction of data and model ecosystems**: - Make training data, models and code public to promote the development of the research community in the field of nuclear fusion and promote more high - quality research results. Through the above methods, this paper attempts to construct an efficient, transferable and physically consistent deep - learning model to meet the complex diagnostic challenges in nuclear fusion research and provide faster and more reliable decision - support for future experiments.

ONION: Physics-Informed Deep Learning Model for Line Integral Diagnostics Across Fusion Devices

Physics Embedded Weight-Sharing Neural Network for Electrical Impedance Tomography

Hybrid deep learning and physics-based neural network for programmable illumination computational microscopy

DeepPhysics: a physics aware deep learning framework for real-time simulation

Supplementation of deep neural networks with simplified physics-based features to increase model prediction accuracy

Enhanced physics‐informed neural networks for hyperelasticity

Multi-Fidelity Physics-Constrained Neural Network and Its Application in Materials Modeling

Data-Enabled Physics-Informed Machine Learning for Reduced-Order Modeling Digital Twin: Application to Nuclear Reactor Physics

A Two-Stage Imaging Framework Combining CNN and Physics-Informed Neural Networks for Full-Inverse Tomography: A Case Study in Electrical Impedance Tomography (EIT)

Interfacing finite elements with deep neural operators for fast multiscale modeling of mechanics problems

Extrapolation of Physics-Inspired Deep Networks in Learning Robot Inverse Dynamics

Electrical Impedance Tomography Image Reconstruction With Attention-Based Deep Convolutional Neural Network

A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics

DIMOND: DIffusion Model OptimizatioN with Deep Learning

Physics-Informed Neural Networks for Prognostics and Health Management of Lithium-Ion Batteries

Physics-Informed Transformation Toward Improving the Machine-Learned NLTE Models of ICF Simulations

PINNsFormer: A Transformer-Based Framework For Physics-Informed Neural Networks

Dual-energy CT based mass density and relative stopping power estimation for proton therapy using physics-informed deep learning

Physics-Informed Deep Ai Simulation For Fractal Integro-Differential Equation

Physics-Informed Neural Networks for Electrical Circuit Analysis: Applications in Dielectric Material Modeling