Abstract:Scientific machine learning and the advent of the Physics-Informed Neural Network (PINN) have shown high potential in their ability to solve complex differential equations. One example is the use of PINNs to solve the gravity field modeling problem -- learning convenient representations of the gravitational potential from position and acceleration data. These PINN gravity models, or PINN-GMs, have demonstrated advantages in model compactness, robustness to noise, and sample efficiency when compared to popular alternatives; however, further investigation has revealed various failure modes for these and other machine learning gravity models which this manuscript aims to address. Specifically, this paper introduces the third generation Physics-Informed Neural Network Gravity Model (PINN-GM-III) which includes design changes that solve the problems of feature divergence, bias towards low-altitude samples, numerical instability, and extrapolation error. Six evaluation metrics are proposed to expose these past pitfalls and illustrate the PINN-GM-III's robustness to them. This study concludes by evaluating the PINN-GM-III modeling accuracy on a heterogeneous density asteroid, and comparing its performance to other analytic and machine learning gravity models.

What problem does this paper attempt to address?

The main problems that this paper attempts to solve are various deficiencies in existing gravity models when dealing with complex celestial bodies (such as asteroids and comets), especially in terms of high precision, low computational cost, and the challenges of dealing with irregular - shaped objects. Specifically, the paper focuses on: 1. **Feature Divergence**: Existing models may produce unstable results in certain specific regions or under certain conditions. 2. **Low - Altitude Sample Bias**: The model has a preference for low - altitude samples, resulting in a decrease in the prediction accuracy of other regions. 3. **Numerical Instability**: In some cases, the numerical solutions of the model may become unstable. 4. **Extrapolation Error**: When the model is used in situations beyond the range of training data, its prediction results may be inaccurate. 5. **Non - Compliance of Boundary Conditions**: The model may not be able to meet the physical boundary conditions. To solve these problems, the paper introduces the third - generation physics - informed neural network gravity model (PINN - GM - III). Through a series of design improvements, PINN - GM - III aims to overcome the above problems and provide a more robust method for gravity field modeling. In addition, the paper also proposes six evaluation indicators to reveal the defects of past models and demonstrates the superior performance of PINN - GM - III on these six indicators. ### Specific Improvement Measures 1. **Solve Feature Divergence**: By improving the network structure and loss function design, ensure that the model can remain stable in different regions. 2. **Reduce Low - Altitude Sample Bias**: Optimize the data sampling strategy so that the model can better handle data at different altitudes. 3. **Improve Numerical Stability**: Introduce regularization terms and other numerical optimization techniques to enhance the stability of the model under extreme conditions. 4. **Reduce Extrapolation Error**: Use a broader data set for training and combine physical constraints to improve the generalization ability of the model. 5. **Ensure Compliance of Boundary Conditions**: By introducing an additional boundary condition loss term, ensure that the model output conforms to physical laws. ### Conclusion By comparing the performance of PINN - GM - III with other analytical and machine - learning gravity models, the paper verifies the modeling accuracy and robustness of the new model on complex celestial bodies such as asteroids with heterogeneous densities. These improvements make PINN - GM - III a more potential tool for gravity field modeling, suitable for a variety of application scenarios.

The Physics-Informed Neural Network Gravity Model: Generation III

PINN Based Data Driven Magnetics Loss Modeling

Machine Learning Through Physics–Informed Neural Networks: Progress and Challenges

Scientific Machine Learning Through Physics–Informed Neural Networks: Where we are and What's Next

GRINN: A Physics-Informed Neural Network for solving hydrodynamic systems in the presence of self-gravity

Modelling Force-Free Neutron Star Magnetospheres using Physics-Informed Neural Networks

Learning Gravity Fields of Small Bodies: Self-adaptive Physics-informed Neural Networks

Improved physics-informed neural network in mitigating gradient related failures

Numerical analysis of physics-informed neural networks and related models in physics-informed machine learning

Learning Physics-Informed Neural Networks Without Stacked Back-propagation.

Quantifying local and global mass balance errors in physics-informed neural networks

Recipes for when physics fails: recovering robust learning of physics informed neural networks

Advancing Physics Data Analysis through Machine Learning and Physics-Informed Neural Networks

Physics-informed neural networks for gravity currents reconstruction from limited data

An Investigation of Physics Informed Neural Networks to solve the Poisson-Boltzmann Equation in Molecular Electrostatics

Characterizing possible failure modes in physics-informed neural networks

Inverse Physics-Informed Neural Networks for transport models in porous materials

Modeling of additive manufacturing processes with time‐dependent material properties using physics‐informed neural networks

Physics-informed neural networks based on adaptive weighted loss functions for Hamilton-Jacobi equations

Neuroevolution of Physics-Informed Neural Nets: Benchmark Problems and Comparative Results

Gradient-Annihilated PINNs for Solving Riemann Problems: Application to Relativistic Hydrodynamics