Physics-informed neural networks
Assembly Line
📐 UCLA Researchers Propose PhyCV: A Physics-Inspired Computer Vision Python Library
In the latest innovation, Jalali-Lab @ UCLA has developed a new Python library called PhyCV, which is the first Physics-based Computer vision Python library. This unique library uses algorithms based on the laws and equations of physics to analyze pictorial data. These algorithms imitate how light passes through several physical materials and are based on mathematical equations rather than a series of hand-crafted rules. The algorithms in PhyCV are built on the principles of a rapid data acquisition method called the photonic time stretch.
The three algorithms included in PhyCV are – Phase-Stretch Transform (PST) algorithm, Phase-Stretch Adaptive Gradient-Field Extractor (PAGE) algorithm, and Vision Enhancement via Virtual diffraction and coherent Detection (VEViD) algorithm.
Hybrid AI-Powered Computer Vision Combines Physics and Big Data
Many computer vision techniques infer properties of our physical world from images. Although images are formed through the physics of light and mechanics, computer vision techniques are typically data driven. This trend is mostly performance related: classical techniques from physics-based vision often score lower on metrics compared with modern deep learning. However, recent research, covered in this Perspective, has shown that physical models can be included as a constraint into data-driven pipelines. In doing so, one can combine the performance benefits of a data-driven method with advantages offered from a physics-based method, such as intepretability, falsifiability and generalizability. The aim of this Perspective is to provide an overview into specific approaches for integrating physical models into artificial intelligence pipelines, referred to as physics-based machine learning. We discuss technical approaches that range from modifications to the dataset, network design, loss functions, optimization and regularization schemes.
Physics-Driven Generative Design: The Future of Engineering
Introduction to Hybrid Modelling for Digital Twins
Physics-informed Machine Learning (PIML) involves embedding established domain knowledge (i.e. physics, chemistry, biology) with machine learning (ML) to effectively model dynamic industrial systems. While these dynamic systems face challenges such as high sensor noise and sparse measurements, they often are characterized by some fundamental scientific/engineering knowledge. There are 3 general ways to embed domain knowledge with ML, including:
- Introducing observational bias to the data
- Introducing inductive bias into the model structure
- Introducing learning bias to how models are trained
Physics-informed neural networks (PINNs) are a novel approach that integrate the information from both process data and engineering knowledge by embedding the ODEs into the loss function of a neural network. PIML integrates data and mathematical models seamlessly even in noisy and high- dimensional contexts.Thanks to its natural capability of blending physical models and data as well as the use of automatic differentiation, PIML is well placed to become an enabling catalyst in the emerging era of digital twins.
With physics-informed AI, machine operators can trust and verify
The first PINN applications are emerging in manufacturing processes with complex models and relations, such as in additive manufacturing, Van der Auweraer said.
Other early adopters will be in the food industry or pharmaceutical processing industry where complex processes may hinder a pure simulation-based approach and where the AI in a PINN approach may yield promising results, Van der Auweraer and Mas said.
PINN models also can complement or replace labor-intensive lab testing and design, Mas said, combining the existing strengths of lab testing and the benefits of physics-based simulations to accurately design new material and products in much less time using less lab testing.
Physics-Informed Neural Networks (PINNs) for Improving a Thermal Model in Stereolithography Applications
Stereolithography (SLA), additive manufacturing (3D printing) technique, is widely used nowadays for rapid prototyping and manufacturing (RP & M). This technique is driven by photo-polymerisation, which is an exothermal process. This may lead to thermal stresses significantly affecting the final quality of printed parts/products. To guarantee high-quality parts printed with the SLA technique, understanding the thermal behaviour is therefore crucial for optimizing the process. In this paper, the recent physics-informed neural network (PINN) methodology was employed to improve a physics-based model for predicting the thermal behaviour of SLA processes. The accuracy of the improved thermal model is demonstrated in this paper by comparing the predicted 2D temperature field with the 2D temperature field measured by a high-speed infrared thermal camera on parts printed on a production machine.