Fluid Flow Assessment Using PINNs for Particle Tracking Uncertainty
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional particle tracking velocimetry (PTV) methods suffer from significant uncertainties in particle localization and tracking, leading to inaccuracies in velocity and pressure field estimates, particularly in high-speed and multiphase flows where inertial particle transport effects are prevalent, and existing systems fail to accurately account for these non-ideal advection effects.
Innovation Solution
A physics-informed neural network (PINN) is employed to simultaneously estimate velocity and pressure fields from error-laden particle positions, incorporating a physics loss to satisfy fluid motion equations and a data loss to compensate for inertial particle transport and localization uncertainties, using particle advection velocimetry (PAV) and stochastic particle advection velocimetry (SPAV) techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional particle tracking velocimetry (PTV) methods are used to track particle positions, then velocity fields can be estimated, but significant uncertainties in particle localization and tracking lead to inaccuracies in velocity and pressure field estimates
Solution Approach 1:
The patent replaces conventional mechanical particle tracking methods with a physics-informed neural network (PINN) that uses deep learning to predict particle positions. This substitution allows the system to handle noisy and incomplete particle trajectory data more effectively, reducing localization uncertainties and improving velocity field estimation accuracy by learning from physical principles embedded in the neural network architecture.
Solution Approach 2:
The patent changes the approach from direct particle position measurement to using a neural network model that predicts particle positions based on learned patterns from training data. This parameter transformation allows the system to compensate for measurement noise and tracking errors by using statistical learning methods instead of deterministic tracking algorithms.
2Ease of manufacture
If particles are used as tracers in high-speed and multiphase flows, then flow visualization is achieved, but inertial particle transport effects cause particles to lag the carrier fluid or travel ballistically, resulting in slip velocity
Solution Approach 1:
The patent introduces an intermediary correction model that accounts for inertial particle transport effects. This model acts as a mediator between the observed particle positions and the actual fluid velocity, using physical principles of particle-fluid interaction to correct for slip velocity caused by particle inertia, drag, and other non-ideal advection effects.
Solution Approach 2:
The patent replaces the assumption that particles perfectly follow fluid flow with a physics-based correction approach using neural networks. This substitution allows the system to explicitly model and correct for inertial effects, replacing the naive tracer particle assumption with a more accurate physical model of particle-fluid interaction.
3Productivity
If particle positions with localization errors are used for velocity field estimation, then flow assessment can be performed, but inaccuracies propagate through the velocity and pressure field calculations
Solution Approach 1:
The patent implements feedback through the physics-informed neural network architecture, where the predicted velocity and pressure fields are continuously refined by comparing with observed particle positions and enforcing physical consistency through loss functions. This feedback mechanism allows errors in particle localization to be compensated by adjusting the predicted fields to satisfy both the observed data and the underlying physics equations.
Solution Approach 2:
The patent performs preliminary training of the neural network on synthetic particle trajectory data before applying it to experimental measurements. This preliminary action allows the model to learn robust patterns of particle motion and error characteristics, preparing it to handle noisy real-world data and reducing the propagation of localization errors into the final velocity and pressure field estimates.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system provides highly accurate 3D velocity and pressure field assessments, even in noisy and error-prone environments, by explicitly modeling particle advection and accounting for localization and tracking uncertainties, thereby improving the accuracy of flow field estimates.
Implementation Method 1
particle advection velocimetry (PAV) and stochastic particle advection velocimetry (SPAV) techniques
Data Source
AI summary
A system for fluid flow assessment that includes a computer device having a processor connected to a non-transitory computer readable medium configured to receive image data from at least one camera device positioned to capture images of a flow of fluid passing through a region of interest. The computer device configured can be to perform a fluid flow assessment process by running code stored in the non-transitory computer readable medium defining the fluid flow assessment process to assess the fluid flow and/or particles within the fluid. Embodiments of a process for fluid flow and/or particle assessment can utilize camera image data as well for performing the assessment. A computer device can be configured to facilitate the assessment of such data.