An AI-Augmented Multiphysics approach to speed up the analysis of a Solenoid Valve
A solenoid valve is a simple on/off valve commonly used in many applications like industrial water control, car wash and laundry equipment, but also hydraulic and pneumatic systems. These important, if often overlooked, components are made up of two basic functional units – the valve body and the solenoid electromagnet, which when energized or deenergized shuts off or allows flow through the valve orifices.
Compared to many other highly complex components that engineers are tasked with designing and validating through simulation, the humble solenoid valve might appear inconsequential by comparison. Why focusing on optimizing something that seemingly has little room for improvement?
Often, product engineers view the function of a solenoid valve through a limited CFD lens. By optimizing the flow path through the valve, you’ve optimized the component. But the actuation of the valve also plays an important role for reliability and safety. The actuation problem, however, is much more complex, requiring an understanding of the interaction between magnetics, hydraulics, fluid dynamics, and mechanics. To optimize the solenoid valve actuation, a multiphysics approach is needed to consider all forces, leading to an optimal design of the entire system.
And what else, if not Altair Activate®?
Activate allows engineers to bring together mixed fidelity models from all domains – electromagnetics, mechanics, hydraulics, and fluid dynamics, into a unified simulation of the entire system. Convenient parameterization and simulation control allows engineers to visualize and adjust essential settings, boundary conditions, and sub-models.
Mechanics
The mechanics of the valve is very simple. It only consists of the mass of the valve’s plunger and a spring. Magnetic force, pressure forces, and flow forces can be calculated in their own domain model and added to the connections.
Electromagnetics
Although the electromagnet is the heart of the solenoid valve and also the subject of the actual investigation, it shall also be mentioned here only briefly. The electromagnetic analysis is solved using Flux®. Because of the tight integration between Activate and Flux, users can perform high-accuracy electromagnetic co-simulation or choose to run quicker simulations by leveraging look-up tables or functional mock-up units (FMUs).
CFD
In the hydraulic part of the valve, two aspects are important – flow and forces. Forces are generated by static pressure, but also by locally decreasing dynamic pressure (flow forces). The model is represented by a combination of hydraulics components and the results of CFD simulations with AcuSolve®.
Look-up tables are used for the flow and forces that represent the steady-state behavior. The flow is modeled and solved using fixed conditions, exploring several plunger positions and different flows. The results of those steady-state simulations are transferred to a look-up table, where the combined data can be examined to determine the mass flow and forces depending on pressure differences and plunger positions.
For a more accurate representation of the phenomenon dynamic reduced-order models (ROMs) are then utilized to represent the solenoid valve’s transient fluid dynamics behavior.
CFD, tackled with romAI
First of all, what is romAI?
romAI is a modeling technique which combines classical system theory and machine learning to produce fast and accurate reduced order models
In the case of a complete system simulation, such as the current electro-valve one, it is often sufficient to reduce component behavior to its interaction with the complete system, improving solver run time while still providing sufficiently accurate results.
By leveraging romAI for the solenoid valve simulation, the CFD 3D model of the valve can be turned into an Altair Activate romAI block:
whose input/output ports can be connected to interact with the complete system.
Hence, the romAI block, which embeds the CFD physics, receives as inputs:
- The plunger position;
- The plunger velocity;
- The delta pressure across the valve.
And provides as outputs:
- The hydraulic force (static hydraulic force + flow force);
- The fluid flow (through the valve).
Generating the romAI model is a straigthforward process, which articulates in the following sequential steps:
1. Perform transient 3D CFD simulations with Altair AcuSolve
4 CFD Transient simulations were run at 4 different operational conditions, given by the combinations of the following parameters:
-
- 2 level of inlet pressure (2 [bar], 4 [bar])
- 2 opening times (10 [ms], 100 [ms])
The romAI model is expected to work in the operational range defined by the just reported parameters.
1 additional CFD Transient simulation was run, for accuracy test purposes, with the following operational conditions:
-
- Inlet pressure equal to 3[bar]
- Opening time equal to 50[ms]
2. Extract the training data from the 3D simulation
The training dataset consists of a unique csv file (which concatenates together all 4 CFD simulations), whose 1st column is the time vector, while the other columns are the inputs, outputs and states of the romAI model.
The states are those variables which describes the dynamics of the reduced order model. In our case, we have set a unique state: the flow through the valve (which indeed is also an output).
Instead, inputs and outputs have been previously defined.
3. Run the machine learning module
This step is entirely carried over trough the romAI app:
Where the user can:
-
- Set inputs, outputs, states
- Define the mathematical model
- Define the training params
- And of course, train the romAI model!
4. Include the romAI model in the romAI block
Leveraging the romAI block, the romAI model can be seamlessly deployed in Activate:
Its accuracy can be evaluated using the dataset generated from the additional CFD simulation (with 3 [bar] as inlet pressure and 50 [ms] as opening time):
5. Use the ROM in Activate for system simulation
Just connect the romAI block to other system blocks through its input/output ports.
By leveraging romAI for the solenoid valve simulation, users get the best of both worlds – accuracy of 3D models with the speed of 1D models.
Results of the AI-Augmented Multiphysics Analysis
A co-simulation with AcuSolve provides the most accurate results. But who has time to wait 18 h for a single simulation run on a 4 core/8 threads PC when optimizing the solenoid? romAI helps to reduce the time to a few seconds while still providing sufficient accuracy even in transient processes.
