romAI verification corner (8) Can you make a ROM for acceleration input? (English Version)
(This is a translated version of a blog, posted by Kosuke Ikeda. The original blog in Japanese can be found here).
Let's check the function of romAI using a simple model. When using actual measurement data, there are many cases where you want to use acceleration, so for the 8th experiment, we will check if you can make a ROM for acceleration input.
Problem setting
In Part 7, we created a dynamic model of a force input spring mass damper system. This time, let's consider the problem of acceleration, velocity, and displacement input. Consider predicting the acceleration of the mass when the boundary on one side is forcibly moved with a 1-degree-of-freedom spring mass damper system.
The model attached to the knowledge article below will be used. By setting N = 1, it is used as a spring mass damper system with one degree of freedom.
Knowledge Article: 1D model of N degrees of freedom spring mass
An acceleration response that resonates after 15 seconds has been obtained for an acceleration input that increases in amplitude and frequency. Let's create a romAI model that reproduces this input / output relationship.
romAI settings
The input is the acceleration in the boundary, the output is the acceleration of the mass, and the state quantity is the displacement and velocity of the mass. The concept of state quantity was explained in Part 7, so I will omit it.
Inputs: accin
Outputs: acc
States: x, v
Since the Activate model is linear, I chose Model Type: linear.
Since the data is noise-free, the data is not divided, referring to Part 6.
Acceleration input calculation result
We compare the acceleration history of the mass from romAI with the original waveform. Unfortunately, it's completely unpredictable.
So what was wrong? As we observed in Part 7, the force input worked well, so this case should also work if the input is a spring damper load. The spring damper load is determined by the relative displacement and velocity of the boundary and mass. Therefore, if we have the displacement and velocity information of the boundary, not the acceleration of the boundary, we should be able to predict the spring damper load internally and, as a result, the acceleration of the mass.
Reconfigure romAI
Change the input to be boundary displacement, velocity instead of boundary acceleration.
Inputs: xin, vin
Outputs: acc
States: x, v
Other settings are the same.
Displacement / velocity input calculation result
We compare the acceleration history of the mass from romAI with the original waveform. This time it was a perfect match.
Summary
This time, I experimented with creating a ROM for acceleration input. Generally, the boundary is connected to other parts by stiffness and damping, so the input load depends not on acceleration but on displacement and velocity. Therefore, ROM can be created correctly by inputting displacement and velocity instead of acceleration.
In the case where it is connected only by stiffness, I believe that the displacement input is enough, and in the case where it is connected only by damping, it seems that only the velocity input is enough. Please try each one.