# Impact acceleration too high

Member

I'm doing an impact simulation. The impactor is a 5cm diameter sphere made from steel 4340, moving with velocity 1m/s. The target is a 20cmx20cm plate with thickness 5mm, also made from steel 4340. The probe acceleration is picked at the center of the plate. The peak acceleration values are very strange. They raise up to thousands of G. Could such slow impact make too high peak acceleration?

• Member

This discussion talks about force, but you can apply the same filtering to acceleration data since F = ma.

https://forum.ansys.com/discussion/comment/99466#Comment_99466

• Member
edited December 2020

@peteroznewman Thanks Peter. In my simulation, the end time is equal to 0.01 s, number of output points is 2000, so sampling frequency is fs = 2x10^5 Hz. Cut-off frequency is set to fc = 2000 Hz. Normalized frequency fn = fc / (fs / 2) = 0.02. I use Butterworth low pass filter from scipy package to filter out the acceleration. The filter order is set to 5. Now, the filtered data is smooth, but the values are still high. Could you explain how to properly choose sampling frequency, cut-off frequency and filter order in such low speed impact to obtain appreciate results? Are there instructions about choosing these parameters in an Explicit Dynamics simulation?

• Member
edited December 2020

Since you do not know the maximum frequency in the signal, to avoid the risk of aliasing, you should write the result data on every cycle. This will create a very large result file.

If the time step is 1e-8 s and your end time is 0.01 s you will have 1 million samples in the result file. Make sure you have plenty of disk space. Increase the time step by raising the minimum Characteristic Length in the mesh if you can.

Low pass filter the original data, then you can decimate the filtered data to get a more manageable number of points to plot. This advice is based on section 2.3 of the paper linked to below.

https://bodietech.com/pdfs/DSP_for_Explicit_FEA_AUC1999.pdf

Read the whole paper. It has examples that show how sampling the raw Explicit data, which is what you are doing when you set output points to 2000, aliases the data and can cause smooth curves to come out of noisy data, but those smooth curves can be very wrong! Low pass filtering aliased data doesn't fix the problem.