Experimental Modal Analysis using FDD?

[chaitugk]

I am a beginner to Experimental Modal Analysis and I have a scaled miniature frame and my objective is to conduct modal identification under the structure's operating condition using Frequency Domain Decomposition (FDD).

However, after I expose the structure and take the reading, the results almost have no information about modal parameters.

Questions:

1. In case 1, there is only one peak in the response, whereas the system is a 2DOF system. What could be the reason?

2. In case 2, I could not see any peaks in the 1st singular values of the matrix, where am I going wrong?

3. What is the sensitivity of the acceleration transducers required for scaled experiments? Are the sensors, sensitive enough?

4. Is the frequency response just due to the intrinsic noise of the accelerometer?

5. Also, why is there a peak in the PSD at close to 0hz?

6. Can I use FDD for laboratory based experimental modal analysis without any external random excitation?

7. Should I excite the scaled model with random input force and then do output only modal analysis?

I have attached my setup, results and questions.
https://drive.google.com/file/d/0B-25vDIilD9FM1NuXzNwX2dEY00/view?usp=sharing
Thank you,

ChaitanyaHow can I do modal identification of a scaled miniature steel frames using FDD technique under ambient vibration? - ResearchGate. Available from: https://www.researchgate.net/post/How_can_I_do_modal_identification_of_a_scaled_miniature_steel_frames_using_FDD_technique_under_ambient_vibration [accessed Jul 6, 2015].

 

[Dr. Courtney]

1. Usually, several modes would be expected. But something about the excitation either only excited one mode. Alternatively, the system may have a symmetry such that the modes are degenerate. Another possibility is that the other modes may be at higher frequencies.

2. It seems likely that the ambient conditions are not exciting any modes. You need to raise the white noise excitement in your ambient conditions.

3. Your full scale is 1G. Sensitivity depends on your ADC. An 8 bit ADC will not be as sensitive as a 12 bit ADC.

4. Possibly. What happens in other scenarios? What happens if you drive the structure with a known frequency, say 20 Hz?

5. The peak at 0 Hz results from a small non-zero average in all the data. It corresponds to an offset issue in the circuit. You can usually make it smaller by subtracting the average value of the whole data set from each point, so that the mean is shifted to zero.

6. You need something to put energy into the system if you expect it to shake.

7. Seems like a reasonable idea.

Also, I find it odd that the frequency spectra have units of dB, but the time domain data is in m/s/s.

Why not do a Fourier transform that retains the amplitude units in the frequency domain? You can plot on a log scale, but then you know what the amplitudes of your resonances are.

 

[chaitugk]

1. Usually, several modes would be expected. But something about the excitation either only excited one mode. Alternatively, the system may have a symmetry such that the modes are degenerate. Another possibility is that the other modes may be at higher frequencies.

2. It seems likely that the ambient conditions are not exciting any modes. You need to raise the white noise excitement in your ambient conditions.

3. Your full scale is 1G. Sensitivity depends on your ADC. An 8 bit ADC will not be as sensitive as a 12 bit ADC.

4. Possibly. What happens in other scenarios? What happens if you drive the structure with a known frequency, say 20 Hz?

5. The peak at 0 Hz results from a small non-zero average in all the data. It corresponds to an offset issue in the circuit. You can usually make it smaller by subtracting the average value of the whole data set from each point, so that the mean is shifted to zero.

6. You need something to put energy into the system if you expect it to shake.

7. Seems like a reasonable idea.

Also, I find it odd that the frequency spectra have units of dB, but the time domain data is in m/s/s.

Why not do a Fourier transform that retains the amplitude units in the frequency domain? You can plot on a log scale, but then you know what the amplitudes of your resonances are.

Thank you very much for your invaluable suggestions Dr. Courtney.

I still have some more issues.
First of all, the units of PSD is in dB as it is a conversion of scale for better visualization, i.e if y is the usual amplitude of power density, then ydB=20log₁₀ y
From your suggestions;
1. Correct me if I am wrong; Degenerate modes occur when there are equal natural frequencies or one of the natural frequency is zero. I will check the possibility of other modes in higher frequencies by increasing the sampling rate. (currently, I am using a sampling rate of 500Hz)

2. I felt the same that the ambient conditions are not exciting any modes. How do I increase the ambient noise? Is there a simple way to do that without affecting the broadband assumptions in system identification techniques?

3. I am using EDX100A Universal Recorder as ADV and CDV 40A type of conditioner card. This has a resolution of 16bit; should this be enough? however, it does not have an anti-aliasing filter.

4. I will try this.

5. Actually, the peak is not exactly at 0hz but is close to 0hz. If it is an offset issue, how exactly can I subtract the average value? should I use a High Pass Filter?

6. How can I put energy into the system effectively? will inappropriate addition cause harmonics into the system?

 

[Dr. Courtney]

1. The upper end of your accelerometer frequency response if 40 Hz. Sampling faster won't catch higher frequency modes if the accelerometer is not sensitive to frequencies above 40 Hz.

2. I don't know. But you are doing a variation of a resonance experiment. To find resonances, you need excitation.

3. 16 bit ADC resolution is plenty.

5. To remove offset, past the data into a spreadsheet, compute the mean for the amplitude column, subtract the mean amplitude from the original amplitude creating a new column of amplitude data with zero mean offset.

6. One way that it is done in other vibration studies is to use a vibration table.