Tweet
I posted this question on BIOMCH-L last week:
"
I'm interested in the force plate data collection practices used by other motion labs. For example, what sampling frequency is used? Are the data filtered in any way? We use two AMTI OR6 force plates in our lab and often see suspicious 'spikes' in our force plate data immediately after initial contact. I'm not sure if these spikes are artificial or if they are real data caused by the impact of heal strike. Do others see this? We sample at 1080Hz and do not filter our signal.
"
These are the responses I received:
------------------------
These spikes are representing heel strike and are very important to the human musculoskeletal system.
Do NOT please filter them out. It is real.
------------------------
Could they be impact transients caused by the "wobble mass" of the acceleration of the organs within the body?
------------------------
We are filtering our analog signals with a zero-lag low pass Butterworth filter using Visual 3D tech. Cut-off frequency is 10 Hz, this will not attenuate the initial impact spike, but should remove other noise. Related note, we collect at a much lower sampling rate. 200 Hz captures all relevant locomotion events.
------------------------
The spikes may be real or artifacts. The "real" question is, how are you going to use the data? If you are doing inverse dynamics to derive what the muscles are doing, then the spikes are artifacts and you should filter with a 10 Hz (or thereabouts) low-pass filter. If you are doing impact analyses, then you should find out if the spikes are real or caused by poor shimming of the force plate to the ground or some other factor. This is difficult to determine. A mechanical engineer might help to determine whether the person or the platform are causing the spikes.
------------------------
It sounds like "heel strike" but a quick test would be to bump the ADC sample rate up as high as it will go and gently tap once on the plate with a small rubber tipped hammer. For bonus points, put a marker on the hammer and you'll have a trajectory that indicates exactly when the impulse is delivered.
------------------------
This may be a really silly question, but have you tried running a simple FFT via Matlab to examine the power spectrum of your signal? If you know the signal content, then you can justify any filtering such that it satisfies the Nyquist Criterion.
We've done this sort of thing with impact tests, where the SAE standards call for less aggressive filtering and thus data with poor visual appeal. By running the FFT we are able to justify our more aggressive filtering and have outputs free of spikes such as you mentioned.
------------------------
I am a researcher at teesside university and we use Kistler forceplates portable and ones fixed into the ground. We usually sample at 1000hz and do high pass filtering. Spikes in the data may be due to movement if the force plate is on and unstead surface and picking up some cross talk. If the rest of the data is good possibly dont use the initial section.
------------------------
I used AMTI OR6 in other University. We collected FRS data with sampling rate of 100Hz. It was for gait analysis and COP measurements.
------------------------
We too use the AMTI OR6 force plate. It is possible that the spike is the transient of heelstrike. However, how are the force plates embedded?
It is also possible that the spike you see is an artifact that results from movement of the forceplate under the influence of heelstrike. I have had a similar problem since our plates are not fixed to the floor (unfortunately) and could become unstable due to dirt getting underneath them.
------------------------
Recording at 1000 Hz or around is fine, but filtering is necessary when you are to do inverse dynamics.
These spikes are not true artifacts, but the impact of the foot on the force plate, F = m_foot.a_foot . Problem is that the foot acceleration cannot be calculated (from kinematics recorded at 50 or 100 Hz) with a similar time resolution as the force plate allows. The usual programs use the filtered accelerations and the unfiltered or slightly filtered (100 Hz or so) force plate data together in the calculation of joint moments etc. Bisseling and I have studied this problem, J. Biomechanics 39:2439-2444 (2006). Our recommendation is to filter the force plate data with the same cut-off as the kinematic data, some 4-10 Hz. It seems wasteful to buy a force plate that can record up to 100 or 120 Hz, and then to LP-filter those valuable signals at 4 Hz, but that is the best way in our opinion. Another option is to filter neither optokinematic, neither force plate data, but the calculated moments.
------------------------
We use dual AMTI force plates aswell recording at 1000Hz and see similar spikes, Id love to hear any feedback you get on the matter.
------------------------
In response to your question about sampling frequency, I seem to recall that whilst I was at the Oxford Orthopaedic Engineering Centre in the 80's Eric Radin and Mike Whittle did a study of heelstrike transient by sampling at 1000 Hz, and I think they were mystified to find that some subjects had a high peak on heelstrike and some didn't. I don't know if they published it, but believe they concluded that some people were "slappers" and some were "creepers" and the normal population was randomly divided this way.
------------------------
Thats the same frequency we use since we have 60 fps cameras (Best to combine the camera fps and FP freq in a multiple). No filter also. There should be a spike at contact but no more than 2-4 times BW for walking or running. If you have a jpg you could send I could give you better feedback.
------------------------
We observe these 'spikes' spikes as well. If we are talking about the same thing, this is defined as the weight acceptance phase of stance during running.
See Besier et al 2002 for a full description.
Note: from pilot work at UWA we find that when filtering kinematic and kinetic data prior to Inverse Dynamics you should use the same LP cut off frequencies to avoid discontinuities between your measurements, which theoretically will reduce errors in the ID process.
------------------------
Adam Miller
Motion Analysis Center Engineer
Mary Free Bed Rehabilitation Hospital
235 Wealthy SE Grand Rapids MI 49503
Phone: 616.493.9770
Fax: 616.493.9834
www.MaryFreeBed.com
Winner of the 2009 Michigan Quality Leadership Award
__________________________________________________ ____________________
Confidentiality Note: This e-mail, and any attachment(s) is intended for use only by the individual or entity to which it is addressed and may contain information that is privileged, confidential, and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient or the employee or agent responsible for delivering the message to the intended recipient, you are hereby notified that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and destroy the material in its entirety, whether electronic or hard copy.
Thank you.
__________________________________________________ ____________________
"
I'm interested in the force plate data collection practices used by other motion labs. For example, what sampling frequency is used? Are the data filtered in any way? We use two AMTI OR6 force plates in our lab and often see suspicious 'spikes' in our force plate data immediately after initial contact. I'm not sure if these spikes are artificial or if they are real data caused by the impact of heal strike. Do others see this? We sample at 1080Hz and do not filter our signal.
"
These are the responses I received:
------------------------
These spikes are representing heel strike and are very important to the human musculoskeletal system.
Do NOT please filter them out. It is real.
------------------------
Could they be impact transients caused by the "wobble mass" of the acceleration of the organs within the body?
------------------------
We are filtering our analog signals with a zero-lag low pass Butterworth filter using Visual 3D tech. Cut-off frequency is 10 Hz, this will not attenuate the initial impact spike, but should remove other noise. Related note, we collect at a much lower sampling rate. 200 Hz captures all relevant locomotion events.
------------------------
The spikes may be real or artifacts. The "real" question is, how are you going to use the data? If you are doing inverse dynamics to derive what the muscles are doing, then the spikes are artifacts and you should filter with a 10 Hz (or thereabouts) low-pass filter. If you are doing impact analyses, then you should find out if the spikes are real or caused by poor shimming of the force plate to the ground or some other factor. This is difficult to determine. A mechanical engineer might help to determine whether the person or the platform are causing the spikes.
------------------------
It sounds like "heel strike" but a quick test would be to bump the ADC sample rate up as high as it will go and gently tap once on the plate with a small rubber tipped hammer. For bonus points, put a marker on the hammer and you'll have a trajectory that indicates exactly when the impulse is delivered.
------------------------
This may be a really silly question, but have you tried running a simple FFT via Matlab to examine the power spectrum of your signal? If you know the signal content, then you can justify any filtering such that it satisfies the Nyquist Criterion.
We've done this sort of thing with impact tests, where the SAE standards call for less aggressive filtering and thus data with poor visual appeal. By running the FFT we are able to justify our more aggressive filtering and have outputs free of spikes such as you mentioned.
------------------------
I am a researcher at teesside university and we use Kistler forceplates portable and ones fixed into the ground. We usually sample at 1000hz and do high pass filtering. Spikes in the data may be due to movement if the force plate is on and unstead surface and picking up some cross talk. If the rest of the data is good possibly dont use the initial section.
------------------------
I used AMTI OR6 in other University. We collected FRS data with sampling rate of 100Hz. It was for gait analysis and COP measurements.
------------------------
We too use the AMTI OR6 force plate. It is possible that the spike is the transient of heelstrike. However, how are the force plates embedded?
It is also possible that the spike you see is an artifact that results from movement of the forceplate under the influence of heelstrike. I have had a similar problem since our plates are not fixed to the floor (unfortunately) and could become unstable due to dirt getting underneath them.
------------------------
Recording at 1000 Hz or around is fine, but filtering is necessary when you are to do inverse dynamics.
These spikes are not true artifacts, but the impact of the foot on the force plate, F = m_foot.a_foot . Problem is that the foot acceleration cannot be calculated (from kinematics recorded at 50 or 100 Hz) with a similar time resolution as the force plate allows. The usual programs use the filtered accelerations and the unfiltered or slightly filtered (100 Hz or so) force plate data together in the calculation of joint moments etc. Bisseling and I have studied this problem, J. Biomechanics 39:2439-2444 (2006). Our recommendation is to filter the force plate data with the same cut-off as the kinematic data, some 4-10 Hz. It seems wasteful to buy a force plate that can record up to 100 or 120 Hz, and then to LP-filter those valuable signals at 4 Hz, but that is the best way in our opinion. Another option is to filter neither optokinematic, neither force plate data, but the calculated moments.
------------------------
We use dual AMTI force plates aswell recording at 1000Hz and see similar spikes, Id love to hear any feedback you get on the matter.
------------------------
In response to your question about sampling frequency, I seem to recall that whilst I was at the Oxford Orthopaedic Engineering Centre in the 80's Eric Radin and Mike Whittle did a study of heelstrike transient by sampling at 1000 Hz, and I think they were mystified to find that some subjects had a high peak on heelstrike and some didn't. I don't know if they published it, but believe they concluded that some people were "slappers" and some were "creepers" and the normal population was randomly divided this way.
------------------------
Thats the same frequency we use since we have 60 fps cameras (Best to combine the camera fps and FP freq in a multiple). No filter also. There should be a spike at contact but no more than 2-4 times BW for walking or running. If you have a jpg you could send I could give you better feedback.
------------------------
We observe these 'spikes' spikes as well. If we are talking about the same thing, this is defined as the weight acceptance phase of stance during running.
See Besier et al 2002 for a full description.
Note: from pilot work at UWA we find that when filtering kinematic and kinetic data prior to Inverse Dynamics you should use the same LP cut off frequencies to avoid discontinuities between your measurements, which theoretically will reduce errors in the ID process.
------------------------
Adam Miller
Motion Analysis Center Engineer
Mary Free Bed Rehabilitation Hospital
235 Wealthy SE Grand Rapids MI 49503
Phone: 616.493.9770
Fax: 616.493.9834
www.MaryFreeBed.com
Winner of the 2009 Michigan Quality Leadership Award
__________________________________________________ ____________________
Confidentiality Note: This e-mail, and any attachment(s) is intended for use only by the individual or entity to which it is addressed and may contain information that is privileged, confidential, and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient or the employee or agent responsible for delivering the message to the intended recipient, you are hereby notified that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and destroy the material in its entirety, whether electronic or hard copy.
Thank you.
__________________________________________________ ____________________