Help with removing transient oscillations in a force signal

Hello all,

I am working on a system that needs to accurately measure the
time-varying force applied by an object to a force sensor. However, the
initial impact of the object against the force sensor induces a severe
oscillation in the system that decays over time (similar to a damped
spring-mass system). We have tried a number of different mechanical
alternatives to reduce this oscillation, but have not had much success
(the solutions have all been undesirable for various reasons). One
approach has been to increase the effective spring constant of the
system thus raising the natural frequency of the oscillations, which
are then removed using low-pass filtering. However, this solution is
not really effective long term.

We have also tried a number of DSP techniques to remove the
oscillations, including using a notch filter, but the frequency of the
oscillation is not constant and depends on the mass of the system
(which is not constant). Does anyone know of a digital signal
processing technique that can be used to attenuate the oscillations?
The system samples at 500 Hz and measures a force signal up to 200 Hz.
The oscillations occur at between 60 Hz to 120 Hz and are often of a
similar magnitude as the initial force applied to the sensor (in other
words, the force signal can, at times, oscillate almost back to the
zero-force level).

Your comments and suggestions would be very much appreciated.

Regards,
John

"John Gibson" <johngibson1968@hotmail.com> wrote in message 
news:1110255829.195451.278250@f14g2000cwb.googlegroups.com...
> Hello all,
>
> I am working on a system that needs to accurately measure the
> time-varying force applied by an object to a force sensor. However, the
> initial impact of the object against the force sensor induces a severe
> oscillation in the system that decays over time (similar to a damped
> spring-mass system). We have tried a number of different mechanical
> alternatives to reduce this oscillation, but have not had much success
> (the solutions have all been undesirable for various reasons). One
> approach has been to increase the effective spring constant of the
> system thus raising the natural frequency of the oscillations, which
> are then removed using low-pass filtering. However, this solution is
> not really effective long term.
>
> We have also tried a number of DSP techniques to remove the
> oscillations, including using a notch filter, but the frequency of the
> oscillation is not constant and depends on the mass of the system
> (which is not constant). Does anyone know of a digital signal
> processing technique that can be used to attenuate the oscillations?
> The system samples at 500 Hz and measures a force signal up to 200 Hz.
> The oscillations occur at between 60 Hz to 120 Hz and are often of a
> similar magnitude as the initial force applied to the sensor (in other
> words, the force signal can, at times, oscillate almost back to the
> zero-force level).
>

John,

Your description is a bit thin to provide much help.  So, here are some 
questions that might help:

- Is the sensor part of the "system"?  Or, is the sensor measuring system 
motion or forces without affecting it?

- Do you have a reasonable model for the system in terms of masses, springs 
and damping?  This would be a key element in suggesting something.

- To what extent is removing the observed sinusoid (i.e. "oscillation") 
simply removing the actual system response?

- Is the oscillation in force or in motion?  What does the system model 
mentioned above tell you about this?

- What is the instrumentation block diagram?  To what extent might the 
observation be in the instrumentation and not in the system?

- What are you trying to accomplish?

Fred

John Gibson wrote:
> Hello all,
> 
> I am working on a system that needs to accurately measure the
> time-varying force applied by an object to a force sensor. However, the
> initial impact of the object against the force sensor induces a severe
> oscillation in the system that decays over time (similar to a damped
> spring-mass system). We have tried a number of different mechanical
> alternatives to reduce this oscillation, but have not had much success
> (the solutions have all been undesirable for various reasons). One
> approach has been to increase the effective spring constant of the
> system thus raising the natural frequency of the oscillations, which
> are then removed using low-pass filtering. However, this solution is
> not really effective long term.
> 
> We have also tried a number of DSP techniques to remove the
> oscillations, including using a notch filter, but the frequency of the
> oscillation is not constant and depends on the mass of the system
> (which is not constant). Does anyone know of a digital signal
> processing technique that can be used to attenuate the oscillations?
> The system samples at 500 Hz and measures a force signal up to 200 Hz.
> The oscillations occur at between 60 Hz to 120 Hz and are often of a
> similar magnitude as the initial force applied to the sensor (in other
> words, the force signal can, at times, oscillate almost back to the
> zero-force level).
> 
> Your comments and suggestions would be very much appreciated.

The oscillation is part of the system response. Most devices that sense 
a "force" actually measure a displacement that the force induces in a 
stiff member. The compliance of that member and the mass of the system 
-- load and support -- set a resonant frequency. Damping is rarely an 
initial design criterion.

The oscillation that you see is real. The force that you read during it 
is real. In general, impacts cause higher forces than resting loads.

Think of an electrical analogy: an inductor, a resistor, two capacitors, 
and a switch, connected in series. With the switch open, one of the 
capacitors is charged. The signal of interest is the voltage on the 
originally charges capacitor. Asking to see that signal without 
oscillation seems rather beside the point, does it not?

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Jerry Avins wrote:

> John Gibson wrote:
> 
>> Hello all,
>>
>> I am working on a system that needs to accurately measure the
>> time-varying force applied by an object to a force sensor. However, the
>> initial impact of the object against the force sensor induces a severe
>> oscillation in the system that decays over time (similar to a damped
>> spring-mass system). We have tried a number of different mechanical
>> alternatives to reduce this oscillation, but have not had much success
>> (the solutions have all been undesirable for various reasons). One
>> approach has been to increase the effective spring constant of the
>> system thus raising the natural frequency of the oscillations, which
>> are then removed using low-pass filtering. However, this solution is
>> not really effective long term.
>>
>> [SNIP]
>>
>> Your comments and suggestions would be very much appreciated.
> 
> 
> The oscillation is part of the system response. Most devices that sense 
> a "force" actually measure a displacement that the force induces in a 
> stiff member. The compliance of that member and the mass of the system 
> -- load and support -- set a resonant frequency. Damping is rarely an 
> initial design criterion.
> 
> The oscillation that you see is real. The force that you read during it 
> is real. In general, impacts cause higher forces than resting loads.
> 
> [snip electrical analogy]

Mr. Gibson, are you using a bridge-style strain gage transducer?
Have you considered a piezoelectric force transducer?

I would suspect that the inherent stiffness of a piezoelectric element 
as compared to a strain gage might force the resonant frequency high 
enough that minimal amount of damping would alleviate the problem.

CAVEAT: above based on fond memories of being lab tech ~40 years ago.

John Gibson wrote:
> Hello all,
>
> I am working on a system that needs to accurately measure the
> time-varying force applied by an object to a force sensor. However,
the
> initial impact of the object against the force sensor induces a
severe
> oscillation in the system that decays over time (similar to a damped
> spring-mass system).


Hello John,

Is your system relatively linear? If so, then your measured signal is
the convolution of the applied signal and the system's impulse
response. Measure the impulse response and then deconvolve your
measured signal with the system's impulse response.

Clay

Richard Owlett wrote:
> Jerry Avins wrote:
> 
>> John Gibson wrote:
>>
>>> Hello all,
>>>
>>> I am working on a system that needs to accurately measure the
>>> time-varying force applied by an object to a force sensor. However, the
>>> initial impact of the object against the force sensor induces a severe
>>> oscillation in the system that decays over time (similar to a damped
>>> spring-mass system). We have tried a number of different mechanical
>>> alternatives to reduce this oscillation, but have not had much success
>>> (the solutions have all been undesirable for various reasons). One
>>> approach has been to increase the effective spring constant of the
>>> system thus raising the natural frequency of the oscillations, which
>>> are then removed using low-pass filtering. However, this solution is
>>> not really effective long term.
>>>
>>> [SNIP]
>>>
>>> Your comments and suggestions would be very much appreciated.
>>
>>
>>
>> The oscillation is part of the system response. Most devices that 
>> sense a "force" actually measure a displacement that the force induces 
>> in a stiff member. The compliance of that member and the mass of the 
>> system -- load and support -- set a resonant frequency. Damping is 
>> rarely an initial design criterion.
>>
>> The oscillation that you see is real. The force that you read during 
>> it is real. In general, impacts cause higher forces than resting loads.
>>
>> [snip electrical analogy]
> 
> 
> Mr. Gibson, are you using a bridge-style strain gage transducer?
> Have you considered a piezoelectric force transducer?
> 
> I would suspect that the inherent stiffness of a piezoelectric element 
> as compared to a strain gage might force the resonant frequency high 
> enough that minimal amount of damping would alleviate the problem.
> 
> CAVEAT: above based on fond memories of being lab tech ~40 years ago.

*Something* has to absorb and then dissipate the energy of impact. I can 
show you a force measurement system that has no more * give than a 
bridge abutment and enough sensitivity to measure a pound or two. If you 
jump from a foot stool onto a scale at floor level and land on your 
heels with your knees locked, be glad the scale will oscillate. Expect 
the scale to read more than your weight at first. The stiffer you make 
the platform, the higher the scale will read.

Jerry
____________________________
* = non-transient
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Clay wrote:
> John Gibson wrote:
> 
>>Hello all,
>>
>>I am working on a system that needs to accurately measure the
>>time-varying force applied by an object to a force sensor. However,
> 
> the
> 
>>initial impact of the object against the force sensor induces a
> 
> severe
> 
>>oscillation in the system that decays over time (similar to a damped
>>spring-mass system).
> 
> 
> 
> Hello John,
> 
> Is your system relatively linear? If so, then your measured signal is
> the convolution of the applied signal and the system's impulse
> response. Measure the impulse response and then deconvolve your
> measured signal with the system's impulse response.
> 
> Clay

Aren't the measured and actual forces on the sensing element the same? 
Assuming that the effective mass of the linkage is small relative to the 
load, deconvolving will yield a (perhaps pretty) fiction. Unless the 
effective mass of the linkage is small relative to the load, the 
oscillatory response depends heavily on the load.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Thank you for your considered responses. To further elaborate:

- The force sensor (a semiconductor piezo-resistive sensor) is part of
the system, and the oscillations are the result of the entire system
vibrating in response to the initial impact.

- We are unable to accurately model the system as a spring/mass/damping
system since we cannot know the effective mass of the system due to
external influence.

Reading your comments has made me revise my perspective on the problem.
The way I see it is that there are two discrete signals being measured;
a short-term transient signal (the sharp step resulting from the
initial impact) and a long-term steady state signal (the force applied
by the object as it is in contact with the system and the signal that
we are really interested in). Whist I agree that the oscillations
resulting from the initial impact are a real part of the signal, from a
system processing point of view they are noise and we would like to
remove them if possible. The ideal system would produce the same force
readings irrespective of if the object coupled with the system with a
hard impact (large oscillations) or a gentle impact (minimal
oscillations).

Thus, the problem is really "how can we remove the effects of the
initial transient signal that is not of interest to the system?" Clay
suggested measuring the impulse response and de-convolving, which is
something we considered. However, we are again hampered by the fact
that the impulse response is dependent on the mass of the system, which
is not constant.

What I was hoping for was some kind of adaptive algorithm that could
detect the magnitude and frequency of a damped oscillation (the same
way an echo canceller might detect the lag and magnitude of an echo)
thus allowing removal. Alternatively, perhaps there is a signal
processing technique that allows the two signals to be separated and
processed separately. Perhaps I am being overly optimistic in thinking
this, but the problem seems like it would be very common in systems
that use sensors (force or otherwise) and thus there might be a
well-known solution.

Best regards,
John

Jerry Avins wrote:

> Richard Owlett wrote:
> 
>> Jerry Avins wrote:
>>
>>> John Gibson wrote:
>>>
>>>> Hello all,
>>>>
>>>> I am working on a system that needs to accurately measure the
>>>> time-varying force applied by an object to a force sensor. However, the
>>>> initial impact of the object against the force sensor induces a severe
>>>> oscillation in the system that decays over time (similar to a damped
>>>> spring-mass system). We have tried a number of different mechanical
>>>> alternatives to reduce this oscillation, but have not had much success
>>>> (the solutions have all been undesirable for various reasons). One
>>>> approach has been to increase the effective spring constant of the
>>>> system thus raising the natural frequency of the oscillations, which
>>>> are then removed using low-pass filtering. However, this solution is
>>>> not really effective long term.
>>>>
>>>> [SNIP]
>>>>
>>>> Your comments and suggestions would be very much appreciated.
>>>
>>>
>>>
>>>
>>> The oscillation is part of the system response. Most devices that 
>>> sense a "force" actually measure a displacement that the force 
>>> induces in a stiff member. The compliance of that member and the mass 
>>> of the system -- load and support -- set a resonant frequency. 
>>> Damping is rarely an initial design criterion.
>>>
>>> The oscillation that you see is real. The force that you read during 
>>> it is real. In general, impacts cause higher forces than resting loads.
>>>
>>> [snip electrical analogy]
>>
>>
>>
>> Mr. Gibson, are you using a bridge-style strain gage transducer?
>> Have you considered a piezoelectric force transducer?
>>
>> I would suspect that the inherent stiffness of a piezoelectric element 
>> as compared to a strain gage might force the resonant frequency high 
>> enough that minimal amount of damping would alleviate the problem.
>>
>> CAVEAT: above based on fond memories of being lab tech ~40 years ago.
> 
> 
> *Something* has to absorb and then dissipate the energy of impact.

Yepp ;{!  Woke up [ *literally* ] to "need damping" about 2 hrs after 
posting. I work graveyard shift and should only post when awake ;)


> I can 
> show you a force measurement system that has no more * give than a 
> bridge abutment and enough sensitivity to measure a pound or two. If you 
> jump from a foot stool onto a scale at floor level and land on your 
> heels with your knees locked, be glad the scale will oscillate. Expect 
> the scale to read more than your weight at first. The stiffer you make 
> the platform, the higher the scale will read.
> 
> Jerry
> ____________________________
> * = non-transient

John Gibson wrote:
> Thank you for your considered responses. To further elaborate:
> 
> - The force sensor (a semiconductor piezo-resistive sensor) is part of
> the system, and the oscillations are the result of the entire system
> vibrating in response to the initial impact.
> 
> - We are unable to accurately model the system as a spring/mass/damping
> system since we cannot know the effective mass of the system due to
> external influence.
> 
> Reading your comments has made me revise my perspective on the problem.
> The way I see it is that there are two discrete signals being measured;
> a short-term transient signal (the sharp step resulting from the
> initial impact) and a long-term steady state signal (the force applied
> by the object as it is in contact with the system and the signal that
> we are really interested in). Whist I agree that the oscillations
> resulting from the initial impact are a real part of the signal, from a
> system processing point of view they are noise and we would like to
> remove them if possible. The ideal system would produce the same force
> readings irrespective of if the object coupled with the system with a
> hard impact (large oscillations) or a gentle impact (minimal
> oscillations).
> 
> Thus, the problem is really "how can we remove the effects of the
> initial transient signal that is not of interest to the system?" Clay
> suggested measuring the impulse response and de-convolving, which is
> something we considered. However, we are again hampered by the fact
> that the impulse response is dependent on the mass of the system, which
> is not constant.
> 
> What I was hoping for was some kind of adaptive algorithm that could
> detect the magnitude and frequency of a damped oscillation (the same
> way an echo canceller might detect the lag and magnitude of an echo)
> thus allowing removal. Alternatively, perhaps there is a signal
> processing technique that allows the two signals to be separated and
> processed separately. Perhaps I am being overly optimistic in thinking
> this, but the problem seems like it would be very common in systems
> that use sensors (force or otherwise) and thus there might be a
> well-known solution.
> 
> Best regards,
> John

John,

Imagine that the system compliance is made insignificantly small. (That 
can be done.) How would impact then affect your measurements? The 
initial oscillations you now read are true measurements of the 
oscillating force on your sample as it and the weighing platform move.

The mass of your sample and its velocity at impact tell you the amount 
of energy that something has to absorb before the transient is 
dissipated. The duration of the dissipation interval will provide a 
useful estimate of the power required to do that with an active absorber.

As it is, you want to avoid measuring what is there and instead display 
what would have been there if only .... That's hard.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������