Is neccesary to add an hysteresisi filter if the signal will be filtered with the peak valley method? Or there are signals that might need both procceses as the complexity is higher? Can other filters be added instead of the hysteresis filttering?
The hysteresis filter is used to remove small cycles. We do it always with at least the size of 1 bin width. So, all cycles that are smaller than 1 bin width are ignored. The peak valley filtering mentioned in the video will remove monotonic parts between turning points. Without the hysteresis filter there would be much more turning points and you would end up with cycles on the main diagonal. These would be cycles with a reference amplitude of zero (after the binning). So, it does not make much sense in keeping them with respect to damage. Other filters can be applied in time domain before doing the Rainflow counting, but they are not embedded in the Rainflow counting algorithm as we have implemented it. But be careful with applying filters that modify the amplitude: this will modify the damage!
thanks for this good explanation how the algorithm is working. One question raised up. I do not understand why the point 3 and 4 of the cycle in minute 11:18 were deleted. I would expect to delete the point 2 and 3. Do you have an explanation for me? thanks in advance
You have a very good eye! You are absolutely right, the removed cycle is not shown correctly. Your expectation is correct. Point 2 and 3 should be removed, not 3 and 4. We have an update of that presentation available, but we missed to update it on TH-cam. Sorry for that. The residue does not change, since instead of the value 1 somewhere in the middle the value 1 more to the end is used.
A half cycle or incomplete cycle is treated differently. It has not fully closed, so the true amplitude and mean of the cycle is not known. In some damage calculation, the cycle will be artificially closed, in other damage calculations it is not closed until it is fully understood. So they main point is that these incomplete cycles should be accounted for differently (ie, the residue).
Example: If a rainflow matrix was repeated 10 times (maybe to go from one hour of damage to 10 hours), the residue can be appended 10 time over and counted, and many of the incomplete cycles become complete.
The half cycles are stored in the residue and can later be used in different ways when calculating damage. You can decide to ignore them or forcefully close them. But by storing the half cycles separate from the Rainflow matrix, you can later decide how to deal with them. If you immediately add it to the matrix or not add it to the matrix, you don't have the option later to evaluate the difference.
In our software, the minimum gate is 1 bin width. This will avoid cycles on the main diagonal which would have a reference physical range of 0. This would not add to the damage anyway. You can use larger gates, then this value should relate to the endurance limit of the material you are measuring/testing. So, the gate should only exclude cycles where you are super sure that they will not add to your damage.
Only the inner points build a closed hysteresis loop in the stress/strain path. To close the cycle built by the inner points, you need the outer points. Later, the outer points might become inner points of some larger points. Then they are counted as well.
Excellent video. The Rainflow method is clearly explained.
Thanks, this clearly explained the whole process of Rainflow.
Thanks for such a clean video
Excellent! Thank you very much!
This is such an amazing explanation
Excellent
Is neccesary to add an hysteresisi filter if the signal will be filtered with the peak valley method? Or there are signals that might need both procceses as the complexity is higher?
Can other filters be added instead of the hysteresis filttering?
The hysteresis filter is used to remove small cycles. We do it always with at least the size of 1 bin width. So, all cycles that are smaller than 1 bin width are ignored.
The peak valley filtering mentioned in the video will remove monotonic parts between turning points. Without the hysteresis filter there would be much more turning points and you would end up with cycles on the main diagonal. These would be cycles with a reference amplitude of zero (after the binning). So, it does not make much sense in keeping them with respect to damage.
Other filters can be applied in time domain before doing the Rainflow counting, but they are not embedded in the Rainflow counting algorithm as we have implemented it.
But be careful with applying filters that modify the amplitude: this will modify the damage!
thanks for this good explanation how the algorithm is working. One question raised up. I do not understand why the point 3 and 4 of the cycle in minute 11:18 were deleted. I would expect to delete the point 2 and 3. Do you have an explanation for me? thanks in advance
You have a very good eye! You are absolutely right, the removed cycle is not shown correctly. Your expectation is correct. Point 2 and 3 should be removed, not 3 and 4.
We have an update of that presentation available, but we missed to update it on TH-cam. Sorry for that.
The residue does not change, since instead of the value 1 somewhere in the middle the value 1 more to the end is used.
@@HolgerZeinert thx for the quick answer. Now my world is back in order again :-)
Has this been published as research?
Not sure the full extent of the question, but rainflow counting is a standardized procedure which is documented in ASTM 1049 standard since 1986.
Is there a sample time series I can use to test my code with?
So If I understand it correctly this method only counts the "full cycles". The half cycles are left in the residue ?
A half cycle or incomplete cycle is treated differently. It has not fully closed, so the true amplitude and mean of the cycle is not known. In some damage calculation, the cycle will be artificially closed, in other damage calculations it is not closed until it is fully understood. So they main point is that these incomplete cycles should be accounted for differently (ie, the residue).
Example: If a rainflow matrix was repeated 10 times (maybe to go from one hour of damage to 10 hours), the residue can be appended 10 time over and counted, and many of the incomplete cycles become complete.
The half cycles are stored in the residue and can later be used in different ways when calculating damage. You can decide to ignore them or forcefully close them. But by storing the half cycles separate from the Rainflow matrix, you can later decide how to deal with them.
If you immediately add it to the matrix or not add it to the matrix, you don't have the option later to evaluate the difference.
how did you decided the length of the hystereis gate?
In our software, the minimum gate is 1 bin width. This will avoid cycles on the main diagonal which would have a reference physical range of 0. This would not add to the damage anyway.
You can use larger gates, then this value should relate to the endurance limit of the material you are measuring/testing. So, the gate should only exclude cycles where you are super sure that they will not add to your damage.
7:08 why?only inner counted?
Only the inner points build a closed hysteresis loop in the stress/strain path. To close the cycle built by the inner points, you need the outer points.
Later, the outer points might become inner points of some larger points. Then they are counted as well.
11:30