LabVIEW
identify the disturbances in a signal
Hello Lynn,
The subtracted Adjusted baseline logic looks good for most part of the signal. Thank you for the same.
But it shows some unwanted corrections in the waveform like half profile as marked in the attached image.
Also the logic behind the choosing specific values is not clear to me. Please elaborate.
- Rahul
Rahul,
The 50 is easy. I guessed at 25. When I looked at the data, it did not look entirely satisfactory so I increased the. 50 seemed like it worke well. The idea is that with a -0.5 threshold, the index where the signal crosses the threshold is not where the signal starts falling or stops rising. Adusting the indexes by an arbitrary constant seems close enough and is much less work than trying to decide programmatically where the spike deviates from the baseline.
The 2147483647 is the largest value which can be represented by I32. So that code simply forces the index to the Replace Array Subset to be >= 0.
Now the hard part. The half profile issue. I saw that after I had posted. This can occur when either a valid disturbance or some noise occurs on the signal so that the threshold is crossed twice in a short time. I tried varying the threshold and could not find a value which did not produce at least one such truncated spike. The first image below is an expanded view of the original signal showing two downward corssings fo the -0.5 thershold between 0.52445 and 0.52450. The second image shows that the spike at that time was cut off. That noise is almost invisible at a scale which shows the entire spike.
I do not have a solution for that issue. If you have a small amount of data, you could just check manually and either adjust the threshold to get the spike or discard. For large amounts of data or if a manual approach is not acceptable, makeing two or more passes through the data with different thresholds might work. Each threshold may miss some spikes but they will likely be different spikes. Another approach would be to not accept a second crossing within ~ 1ms of the first. Actually you would need to require that the signal stayed low for most of that ms. Just blocking for a millisecond could have problems with a noise impulse which crossed the threshold - and there are a few in the data you posted.
Automated processing of complicated, real signals is often not straightforward. When you think you have all the possibilities covered, a new twist shows up. Only you, or the person wanting the data, knows what is important in these signals. Your normal spikes seem to have two negative peaks with the later one larger than the first. The disturbed spikes seem to have a portion of the later peak inverted. But what parameters are important? The peak value of the disturbed spike? The magnitude of the inverted portion? The time the inversion starts or stops? If that timing is important, what defines the reference time for that spike? Or many other possible questions I have not thought to ask?
Even with the baseline correction I proposed, it appears that you still have variation of the peak amplitude between 10 and 15% on the normal spikes. While there are small differences compared to the original, this variation appears to be significantly larger than the baseline variations.
Lynn
Hello Lynn,
Yes, there is a variation of about 10 to 15% in the magnitudes of all peaks. I am thinking of defining a bandlimit for lower and upper value of the signal for each small time phase. Right now, each negative spike has about 350 such time intervals. With defining the acceptable limits for the negative spike values at a defined time interval, I think it will be easier to separate out the disturbances.
For this I am thinking of using the Triggered Signal Array Display and band limit. And comparing the values, to find out the disturbances. Once I am sure about the proper profile of the signal wave, it will be possible to difine the band limits for specific time phase.
Rightnow, I am just concerend about the identification of the disturbaces in the signal for identifying the presence of chip and hence its length.
Its magnitude is not important as far as I think so. At the moment, I am working with a defined speed, so the negative spike profile is visible at the defined time interval depending upon the rotational speed of the tool. This time I am trying to find out the disturbances with only one pair of light source and photo-detector. Once I am sure about the process, in coming days, I am going to implement 3 pairs of light source and photo-detectors covering complete perimeter of the tool. And also running the tool with different speeds is also a part of next experiments depending upon the tool progress in to the workpiece.
Considering all these things, its seems to be a complicated task but will be surely going work-out with proper planning.
I hope you have now got an idea about what I am looking for from this analysis.
- Rahul
Lynn,
I tried to apply Lowpass and Smoothing filters in the start of signal analysis and it provided good results without half profile pattern. Now, the results as per the VI provided by you are looking good. 
I am still thiking of any possibility to stretch the wave so as to balance the 50Hz wave profile or may be any possibility to get over this unwanted noise. I think if anyhow if I am able to remove this noise, it's effect on the wave pattern as hown in the image below will be reduced and all wave patterns will be in a narrow band and that will be good to define precise bandlimit to identify the correct / standard wave profile.
- Rahul
