Spike Peak finding from spectral data

Question

Hi everybody,

I need some help: is there a way to find the min and max value from spikes and to do a delta calculations of those 2 values. i have attached the dataset. In a nutchel, the customer would like to achieve the following:

we need to answer this equation delta y/delta x where y is the temperature and x the time.

And the delta is than difference when a batch is started and when the same batch is finished.

The file attached is the raw data. When you plot it you will see a lot of peaks. And from the valley to the peak, that is one batch. So we would like to get out the batches Starting temp(aka minimum temp) and finishing temp (aka maximum temp) and at what speed this difference occurred.

So the math is very easy. We just need help to find these points in this raw data that we send to you. And to get the analysis automated.

2018-05-20.jmp

Valerie_Nedbal · Answer

Thanks Martin, that helps!! and yes, i agree if we could implement a sort of formula or so in the next version.

Chris_Kirchberg · Answer

I have to agree wiht Martin in regards to the noise of the data. There are 4 patterns/trends that I see: 1) A general drift upwards over time. 2) a series of quick oscillations followed by a long declinding slope 3) Patter of item 2 repeating itself. 4) A drift upwards withing the quick oscillations over time.

So I have a few questions about the data.  Where does the real batch begin and end? There are a lot of peaks and valleys, so is every single one a batch? Maybe only 4 batches given the slowly declining slope after each series of quick oscillations? Is every little oscilation a batch and the slowly declining slope and flush/reset of the system? This looks like measurement of temp over time, is that what it is? That would make sense given the drift upward over time and within each series of quick oscillations, then a slow drift down afterwards.

More information about the process would help. Otherwise all one would have is what Martin provided. Moving average and find min and max using sign change of slopes.

martindemel · Answer

Hi Valerie,

hopefully not the best approach but at least something that should work. Attached you'll find a table with the peak finding for Batch start and Batch end. More I couldn't do because of other work.

Basically you need to calculate the deltaY and deltaX and divide them (to get the slope). That's easy. The data is quite noisy so you definitly want to smooth it, e.g. with a moving average. Then you need to look for sign changes of the slope and you have to take care of the observations where no change in the slope occurs. For Batch start and batch end you need to set lower/upper thresholds too to eliminate the slope change peaks for each. This results in the peaks to identify the batch start and batch end.

Based on the data there are still cases where two consecutive peaks belong to the same Batch type (Start or End), for those you need to set the first ones of these two or more consecutive peaks to 0 and only the latest to 1. If you look at the visualization you'll see what I mean. This should work using a subset where there is Batch Start not equal Batch End. THen search for the consecutive entries either in Batch Start or Batch End and set the  values to 0. May this can be done also with Brady's Data Table tools Add-in.

As soon this has been done you could calculate the differences from row to row (of course only for the ones which belong to each other :)  )

Hope this helps and I'd like to see some kind of this in JMP in future releases, if there isn't yet a more gentle solution ;)

In Pharmaceutical/chemical/Sensory and in batch processing I hear this question where does my batch start and what is the difference more and  more often.

FindingBatchPeaks.jmp