Investigate Your Data

Note the data is loaded. The number of rows varies between the different data sources (17787 and 65533). You can review the measurement/point names that describe our two turbines (Unit 1 and Unit 2). Statistics are presented for each point. For example Unit 2 Stack NOx has 10273 valid rows (out of 17787), a minimum below 0, a maximum of 30.98, a mean of 3.16 and a standard deviation of 3.44 (which is 108.7% of the mean).

Select

the box outline icon ( Select All). Then right mouse-click within the greyed statistics and

select

Plot and Trend Plot.

All data is presented in row plots, which are not aligned across the different datasets.

Change

the plot presentation option from By Row to By Time using upper right button.

Now data from our data files are aligned in the plots by time (but no longer by rows). If you

run your cursor

across the data near any visible point you can note that Unit 2 has results (running?) from September 18 through 21 and from September 27 through 28.

Scroll down

on this analyzer data to locate Unit 1 data using the scrollbar on the right.

Note that the Unit 1 turbine seems to be running from 9/21 through 27 (running your cursor over trend plots provides a pop up of time and value).

Drag

your mouse (clicked) across a section of the displayed Unit 1 data (this will zoom over that range).

Zoom is flexible by dragging or using selected ranges (row or time as presented) with the find (flashlight) icon. In the upper left additional zoom icons to zoom in (step-wise) or zoom out (where present to your previous zoom level) or zoom back to display all data. Next to these zoom tools are additional plotting options,

select

(turn off) the line button.

Note that points are interpolated and Unit 1 is not running the entire interim time. Gaps are filled in with lines between populated times.

Experiment

with Zoom and Lines/points. You may notice that average points are not plotted where there are insufficient pixels to display all, but preferentially display extremes (to support identifying events or relationships). Also when in ‘line’ rather than point mode – if zoom is sufficient to display points separately they are displayed with lines automatically.

Scroll

back to our first points,

zoom

[] fully out and

turn

[] line display back on.

Maximize the windowand change number of plots displayed to 6 (from 4). While there are apparent spikes in data for now we’ll leave most of these as is.

Select

both Inlet O2 and Stack O2 (with mouse-click and ctrl-mouse-click).

Right mouse-click

,

select

Cut and Cut Below Options. Ctrl-click selects individual plots and Shift-click selects a range of plots.

Drag

your cut Value below the visibly apparent real data on both plots and “

Apply”

your cut.

The two O2 measurements are correlated and we can see (put cursor over any trend values to understand colors) that Stack O2 is sometimes higher than Inlet O2 (duct burner damper?). A zero excess O2 is bad data. A high O2 (up to 21.6% here) is more likely ambient, i.e. without combustion. You may notice controls in the upper left to change from cut (remove data) to clip (set data to limit), from Below (to Above or Both high and low) and to set a specific value (numerically, not graphically)

Data is cleaned.

Scroll

down to check the O2 data on Unit 1. No cuts are required. Continue to check the rest of the data (

Scroll down).

Observe your data and look for more data/time or other issues.

There are problems with the plant weather station data. Apply cuts (

multi-select, right mouse-click,

Cut, Cut Below Value) on the problematic Outside Air Humidity and Temperature. Remember multi-select is Ctrl-mouse-click or you may select a range with mouse-click, shift-mouse-click.

Cut

the data (

drag

cut line) below the least value, but above 0 (or

set

the cut below value to 5).

Select

Apply.

This

change did not eliminate all our problem data and it’s possible that the dry-bulb or web-bulb temperature gets below 5 Deg F (it’s only September).

Select

the Dataset, Transforms option from the top icon.

Select

our TurbineEmissionsAnalyzer dataset (you may work with more than one dataset at the same time).

Select

(

Ctrl-Click

) the three ‘Outside Air” condition transforms where values were Cut below 5. This is our transform list of how we have cleaned/processed the raw data. At the end of an exercise if/when you export part of the dataset for developing a Machine Learning application you will want to execute these same transforms prior to scoring your ML algorithm. In addition you may note that in Interactive Data Exploration data is not eliminated so almost anything is reversible.

Right mouse-click

and

Delete

these transforms. Then

select

Commit and close.

Multi-Select

the three Outside Air plots again (if needed) and with a

right-mouse click

options

Cut

Rows or Time.

As needed, you will

Change

the upper left plot option from “Drag Cut” to “Drag for Zoom”.

Zoom

in (by dragging) around the first data drop. You may do this multiple times and if needed reverse one step with the Zoom Out button.

Change

context to “Drag Cut” and

Drag

across the rows that should be eliminated. This can be quicker and simpler if you include one or two extra ‘good’ values on each side. Zoom out and

repeat

(Zoom & Drag Cut)

for the second data drop later in time on these three measurements.

Select

Apply with two cuts selected leveraging the zoom tools.

Scroll

down

through your data further to look for more issues or information.

You may notice that other than the analyzer file (emissions) the

turbine historical data only goes through 9/24. This is important because we now want to align data on one integrated time-index. This is needed for most Machine Learning tools and for some of our Correlation analytics.

Select

the dataset overview tab (TurbineEmissionsAnalyzer) and

select

the Time Merge button at the top (also available under the Dataset Icon below the menu bar).