12.03.2016 Views

Anomaly Detection for Monitoring

anomaly-detection-monitoring

anomaly-detection-monitoring

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

ack again, meaning that they smooth data using both past and<br />

future data. Smoothing bidirectionally can eliminate the effects of<br />

large spikes.<br />

Figure 3-5. A window function control chart. This time, the window is<br />

<strong>for</strong>med with values on both sides of the current value. As a result,<br />

anomalous spikes won’t generate abrupt shifts in control limits even<br />

when they first enter the window.<br />

The downside to window functions is that they require a larger time<br />

delay, which is a result of not knowing the smoothed value until<br />

enough future values have been observed. This is because when you<br />

center a bidirectional windowing function on “now,” it extends into<br />

the future. In practice, EWMAs are a good enough compromise <strong>for</strong><br />

situations where you can’t measure or wait <strong>for</strong> future values.<br />

Control charts based on bidirectional smoothing have the following<br />

characteristics:<br />

• They will introduce time lag into calculations. If you smooth<br />

symmetrically over 60 second-windows, you won’t know the<br />

smoothed value of “now” until 30 seconds—half the window—<br />

has passed.<br />

• Like sliding windows, they require more memory and CPU to<br />

compute.<br />

Statistical Process Control | 23

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!