The PROGNOST®-Predictor system implements an innovative analysis called Confidence Analysis. Confidence Analysis is a pattern recognition technique that quantifies the similarity of the measured spectral peaks to the expected fault peaks. The confidence provides a strong indicator of whether a measurement really represents a real fault or damage. If the confidence is low, it most likely means that the specific fault is not present, but if the confidence is high, then it is likely that the fault indicated is real. The results of the signature calculation and the confidence are graphed in a confidence plot.

The plot is a data mining tool to identify false alarms and faults which are present but still are in a good status (false pass). The signature amplitude is normalized relative to the threshold values and the amplitude is plotted vertically and the confidence horizontally (figure below). The plot is populated with the most recent amplitude / confidence pairs for each signature in the database. This plot is a convenient method to display whether a significant signature is present and also that the amplitude is larger than normal indicating that a component fault is developing.

Confidence Analysis

The colors of the signature points can be a solid black point or a blue circle, this depends on the element. In case the element is added to the watch list the color is solid black (represents a element which is still in the focus of the user). All blue circles represent points which need to be reviewed corresponding to the following four corners principle.
The four corners of the scatter plot correspond to different possible situations. Points in the bottom left corner are signatures with low amplitudes relative to alarm levels, and a low confidence. These signatures are in a “good” status.

The bottom right corner contains signatures with low amplitudes relative to alarm levels, but a high confidence. These points are false passes and require immediate attention and likely have alarm thresholds that are set too high. These signatures, if left unattended, would be missed alarms in a typical monitoring system since their alarm thresholds are too high to alarm even though a fault signature is present. The user should analyze each of these points and adjust alarm thresholds as necessary to monitor the upcoming failure.

The top left corner contains signatures with high amplitudes relative to alarm levels, but a low confidence. These points are false alarms that need analyst attention and possibly their alarm thresholds increased.

The top right corner contains signatures with high amplitudes relative to alarm levels, and a high confidence. These signatures should be managed carefully to avoid machine failures. They show true component faults.

The scatter plot is divided into alarm regions Warning, Alert, and Alarm. The user can define additional informational regions. For each trend a signature amplitude and confidence is computed. If the new pair of values falls into a defined region that status is assigned to the trend. Hovering over a point in the scatter plot will show a pop-up of information about that point. Double clicking while the pop-up is open moves the selected point into the list to the right. This is a convenient place to save important points for analysis.