The identification of the turning point is particularly difficult to automatically detect in non-stationary data however it is essential because without this knowledge accurate trends cannot be determined.
The approach taken to Auto-trending is to utilise a proprietary enhanced Box-Car technique; principally because this processing automatically detects the turning point in the data.
The essence of the technique is shown in the diagram to the right. The Box trend is calculated using the latest data in the data stream whilst the Car trend is calculated on the remaining data. The point where these two trends separate is the turning point. For every new point, the trend rate between the box and the car are calculated and compared using null hypothesis testing to determine if there is a significant difference between them.
If there is a significant difference the alert is set and the trend and the turning point are identified. The turning point is then fixed for further estimation of the trend.
Adaptive filters are used to compensate for data characteristics such as the data rate, variance and correlation. The processing is robust and adapts automatically to changes in this data over time and is not confused by frequency content in the data.
The sensitivity to defects and false alert performance is comparable to the CFAR.