Anomaly Detection

AVEVA Vision AI Assistant uses an Unsupervised Machine Algorithm to learn what normal is, and then applies a statistical test to determine if a specific data point is an anomaly. This skill detects any type of occurrence, including ones which have never been seen before. Here you provide AVEVA Vision AI Assistant a collection of images of the expected outcome (or good images) and the solution determines an anomaly when the real-time images do not match the training images.

This detection model is useful when there are two states, where the second state is anomalous and the data representing a negative state is not always available or is a super-set of too many sub-states. In these cases, identifying the good case is more intuitive rather than labelling all possible anomalous states.

For example, the grid pattern of a security gate. There could be any number of manufacturing defects observed during the quality check. In this case, Anomaly Detection is best at determining all not good states.