MAPS

A mechanism-aligned pseudo-anomaly screening strategy that constructs hard negative samples by simulating real fault generation mechanisms, such as broken dependencies between sensors, making the training supervision of unsupervised anomaly detection closer to the true anomaly distribution.