Monitoring the propagation of mechanical discontinuity using data-driven causal discovery and supervised learning

•Data-driven workflow successfully monitored the crack location, orientation, and length.•Monitoring of three stages of crack propagation has a median RMSE lower than 2.5 mm.•Numbers of zero-crossing is the strongest causal signatures of crack propagation.•Not all highly correlated features exhibit high causality to crack propagation.•Several high importance features for monitoring task exhibit low causality.