<p>The evolution of transient two-body abrasive wear was investigated using a machine vision system. In the early stages, abrasives with sharp tips produced abrasive wear patterns characterized by isolated elongated grooves. As the abrasive wear progressed, these grooves disintegrated into shorter grooves owing to the overlap of existing and newly formed abrasive wear patterns. This increased the optical image features that correlated with both roughness and mass loss. The relationship between optical image features and mass loss was analyzed from the perspective of abrasive wear theory, and the transition points in the relationship was detected by Kalman filtering. Furthermore, a data-driven framework was developed to predict spatiotemporal variations in surface conditions and the associated optical image features.</p>