Reverse engineering of part geometries and machining process parameters from CNC machining time series measurements

Abstract Manufacturing companies increasingly share computerized numerical control (CNC) machining data via industrial data ecosystems and vendor cloud platforms. While these approaches promise improved productivity, they also raise confidentiality risks: seemingly low-risk time series measurements can leak confidential information about part geometries and process parameters if obtained by malicious actors. In this work, it was investigated to what extent confidential geometric and process information can be reconstructed from CNC time series measurements acquired during 2.5-axis peripheral and face milling jobs, and how privacy-preserving technologies can be used to mitigate this reconstruction risk. A five-step methodology was applied: (1) acquisition of CNC data, (2) protection of the axis time series through privacy-preserving technologies, (3) evaluation of the remaining data utility, (4) reversal of the protection using a filter-based approach, and (5) reconstruction of the confidential information using machining-domain knowledge. The approach was evaluated on time series measurements of two reference parts. The results show that a suppression of only the axis position data is insufficient and that a reduction in sampling rate combined with a moderate noise addition hinders the reconstruction at the cost of a substantial loss in data utility. At the original sampling rate, the feed per tooth and certain geometric features could still be reconstructed even under high noise addition. The necessary degree of protection and the remaining data utility varied between the two reference parts, indicating that the reconstruction risk needs to be evaluated on a per-part basis.