G-code informed reconstruction of high-resolution CNC machine data from low-sampling rate industrial data streams

Abstract

Industrial Computer Numerical Control (CNC) machines provide access to operational data through standardized communication protocols like OPC-UA and MTConnect; However, these data streams are typically sampled at low and irregular rates, limiting their direct use for spatially resolved process analysis. Sparse reporting, variable sample counts across nominally identical runs, and asynchronous updates obscure the geometric context of machining events and hinder feature-level comparison. This study presents a deterministic reconstruction framework that reparameterizes low-frequency industrial machine data over a geometry-derived spatial reference domain constructed from the executed G-code. Motion blocks are parsed and discretized to form a consistent toolpath representation and reported machine samples are projected onto this domain using a geometry-first, G-code block-aware mapping strategy. Temporal information is regularized through controlled redistribution between mapped indices, while operational attributes are propagated according to controller-reported state persistence. The approach was implemented on two CNC platforms using 1 Hz FANUC FOCAS data and 10 Hz OPC-UA data and validated against 500 Hz controller-native SIEMENS Edge measurements. Reconstruction errors ranged from 0.62 mm to 1.92 mm, which is smaller than the typical spacing between raw industrial samples. A feedrate override case study further demonstrated that productivity deviations can be localized to specific geometric segments using only low-frequency data. The results establish a controller-independent and spatially consistent analysis framework that enables repeatable feature-level comparison across nominal runs and structured diagnostic investigation without reliance on proprietary controller interpolation behavior.