Blind separation of manufacturing variability with independent component analysis: A convolutive approach

In this paper, we apply the blind source separation model to the scope of extracting information from a workpiece about the process that made it. Given any manufactured workpiece, we may think about it as the carrier of the information built in the process that made it. Using recent inspection technologies such as stylus profiler, we are able to generate signals from a workpiece. We analyze these signals using independent component analysis (ICA) in its various formulations. In doing this, we develop a convolutive version of ICA to overcome technical and metrological problems arisen. By using this convolutive modification of ICA we are able to demix the recorded signal and to recover the technological fingerprint over it. Simulations on NIST benchmarks are included, as well as a case study on a turned workpiece.