Neural network applications in surface topography

Abstract There is an industrial need for the measurement and classification of the topography of engineering surfaces - two-dimensional (2-D) analysis is fast, but limited in the usefulness of the results obtainable whilst the three-dimensional (3-D) approach offers greater scope. Unfortunately, the data analysis step for 3-D data characterisation can be time-consuming, and often requires a skilled metrologist. The approach has therefore been adopted only with reluctance for on-line work. This Paper presents an approach to on-line 3-D surface characterisation/classification (for quality control purposes) based on artificial neural networks (ANN). Inter-surface classification has allowed the authors to place any new surface received by the system into its correct manufacturing process group (honed, turned, shot-blasted etc.). The second part of this work - reported here - has enabled the authors to reliably and consistently place surfaces in different roughness categories (for example, worn and unworn). Such a system, together with the classification module previously developed, would serve as a quality control tool, by interrogating surfaces presented to it and making reliable predictions on the state of the manufacturing process.