Investigation of surface roughness characterization using an ART2 neural network

Document Type

Conference Proceeding

Date of Original Version



Surface roughness is an important quality of surfaces. A shortcoming of currently used techniques for characterizing surface roughness is the absence of a single descriptive number that can be used to differentiate the roughness of different surfaces. This paper discusses the development of an Adaptive Resonance Theory (ART2) neural network-based approach for surface roughness analysis. The objective is to develop a unique, more descriptive parameter of surface roughness. ART2 neural networks offer an attractive approach to classify sensory data such as surface roughness profiles, since they have the capability of classifying the data in a self-scaling, unsupervised fashion, and of deciphering the data's invariant properties. In this paper, an ART2 network structure is enhanced to enable the network to address the need of attributing a single numerical value to each classified category. The above methodology was tested on six simulated surfaces and the results obtained showed that the ART2 network is capable of classifying each surface into a distinct category and assigning a roughness index to each category that is more descriptive than the commonly used CLA parameter. However, further work is still needed to obtain the ultimate single number index.

Publication Title, e.g., Journal

American Society of Mechanical Engineers, Production Engineering Division (Publication) PED



This document is currently not available here.