Affordable Access

Optical method and neural network for surface roughness measurement and surface pattern classification

Authors
Publisher
Dublin City University. School of Mechanical and Manufacturing Engineering
Publication Date
Keywords
  • Materials
  • Mechanical Engineering
  • Surface Roughness Measurement
Disciplines
  • Computer Science
  • Design

Abstract

In this present study, two optical methods employing diffusive and specular reflections from the steel surfaces are considered to measure the surface roughness value (Ra). The first method is introduced for Ra > 2 fim while the second method is employed for 0.1 (im < Ra < 2 (im. The peak intensity of the reflected beam, and a Gaussian curve parameter of a Gaussian function, approximating the peak intensity of the reflected beam, are measured for the first and second methods, respectively. Since a unique Ra value exists for a surface, the data collected for each profile were combined to produce a profile representing the Ra value for that particular surface, which is Gaussian in nature. The relationship between Ra and the standard deviation of Gaussian function (SDGF) was developed. An experimental set up associated with both methods has been designed and built. In this case, a He-Ne laser beam was used to scan the workpiece surface while fiber optic probes were employed to collect the reflected beam. To calibrate the fiber optic probes, Ra is measured initially using a Bendix surface proficoder. A back-propagation neural network classifying the surface patterns resulting from the first method was developed. A network simplification based on the self-pruning of the weights was employed. Control chart patterns resembling the possible surface profiles were developed when training the network. It is found that, the resolution of the surface texture measurement improved considerably in the case of presently employed optical method. The neural network developed for this purpose could classify the resulting surface patterns successfully. Newly introduced selfpruning method results in an improvement in the network performance and minimisation of the network structure and computing time. The first scheme used in the second method gives an improved standard estimate of error. The linear relationship was found between the Ra values and SDGF of the reflected beam intensity. Higher the SDGF values result in higher surface roughness. However, the measurement is limited to a certain range of Ra values; in this case, the accuracy of the measurement drops considerably as the Ra value reduces below 0.1 um.

There are no comments yet on this publication. Be the first to share your thoughts.