Abstract Neural networks have been successful at pattern recognition and discovery of hidden relationships amongst parameters and as such are likely supplements to the sensory systems employed in industrial applications. This paper examines four resulting issues imposed upon any industrial inspection system using a neural network: the feature set which the sensory system must provide, the accuracy of neural-network-based inspection, the robustness required of the sensory system for accurate inspection, and the computational burden imposed by accuracy requirements. This is accomplished in the context of web-process inspection, which requires rapid examination of vast amounts of data for on-line detection of faults in the sheet material. Each of the four crucial issues is addressed: 1. (i) Feature vectors with nine or 17 dimensions, created by a simulated segmented photodetector using measurement of the angular distribution over a 25° cone angle of the scattering were evaluated for inspection CrO 2 -coated sheet steel samples. The scattered coherent light from the surface of the material being processed could be directly conditioned by a photodetector so as to produce this small set of features which are then examined by a neural network trained to find and categorize unsatisfactory surface conditions. Details are presented to show how a modified feature set was developed and tested after an examination of feature space. This new, smaller set proved to be more accurate than the larger set. 2. (ii) Classification by fault or no fault categorized 133 samples correctly out of 135, while there were seven errors in one attempt at classification into the various common surface faults out of the same number of test samples and nine in another. It is shown that a bit of insight in feature selection can improve the capability of the network to recognize faults. 3. (iii) The robustness issue is important since the inspection system must function in the industrial environment, where maintaining an exact alignment of the optics is not feasible. Tests are described where it is shown that fault classification using the proposed system is reasonably robust to slight variation of the angle between the laser beam and camera. 4. (iv) The computational issue is discussed in the context of the data handling requirements of the inspection system.
[1]
Alex Pentland,et al.
Fractal-Based Description of Natural Scenes
,
1984,
IEEE Transactions on Pattern Analysis and Machine Intelligence.
[2]
Sheldon Gruber,et al.
Measurement of surface roughness parameter using a neural network and laser scattering
,
1991
.
[3]
Koichiro Deguchi,et al.
Texture Characterization and Texture-Based Image Partitioning Using Two-Dimensional Linear Estimation Techniques
,
1978,
IEEE Transactions on Computers.
[4]
Teuvo Kohonen,et al.
Self-Organization and Associative Memory
,
1988
.
[5]
Stephen Grossberg,et al.
Art 2: Self-Organization Of Stable Category Recognition Codes For Analog Input Patterns
,
1988,
Other Conferences.
[6]
Theodore V. Vorburger,et al.
Optical Roughness Measurements Of Industrial Surfaces
,
1986,
Other Conferences.