Neural Networks and Robotics Applied to the Non-Destructive Inspection of Aircraft
Abstract
We are converting the in phase and quadrature measurements into an image and then using the image to classify fault or no fault eddy current. This has been done using old OCR features developed to recognize handwritten numerical characters which are applicable to this situation. The raw data was first smoothed out by taking the average of the nearest 10 measurements, then plotted on a 50 by 50 grid. The X axis is the in phase measurement normalized to accommodate the maximum and minimum in phase values in the data set. The same was done for quadrature values on the Y axis. An efficient path planning method for collision avoidance has been developed. Fast path planning is achieved by decomposing the 3-dimensional space into a number of 2-dimensional subspaces. A method is devised to work directly with arm postures (configurations) instead of dealing with individual joint angles. These two aspects, namely decomposition and posture control, greatly speed up the path finding procedure and make it possible to perform near real-time planning in a moderately cluttered environment. Meetings were held at McDonnell Douglas, McClelland AFB, and Physical Research Inc., manufacturers of Magneto-Optic Crack Detectors to assess the utility of our research and plans for prototyping during Phase II. Discussions have begun with United Technologies concerning the use of our research for a large Air Force NDI program. Prospects for Phase III follow-on look promising.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1991
- Accession Number
- ADA238312
Entities
People
- Dan Greenwood