With a probe for gas, oil and other pipelines a huge number of ultrasonic readings of the wall condition is collected. Based on the recorded wall thicknesses of this so-called pipe pig the Re search Center for Computer Science (FZI) has de veloped an automatic inspection system called NeuroPipe. NeuroPipe has the task to detect defects like metal loss. The kernel of this inspection tool is a hybrid neural classifier which was trained using manually collected defect examples by the Pipetronix company. This paper focuses on some aspects of successful use of learning methods in an industrial application and on the difficulty of interpretation of sometimes faulty sensor measurements.
Neural network, interactive learning process, hybrid pattern recognition
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).