Automated detection, measurement and classification of defects in CANDU reactor pressure tubes
Abstract
During scheduled maintenance outages of CANDU nuclear reactors, detailed
ultrasonic inspections are undertaken of selected pressure tubes in
order to confirm their integrity. The requirement for detailed analysis
of the large volume of generated data prior to restart of the reactor
results in the analysis process playing a significant role on the outage
critical path. In order to support the existing analysis process, a
prototype knowledge-based system has been developed which can automate
aspects of the analysis process.
Based on knowledge captured from interviews and testing conducted alongside human experts, the system undertakes the analysis process significantly faster than is possible for a human operator, albeit without the extensive historical and contextual information available to a human expert. Testing has demonstrated the viability of the automated analysis process to provide useful decision support to the human expert. Work is ongoing to improve the accuracy of the system, particularly targeting edge cases, which will in turn lead to improved overall efficiency and consistency.
This paper presents an overview of the existing ultrasonic inspection techniques, the analysis and interpretation of the ultrasonic data and describes the process of codifying expert knowledge into the prototype system. Discussion is also presented of the potential to include data driven techniques to supplement the existing knowledge based approach, in order to benefit from the archive of historical data which could lead to further improvements in accuracy and reliability.
Based on knowledge captured from interviews and testing conducted alongside human experts, the system undertakes the analysis process significantly faster than is possible for a human operator, albeit without the extensive historical and contextual information available to a human expert. Testing has demonstrated the viability of the automated analysis process to provide useful decision support to the human expert. Work is ongoing to improve the accuracy of the system, particularly targeting edge cases, which will in turn lead to improved overall efficiency and consistency.
This paper presents an overview of the existing ultrasonic inspection techniques, the analysis and interpretation of the ultrasonic data and describes the process of codifying expert knowledge into the prototype system. Discussion is also presented of the potential to include data driven techniques to supplement the existing knowledge based approach, in order to benefit from the archive of historical data which could lead to further improvements in accuracy and reliability.
