Techniques for High Temperature Ultrasonic Inspection of Arc Welding – State of the Art Review
By Linghao Zhou and Kai Yang
CEng, MWeldi, MBCS
Background
Inspections of welds using Ultrasonic Testing (UT) and advanced UT techniques such as phased array UT (PAUT) and Full Matrix Capture (FMC) usually take place after the welding process is completed and the surface condition is prepared for inspection. Indications that are discovered and considered to be unacceptable at this stage often have significant adverse impact on subsequent schedules and induce additional cost. Therefore, there is an industry need to develop the capability of weld inspection in elevated temperature condition, during or shortly after the welding process. However, it is known that high temperature conditions will change the ultrasonic wave travel speed in materials and increase the attenuation of wave energy due to intensified molecule vibrations, hindering sensitivity and accuracy of the inspection. Dynamic temperature conditions will make the issue even more challenging, where the ultrasonic wave speed, propagation angles and attenuation changes are varying in time during the welding process.
In the Core Research Programme (CRP) project, experimental work will be carried out to simulate an arc welding environment, with testing samples heated up to 450°C. A high temperature jacket (HTJacket) system holding a linear array probe and high temperature wedges, as well as a temperature recording system, will be deployed to take measurements of ultrasonic echoes and the dynamic temperature.
Utilising the experimental data obtained, the project will investigate the feasibility and effectiveness of machine learning (ML) techniques to address the adverse impacts of the dynamic high temperature field on the quality and accuracy of the FMC inspection images. Additionally, a comparative study using conventional image processing techniques will be undertaken.
Key Findings
- In BS EN ISO 16809:2019, it is recommended that the speed of longitudinal wave be compensated by 0.8 m/s/°C. In contrast, ASTM E797 recommends that the ultrasound velocity in steel in elevated temperature condition is reduced by 1% per 55°C (1.07 m/s/°C equivalent). Both recommendations are made for longitudinal waves but not shear waves. At the time of writing, no standards are found to instruct on shear wave speed compensation in elevated temperature condition in this stage and so these values will need to be either measured or calculated from known ambient relationships.
- Regression, classification and neural networks ML methods were reviewed and discussed. Non-neural network regression and classification methods are generally simpler to implement and train, but cannot compete in performance to well-designed neural network (NN). Multilayer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs) are promising methods that are being explored for use in NDE and ultrasonic inspections.
- Conventional image processing techniques including stretch-related approaches and matrix decomposition methods were reviewed and will be explored for use in the CRP.
- Commercial UT systems that are designed for high temperature applications were reviewed using publically available information, as well as related research projects. High temperature probes and wedges from different manufactures can be adopted in the event that the high temperature experiemtns damage current experimental equipment. No commercially available system equipped with dynamic temperature correction capability was identified during the review.
- Python has been selected for algorithm development as it is widely supported within the data science and ML fields, and it facilitates fast prototyping. C++ and Java are prominent supplementary languages when speed and scalability become essential.
Impact
The literature review fulfils the objectives listed below:
- To understand what is established in current industrial codes and standards on the effect and rectification of impact due to high temperature during welding processes.
- To enrich knowledge pool based on published research in peer-reviewed journals and high quality conference proceedings.
- To have an understanding of technology currently available in the market that is designed to mitigate high temperature effects.
- To study the state-of-the-art ML methods and to investigate the feasibility of applying them to enhance the TFM image quality.
- To study the conventional, well-established image processing techniques and evaluate their effectiveness, facilitating comparative analysis to ML methods.