Dr.-Ing. Daniel Algernon, MBA is the head of the Nondestructive Evaluation (NDE) Laboratory of the Nuclear Inspectorate with the Swiss Association for Technical Inspections. The section focuses on analysis and R&D of NDE systems, comprising the field of application on nuclear components as well as infrastructure in general.
Nondestructive Testing of Concrete Structures – Processes, Technique Development and Machine Learning
The nondestructive testing (NDT) in civil engineering (NDT-CE), addressing concrete components specifically, comprises interesting and essential tasks, both in the context of aging infrastructure and modern construction. NDT-CE is a subsection of NDT, which still has a low level of standardization. On the one hand, this leads to a high affinity for innovative solutions, while on the other hand, it is facing high cost pressure. With the growing range of test devices, the inclusion of NDT in quality assurance during construction or in condition assessment is continuously increasing. Furthermore, NDT is used to determine unknown component geometries or to localize components inside the structure precisely.
This talk addresses building-diagnostic NDT processes and presents current developments and tasks. While the focus is set on elastic wave methods, such as ultrasound and impact echo, key applications of electromagnetic methods, such as Ground Penetrating Radar (GPR), are also included.
The topic of Artificial Intelligence in terms of Machine Learning is particularly crucial for NDT-CE. It opens up opportunities and comes with particular challenges, which is exemplified in concepts and applications. Deep Learning solutions, consisting of several layers of neural networks, have proven their capabilities. To provide the possibilities of Machine Learning to NDT users in a practical manner, the software tool ECHOLYST A.I. is continuously developed at SVTI. It implements common Deep Learning approaches and targets the NDT-CE methods, including but not limited to impact echo as well as ultrasonics. It enables effective models to be generated efficiently. The procedure is illustrated for detection as well as for regression tasks.