In sectors where the safety requirements are exceptionally high, such as aviation, errors in manufacturing components must be prevented at all costs. The cost involved in checking components for flaws is always a significant factor. A project based in Bremen is creating an algorithm that will work as an adjunct to visual inspections in this sector.
In recent years, 3D printing has moved on from prototyping to being used in industrial applications, even in sectors with exceptionally high safety requirements, such as aviation and aerospace. Nowadays, a number of extremely complex components are produced using additive manufacturing, often from materials such as titanium.
These printed components are not always perfect: they can have cracks, pores or inclusions of other materials which could cause critical failures. For this reason, additively manufactured components undergo a thorough inspection process (as do components produced using "traditional" methods) before being installed in an air plane.
These inspections are performed by specially trained and certified inspection engineers, who use methods such as X-ray computed tomography to screen components.
Joint research project run by ECOMAT members
"Inspection costs are enormously high and can account for up to 50 percent of a component's total costs", said Stefan Neumann, Data Scientist at Testia GmbH, a Bremen company specialising in material inspection, training courses for inspection personnel and engineering, primarily in the aviation sector.
Neumann is part of the "Automated fault detection and evaluation using X-ray computed tomography (CT) data for highly complex 3D metal and fibre composite components" research project, which Testia is running together with the Faserinstitut Bremen (Fibre Institute, Bremen)(German) and the company Kolbes Messtechnik (German)up to the end of 2021.
The objective of this research project is to reduce the enormous amount of time and money involved in inspections and to make additive manufacturing affordable for applications in which its use was previously too expensive.
How algorithms learn to identify flaws
In their efforts to automate the identification of flaws and errors in future, the research team has combined two approaches: one is computed tomography (CT), which is used after a component has been manufactured. It shows where faults have occurred in the component. "International regulations require us to be able to detect very small faults, only slightly thicker than a human hair", said Neumann.
To do this, the CT data is correlated with data gathered by sensors during the printing process itself. "The laser's position is recorded, as is data from other sensors which measure, for example, sounds, temperature and radiation", he went on to explain.
The CT data and sensor data are transferred into a neural network. "Our goal is to teach this neural network how to use the sensor data collected during the 3D printing process to predict the position and size of faults."