At STRUCINSPECT, we have set ourselves the goal of advancing digital structural inspections. A significant advancement in the trustworthy use of AI was achieved in the course of the project “Breakthrough in AI for damage detection on concrete structures“. The project was supported within the framework of a funding program of the Austrian Wirtschaftsservice (AWS). On the one hand, the focus of the project lied on the development of support opportunities during everyday work tasks and secondly on an objective approach to the examination procedure of concrete structures.
Our project is currently nominated for the eAward 2022, an Austrian business prize for IT-related projects, which is awarded annually by the Report Verlag in cooperation with the Plattform Digitales Österreich des Bundeskanzleramts
AI-Level 2: now even more trustworthy!
The automated detection of damages was possible even before the project with the help of the Infrastructure Lifecycle Hub – the online collaboration platform by STRUCINSPECT – thus building the basis for digitalized damage reports for structural inspections.
The aim of the project was firstly to develop a method to identify identical damages pictured on an array of photographs as the same ones and secondly to achieve the ability of recognizing and thereupon categorizing similar damages.
Up to now, the recognition of multiple imaged damages (i.e. the same damages on several images) was a major difficulty in dealing with the results. Especially due to the specification of a necessary overlapping of the images during the acquisition, this situation occurs in almost all digitised inspection processes.
In making use of the new technological possibilities, the project looked at the already available manuals and guidelines regarding the trustworthiness of AI systems. Special attention was paid to the European Union’s “Ethics Guidelines for Trustworthy AI“, which emphasises the following three main components:
- lawful
- robust
- ethical
The existing AI was evaluated and newly assessed against the backdrop of these guidelines. As a result, additional measures were taken to ensure the ability to meet the requirements. In addition to the technical requirements, topics such as transparency, human supervision during the process and the control of results were also dealt with.
To learn more about the trustworthy use of AI, click here.
Identification of same damages with the help of the duplicate search
A first achievement of the project is the duplicate search. As already described, so far, the AI was not able to recognize and tag identical damages as such. The therefore much higher number of identified damages had to be examined and corrected by the test engineer in retrospect. For this, the inspection engineers had to manually check the results of the AI in different images and then correct them.
The automatic detection of the same damages can massively reduce the number of raw results of the AI and subsequently also the manual correction effort for engineers.
Figure 1 below shows a duplicate search. It shows the same damage on different images with different lighting conditions, sizes and image details. The now revised system was able to automatically identify the same damage as a duplicate.
In addition to duplicate search, which allows engineers to identify damage seen in multiple images as the same, another significant skill was gained during the project.
Group comparable damages with the similarity search in one step
STRUCINSPECT works with a large amount of information (especially image data) from building inspections. Similarity search is an AI methodology that is capable of grouping large amounts of data with similar criteria. This capability is particularly relevant to the technology sector.
Fig. 3: Automatic clustering and condition assessment
“Corrosion damage class 2 – medium”
Fig. 3: Automatic clustering and condition assessment
“Corrosion damage class 2 – medium”
With the help of this functionality, the following use cases can now be covered:
- By selecting a damage, similar damages can be found
- All damages are automatically subdivided into useful groups
- These can in turn be filtered according to the components (or similar) of a structure
- Classification of cracks of different widths
- Improved search for “false positives” (detections that are not damage)
Thus, in addition to the damage detected by the AI, manually added damage can also be categorised.
The main advantage is that, for example, damages of similar severity can be grouped and assessed in just one common step. This not only contributes to a faster assessment of the condition of the structure, but also to a clearer and, above all, more accurate procedure.
How you can take advantage of digital damage detection with STRUCINSPECT
Over the past year, there has been a particularly strong increase in interest in AI-based services in the field of structural inspection. With the STRUCINSPECT Infrastructure Lifecycle Hub, we offer you a platform where you can complete your infrastructure building management – from automated damage detection and inspection to lifecycle analysis – all in one place.
The new AI Level 2 achievements – duplicate and similarity search – will be made available to you, the inspection engineers and building operators, as filters on the platform. In addition, in phase 1 of the launch, all inspections in which this technology is used will initially be supervised by STRUCINSPECT experts. This serves to increase traceability and trustworthiness. In this way, it is possible to test the new functions for their practical suitability and at the same time significantly improve quality control.
Above all, the practical application of the similarity search represents a real AIvolution and an extreme simplification of work for test engineers. Whereas they previously had to “manually” sort through an unbelievable amount of similar damages – up to several thousand(!) depending on the project – the AI can now make a suggestion for filtering similar damages.
However, the guiding principle of human autonomy is always central – the human makes the decision, the AI supports him or her in working objectively and efficiently. Thus, the AI’s filter suggestions are made exclusively on the basis of examples that the human shows or teaches it. Humans and AI thus interact and can benefit from each other even more than before.
Where AI Level 2 already finds its application
In practice, this method is already being used in a pilot project in the USA. Tunnels of one of the largest infrastructure operators in the country are being inspected and enormous increases in efficiency and objectivity have been observed. This also shows that the existing system has been massively developed further, especially through the similarity search, which is why we speak of AI level 2 in this context.
STRUCINSPECT has thus once again succeeded in positioning itself as a global pioneer in the field of AI in building inspection.
Would you also like to benefit from the advantages of AI Level 2?
