BauKI

BauKI • Capturing and plausibility check of construction plans

As part of the "Baden-Württemberg AI Innovation Competition" for the research and development of innovations in the field of artificial intelligence, d-serv GmbH from Tübingen is starting the joint research project "KI4VPB - Virtual Planning Office" in spring 2021. Together with the development partners, the aim of the project is to develop an AI-supported recording and plausibility check of construction plans and to make it available as a prototype or basis for later product development by September 2021.

During the planning and construction phase of construction projects, it is not uncommon for several thousand plans to be created and must be available to a large number of project participants (property developers, construction companies, trade companies, architectural offices, planning authorities) at all times. The number of plans often exceeds the capacity available for viewing and evaluating the sometimes heterogeneous data material. The most time-consuming factors in the viewing and evaluation process include the manual categorisation and structuring of the building plans and the addition of descriptive meta-information.

One aim of the project is to simplify and accelerate the processes by using an AI-supported recording and plausibility check of construction plans by automatically structuring, indexing and annotating construction plans. Ideally, new plans that are added can also be compared with existing, recorded plans on the basis of a plausibility check and validation, making it possible to identify any planning errors in construction projects at an early stage. In addition, the plans can be found quickly and used easily by all those involved in a construction project without the additional help that is otherwise often required.

Thanks to its 20 years of software and product development, d-serv GmbH is able to contribute its experience and expertise in digitally provided and editable plans in PDF format, which, in combination with the knowledge of its development partners, will provide a web-based software solution for recording and checking the plausibility of construction plans for architects' offices, planning authorities and property developers.

Development partner

• The Chainless GmbH: Expertise in deep learning-based classification, regression and semantic segmentation tasks, as well as in clustering methods of high-dimensional data and in the development of metric learning methods (Few-shot) to reduce training complexity.

• Loidl GmbH: Role of the consultant in the field of digital construction. In connection with the increased demands on the realisation of commercial buildings, Loidl provides its expertise as a data, planning and project manager in the realisation of large and very large construction projects.

• Uni Mannheim - Institut für Enterprise Systems (InES): The aims of InES are to achieve high-quality research results by pursuing an interdisciplinary research approach and the transfer of knowledge into practice with a focus on manufacturers and users in the SME and enterprise sector, as well as the qualification of young scientists.

• bwcon research gGmbH: The non-profit research organisation, which does not pursue any commercial interests, has the appropriate methods and experience to successfully coordinate complex projects. In addition, it provides training in thinking in data-based business models as well as associated AI experts.

Funders

This product development was funded by the Baden-Württemberg Ministry of Economic Affairs, Labour and Tourism as part of the "Baden-Württemberg AI Innovation Competition". All recognised and funded projects were officially announced on 10 January 2021; More information can be found on the Ministry's website - only available in German - at: Wirtschaftsministerium fördert 44 Innovationsprojekte für mehr »KI - Made in Baden-Württemberg«

Result: The "Virtual Planning Office"

The web-based "Virtual Planning Office" - available as a stand-alone SaaS solution or as a fully integrable white label solution - can record, evaluate, comprehensively categorise and promptly provide construction plans, whereby the developed "BauKI" differentiates between plan content and plan header.

In the plan content, "BauKI" recognises the specific representation type of a plan and in the plan header - regardless of its position in the plan - the scale, the index and a few more parameters. The metadata obtained is stored in a database and assigned to the respective plans.

The recording of such data, which can be time-consuming, error-prone and incomplete in practice, has thus been significantly optimised. In addition, the retrieval and utilisation of plans is accelerated or made possible without additional resources.

Thanks to the "Virtual Planning Office" and its specific "BauKI", construction plans can be provided more promptly and categorised more comprehensively than with conventional working methods.

Functionality

• The "BauKI" can differentiate between plan content and plan header in a plan.
• In the plan content, it recognises the representation type of a plan. In the plan header, information such as description, scale, index etc. is read out. The position of the plan header in the plan is not relevant.
• The plan recognition also works with scanned plans, which means, for example, that plan inventories in paper form can be recorded and analysed.
• The "BauKI" analyses the plans provided individually and transfers the metadata read back to the "Virtual Planning Office" database.
• The "Virtual Planning Office" then assigns the metadata to the corresponding plan.

Visualisation

• The Gradcam visualisation makes the areas recorded and analysed by the "BauKI" visible.
• The "BauKI" recognises, for example, whether it is a view and/or a section.
• If several representations are recognised, the analysed plan is assigned several times accordingly.

Planning documents were provided by: www.bgs-architekten.ch

Advantages

• The "Virtual Planning Office" gives users access to the management of required plans.
• The "BauKI" analyses uploaded plans and then transmits the results of its categorisations and evaluations.
• This results in the following advantages:
  • Central storage location for all plans.
  • Accessible to all participants at all times.
  • Quick capture of large plan inventories.
  • Categorisation of several plan types.
  • Analyses in ongoing projects.
  • Significantly lower error rates than with manual analyses.