ANYstructure: A Steel Structure Design Optimization Tool Integrating Machine Learning and DNV Standards

ANYstructure is a professional steel structure design optimization tool based on DNV standards, focusing on plate panel and cylindrical structure design in the offshore engineering and shipbuilding fields. It integrates machine learning technology to improve buckling prediction accuracy, supports multi-objective optimization of weight, welds, and costs, and helps engineers achieve an optimal design balance under compliance requirements. The project is open-source (GitHub), maintained by Audun Arnesen Nyhus, and released in June 2026.

In offshore engineering and ship design, structural weight affects load capacity and operational costs, while welding workload is related to construction cycle and costs. Traditional design relies on empirical trial calculations, making it difficult to find optimal solutions under complex constraints. Addressing this pain point, ANYstructure is developed based on Det Norske Veritas (DNV) standards, aiming to solve design optimization problems for plate panels and cylindrical structures through automation and intelligent algorithms.

Multi-objective Optimization: Supports weight (minimize total weight), welds (optimize layout/length), cost (comprehensive material/processing/welding) optimization; DNV Standard Calculations: Covers minimum plate thickness (DNV-OS-C101), section modulus, shear area checks; Buckling Assessment: Includes code-based methods (DNVGL-RP-C201/C202), semi-analytical methods, and machine learning prediction (neural network based on PULS simulation); Fatigue Analysis: Evaluates fatigue life at plate panel/stiffener joints based on DNVGL-RP-C203.

ANYstructure innovatively introduces machine learning to improve buckling prediction. Traditional methods rely on classification rules or time-consuming simulations, while this tool uses neural network models: early versions were classification models, and the latest upgrade is numerical prediction (directly outputting continuous buckling capacity), which has higher accuracy and supports more refined optimization. In addition, it provides model visualization verification functions to compare prediction results with code/simulation results, enhancing engineers' trust.

Automatic Compartment Modeling: Supports automatic boundary recognition and pressure load generation for tank structures, reducing repetitive modeling; Custom Pressure Equations: Users can define mathematical equations to describe external pressure distribution, adapting to special loads; Multi-level Optimization: Supports optimization of single plate panels, multi-panel combinations, cylindrical shells, and double-bottom structures; 3D Visualization: Intuitively displays geometric shapes, stress distributions, and optimization results, facilitating review and reporting.

ANYstructure is developed in Python and maintained as the anystruct package. Installation steps: create a virtual environment → activate → install dependencies → install the package → run tests/start the desktop application (using the ANYstructure command). Modular design: supports core functions, ML modules, Excel interfaces, and other components (Excel import requires local installation of Microsoft Excel).

For engineers: Integrates code calculations, simulation, and intelligent optimization, enabling rapid evaluation of numerous schemes in the early design stage and tapping into weight/cost optimization potential; For researchers/students: It is an excellent case of ML application in traditional engineering fields, demonstrating the path of combining intelligent algorithms with professional codes; The open-source feature promotes industry collaboration and continuous improvement.

ANYstructure represents the development direction of engineering software: adhering to industry standards while incorporating AI technology to improve efficiency and quality. Future plans include supporting Python 3.14, gradually decoupling calculation code from the GUI to enhance maintainability and scalability. It is suitable for in-depth exploration by offshore structure engineers and ML engineering application researchers.