Shimizu Develops AI Program Improving Structural Review in the Early Design Stages

Shimizu Corporation (President: Kazuyuki Inoue) developed an AI program called “SYMPREST” to support the structural review of steel structures in the early stages of design process as part of its digital design strategy. At the same time, it developed a cloud-based web application and launched it internally. This will advance workstyle reform for designers by simultaneously improving and streamlining the work ahead of the application of the revised Labor Standards Act in the next fiscal year. Headwaters Co., Ltd., which is engaged in the AI solutions business, and Headwaters Consulting Co., Ltd. cooperated in the development of the program.

Structural review refers to studying and configuring the structural framework and member sections according to the building plans, including the shape and size. Design works have been on the rise in recent years. Frequent changes to plans are still the norm in the early stages of design process , and it is not uncommon for multiple plans to be reviewed simultaneously. These changes and reviews include structural review, increasing the workload of structural designers. In this situation , SYMPREST was developed as a digital design method to streamline structural review and enable the provision of advanced and timely proposals to business operators.

This AI can be applied to relatively regular-shaped steel-framed office buildings up to 60 meters tall. When the shape and dimensions of the office building under review are input, it instantly identifies multiple structural frameworks from the database that roughly match the shape and displays the quantity of the framework members. Ultimately, once the designer selects a structural frame that fits their vision and fine-tunes things such as the span configuration, the AI constructs a 3D model on the monitor that reflects the assumed cross-section of all the members that make up the framework. When a part of the 3D model is clicked, the system displays the cross-sectional dimensions of the member at the clicked location. This 3D model can be converted into input data for the structural analysis program used in the implementation design, substantially reducing the time and effort required to prepare for analysis.

In the development process, Shimizu developed an algorithm to automatically generate structural frameworks data to train the AI. Under the conditions of 15 floors or less (no basement) and a maximum of 10 spans in any direction, the system has generated structural framework data for approximately 1,500 office buildings of various sizes and shapes. By training SYMPREST with this data, Shimizu has improved its proficiency in predicting member sections. It is planned to expand the AI’s functionality by generating structures for high-rise offices and other uses and structural types as a next step.

Shimizu will continue to expand its proprietary digital platform, Shimz DDE, which combines the DNA of monozukuri (superb craftsmanship) with the latest in computational design. The goal is to go beyond structural design and promote the improvement and streamlining of design tasks in general.