Product Development How AI is Advancing Ship Development

Shipbuilding is one of the oldest industries, with humans having built ships for thousands of years. However, it is time to change the approach to the development of ships, and specifically ship hulls.

Claudia Becker is the Managing Director of Edgize GmbH.

Today, about 90 percent of global trade is conducted via waterways. In 2020, the shipping sector emitted almost 3 percent of CO₂ emissions. Without extensive changes, annual emissions from shipping could rise to 18 percent of total emissions by 2050. "Net zero" cannot be achieved solely with novel fuels such as hydrogen; ship design and engineering must radically change.

Quickly Generate And Optimize Variants

A novel tool for this has been developed by the company Compute Maritime with the Neural-Shipper. Neural-Shipper is a design space exploration tool that can generate and optimize shape variants of ship hulls, propellers, hydrofoils, wind-assisted propulsion systems, and energy-saving devices (ESD) within minutes. The generated designs are so efficiently flow-optimized that sustainability goals can be incorporated into the optimization of the design process in the future.

Gallery

Unlike conventional parametric models that are optimized for only one ship type, the Neural-Shipper solution is based on transformer and diffusion models trained on over 100,000 different ship shapes. With deep learning, the software accelerates real-time insights into design decisions and complex simulations, aiding designers in making informed decisions that improve fuel efficiency, reduce emissions, and support the achievement of long-term net-zero goals.

In A Boat With HP

Compute Maritime relies on the Z by HP Workstation Z8 Fury. With up to four Nvidia RTX 6000 Ada GPUs, it handles computationally intensive machine learning. "We need CPU power for the simulation processes in Neural-Shipper, but powerful GPUs are essential for transformer and diffusion models. Cloud computing was an option, but it would have been too insecure, inflexible, and expensive," says Junaid Awan. Awan, who was involved in the development of the predecessor to Neural-Shipper, founded Compute Maritime and is now driving the commercial development of Neural-Shipper with his team. "Without the Z8 Fury Workstation and its computing power, we could never have developed this tool," says Awan. The workstations offer exceptional performance for local AI training and inference or to complement data center and cloud resources for larger workloads.

Why the HP Workstations are Used

"Neural-Shipper was trained on over 100,000 ship designs. When customers saw our progress, they were so impressed that some of them provided us with their sensitive, IP-protected data for the model's training. We couldn't upload these data to the cloud, but with the Z8 Fury, we were able to securely process and store the data, which further built trust and brought us even more data," explains Awan. "At the start of our development, we had to conduct many extensive tests," recalls Junaid. "Therefore, we couldn't use the cloud for two reasons: We didn't need continuous computing power, but rather just-on-the-spot power for our extensive tests. The costs of a continuous cloud setup with the corresponding computing power would have exceeded our budget since we were still in the research phase. Secondly, data security was a major reason for working with the Z8 Fury from Z by HP. It was crucial for us that our data is secure. These were the main reasons to develop the tool on the Z by HP workstations."

It was crucial for us that our data is secure. These were the main reasons for developing the tool on the Z by HP Workstations.

Junaid Awan

More than Parametric Modeling

Neural-Shipper goes beyond parametric modeling—the software understands the semantics of the shape of objects in motion, evaluates performance, and optimizes hulls with respect to flow. The Neural-Shipper model is based on CAD 3D hull geometries rather than images. This means that developing Neural-Shipper requires more than just a powerful GPU and CPU. Instead of image processing, Neural-Shipper processes geometries in the form of CAD geometries.

"The ship designs and the way these designs are fed in are a problem in the 3D and deep learning environment. Therefore, the resolution at which you provide these designs to the AI model is crucial. When we input a higher-resolution 3D design, the architecture of the models becomes more complex and we need more RAM. The 2 TB ECC RAM of the Z8 Fury leads to a higher resolution, which increases the quality of our end product," adds Awan.

Technology Also Suitable for Other Engineering Disciplines

The truly fascinating aspect of Compute Maritime's technology is its potential application in other engineering disciplines, where it could bring about revolutionary changes in the design process. Compute Maritime has developed a technology with Neural-Shipper that understands the semantics of design shapes in flows, evaluates the shapes in terms of their performance, and flow-optimizes the generated hulls.

Easy Use of the Tool

For the user, Neural-Shipper is very easy to use. They only need to enter their respective parameters and constraints into the interface and can quickly generate different designs. The design constraints can be narrowed down with additional parameters to explore a tighter design space. Hydrostatic parameters become immediately visible in Neural-Shipper. At the end of the process, the designer receives a CAD file for further processing. This way, with Neural-Shipper, a development task that can take several days and weeks with conventional CAD design methods is reduced to just a few minutes.

Significance for Engineering

These new thinking and design tools will change the engineer's activities in the design process: In the future, they will need to formulate their ideas and concepts precisely so that they can be implemented by the machine. The engineer will select, evaluate, and curate the generated designs. In this way, generative AI will not only revolutionize shipbuilding but also the engineering discipline.

A major European aerospace company asked us if we could also design spacecraft.

Junaid Awan

Although Compute Maritime was only founded in mid-2024, the company can already boast considerable successes. In addition to the partnership with HP, Neural-Shipper has recently been integrated into the Siemens Simcenter Star-CCM+ platform. Other industries are also already seeking support.

Smart-Factory

Access solutions for the Smart Factory - Industry 4.0