Burns & McDonnell

Revolutionizing Engineering Design: Unleashing the Power of Large Language Models

Written by James Crouch | August 1, 2023

Recent and sudden advances in the capabilities of artificial intelligence (AI) have attracted a lot of excitement and public attention. Notably, the emergence of large language models (LLMs) such as Google Bard or ChatGPT has been highlighted for their ability to create substantial volumes of natural text in a short space of time.

Technology Beyond Words

However, the potential uses for LLMs extend far beyond text generation, holding the significant potential to transform how engineers approach the design of projects. The ability of LLMs to analyse, interpret and even generate complex engineering documentation and instructions is an opportunity for engineers to automate repetitive and administrative tasks, giving them the ability to focus on more innovative and creative problem-solving tasks. They should therefore be actively exploring how they can harness AI and integrate it into their existing processes.

LLMs use Natural Language Processing (NLP) techniques to process input data, which allows them to interpret the semantic meanings of codes, clauses, formulas and standards. If fine-tuned correctly, this can allow models to develop accurate relationships between engineering variables and requirements within their neural networks. In turn, engineers can leverage this capability to develop systems that can automatically carry out appropriate calculations, based on the relevant formulas and clauses required for engineering design.

Just as a human would, such systems would be able to summarise proposals in a way that is simple to read and understand, including technical descriptions and suggestions. AI could act as an engineering design assistant, capable of making informed decisions and applying codified standards to repetitive types of work. This would significantly enhance productivity and efficiency in engineering workflows.

Benefits of Using AI as a Design Assistant

  • Ability to rapidly and efficiently generate numerous potential solutions: If given a problem statement or a design specification as input, LLMs could generate several potential solutions, which would be a valuable starting point for refinement and further exploration by humans, resulting in a significantly broader design space to uncover innovative approaches.
  • Automation of many tedious engineering process tasks: For example, CAD models are utilised for structure design and analysis, but can be time-consuming to produce and require a high level of knowledge to be used effectively. However, engineers have the option to leverage LLMs as an alternative approach. By inputting fundamental design parameters into an LLM, engineers can prompt the system to instantly generate a CAD model — this is done by drawing upon the training on structural engineering principles. Subsequently, the LLM can further refine the design based on additional feedback and input provided by the engineer, streamlining the iterative design process and reducing the burden on the engineer.
  • Potential for designing more innovative solutions than those devised solely by human engineers: The vast amounts of data that LLMs are trained on means that they can identify patterns and relationships that are not immediately apparent to human designers. Through this unique capability, AI could suggest a material or configuration that increases efficiency or has superior performance compared to more conventional human-led designs. By leveraging AI's ability to uncover hidden insights within complex data sets, engineers can explore novel design possibilities that push the boundaries of conventional engineering practices, leading to innovative solutions that may have otherwise been overlooked.

Beyond producing designs, LLMs’ ability to generate text will also be valuable for engineers. Based on brief input data provided by the engineer, LLMs can be used to create well-crafted, detailed technical reports on a variety of topics ranging from the design of a new product to the performance of an existing system. By leveraging LLMs for this purpose, engineers can save a significant amount of time and effort when carrying out the formalities of the design process. This newfound efficiency allows engineers to reinvest their saved resources into further refining design solutions or advancing other project deliverables ahead of schedule, thereby enhancing overall productivity and project success.

Potential Challenges With AI

From a Data Collection and Information Standpoint:

As many opportunities as AI offers, there are some associated challenges that will need to be addressed if AI is going to be implemented into the engineering design process. To be effective, LLMs usually need to be trained on large data sets of relevant information. This imposes significant investment in the collection and processing of applicable technical documentation and examples of how to apply engineering principles and material science. It has been demonstrated in recent research that equivalent — and potentially improved — performance of LLMs can be achieved through training on a much smaller but more highly tuned data set. While this saves time in data collection and processing energy, those savings will have to be weighed against the significant additional effort required by engineers to correctly refine such a data set.

Regardless of approach, data will need to be carefully selected to mitigate the risk of undesirable bias, as it may include historical data that reflects incorrect actualities or antiquated premises. For example, a predictive model that uses historical traffic patterns to forecast future traffic may reflect historical segregation and inequitable access to transportation. Failing to properly account for these biases can lead to generative AI systems that perpetuate inequality rather than innovatively solve it.

Engineers will need to be confident that the tools they use are trained on a diverse and representative data set. This underscores the critical importance of meticulous curation and thoughtful consideration when selecting data to train AI models, allowing for a more inclusive and unbiased foundation upon which AI-driven engineering design processes can be built.

Striking the right balance between data set size, quality and human effort remains a crucial consideration in effectively harnessing the potential of LLMs for the engineering design process.

From a Decision-Making Standpoint:

Engineers will also need to be able to understand how the system has produced its outputs so that the necessary engineering due diligence can take place. This will require interpretability tools to be built into the AI system that provide design approvers with visibility into the decision-making process employed by the model. By providing transparency and insights into the internal workings of the LLM, engineers can effectively assess the validity, reliability and safety of the generated design outputs. This allows the necessary engineering rigor to be upheld, granting informed decision-making and permitting the integration of AI technologies into the engineering design process, which aligns with established standards and practices.

Careful oversight of LLMs’ decision-making will be a very important role for engineers, given the potential for errors. It is essential to acknowledge that these models are not infallible and may occasionally ‘hallucinate’ or generate inaccurate outputs. To address this concern, it is important to carefully evaluate the outputs of these models using techniques such as ensemble methods, which aggregate predictions from multiple models, to help enhance accuracy and robustness.

Additionally, incorporating human-in-the-loop verification mechanisms becomes vital. Engaging and leveraging knowledgeable individuals to review and validate the LLM outputs against expected standards and requirements will remain crucial for design assurance.

By combining ensemble methods and human verification, engineers can enhance the reliability and trustworthiness of LLM outputs, mitigating potential errors and promoting the responsible integration of AI into the engineering design process.

Keeping Up With the Ever-Evolving World

Despite these various difficulties, AI is being taken up by almost every industry, and engineering is unlikely to be an exception. Engineers can expect a continuing rate of adoption and improvement with AI-enhanced project deliveries becoming more commonplace. They will need to be conscious of the limitations of LLMs and put in place measures to mitigate them. By leveraging AI capabilities, engineers can augment their design skills and capabilities, enabling them to deliver high-quality outcomes more efficiently. The integration of AI into the engineering design process empowers engineers to tackle complex challenges, improve decision-making and enhance overall project performance.

Overall, while being mindful of the limitations, engineers can embrace the potential of AI to reshape the engineering landscape, paving the way for improved project deliveries and enhanced client satisfaction.

 

There are manageable actions you can take when considering automation in design or engineering.