AI enters the design department: What is truly lacking is not CAD data, but the logic of design decision-making

Generative Artificial Intelligence is entering the design departments of the manufacturing industry, with a growing number of application scenarios ranging from drawing search, design document compilation, historical defect retrieval, to DRBFM draft generation and design review problem sorting.

Many enterprises therefore believe that the design department has accumulated a large number of 3D models, 2D drawings, BOMs and technical specifications, and as long as these materials are handed over to AI, design efficiency can be significantly improved. However, before introducing AI, enterprises need to answer a question first: Is the existing CAD data really sufficient to support AI to make design judgments?

Drawings can tell AI "what the product is designed to be", but cannot necessarily explain "why it is designed this way". And the latter is the key that determines whether AI can be truly applied to R&D and design.

Enterprises have preserved design results, but not necessarily the design logic

CAD data is the result of design, while design knowledge is the decision-making logic that forms this result. A drawing can record dimensions, shapes, tolerances, materials and technical requirements, but usually cannot fully explain why this structure is adopted, and what the dimensions and tolerances are determined based on.

What functional problems this part solves:

Which parts can be modified,

Which parts cannot be modified,

And what risks may arise after modification.

Many enterprises seem to have a huge design database, but in fact what they store are mainly the final files. As for how customer requirements are converted into functions, how different solutions are compared, how design risks are judged, and how test results affect the final decision, these contents are often scattered in meeting minutes, emails and personal experience, and even only exist in the minds of senior engineers.

- When the personnel are still in the company, this knowledge can still be obtained through oral inquiry;

- After the personnel leave or retire, what the enterprise leaves may only be a "correct drawing", but no one can explain why it is correct.

Therefore, after generative AI enters the design department, what is first exposed may not be a technical problem, but the long-standing knowledge management gap of the enterprise: there is a large amount of data, but the knowledge that can support design decision-making is incomplete.

AI may improve search efficiency, but also accelerate the reuse of errors

After receiving a new project, designers usually look for similar products from the past, and modify the dimensions, materials or local structures on the original drawings. This method seems efficient, but "similar in shape" does not mean "reusable".

To judge whether a design can be reused, we cannot only look at the drawings, but also compare the functional requirements, usage environments, loads, service lives, materials, manufacturing conditions and verification standards of the new and old products. At the same time, we need to understand whether the original design has had quality problems, which structures have been fully verified, and which areas are non-changeable zones.

In reality, designers tend to choose products they are familiar with, projects they have done before, or drawings that are easiest to find. However, the solution that is truly suitable for reuse is not necessarily the most familiar one, but the one that best meets the current needs after comprehensive judgment of functions, risks, verification and manufacturing conditions.

If enterprises do not establish these judgment bases and directly let AI search for and recommend similar drawings, AI may give a solution that is highly similar in form and very reasonable in description, but cannot identify the hidden usage boundaries and historical risks behind the drawings.

This means that although AI improves the search speed, it may also accelerate the reuse of wrong designs and historical defects. The problem is not that AI does not search fast enough, but that enterprises have not told AI what can be reused, under what conditions it can be reused, and what must be re-verified after changes.

For AI to truly serve design, enterprises need to build a design knowledge base

The design knowledge base cannot only be a centralized storage of drawings, CAD models and technical documents, but should further record the basis behind the design results.

Taking a key part as an example, in addition to storing the drawing, the enterprise should also record the functions it needs to achieve, the reasons for adopting the current structure and material, the determination basis of dimensions and tolerances, applicable usage conditions, design calculation and verification results, historical failure cases, actual fluctuations in mass production, as well as the conditions for reusability and prohibition of modification.

Only when this information is associated can past design achievements be transformed into design knowledge that can be reused by the organization. On this basis, when designers input new customer requirements or change requirements, AI can assist in retrieving similar solutions, historical defects, design standards and verification records, and further prompt the functions, parts, processes, tests and risks that may be affected by the change.

In the conceptual design stage, AI can help designers quickly compare the technical routes adopted in the past, understand the applicable conditions and historical problems of different solutions, and reduce repeated discussions and waiting for confirmation. In the detailed design stage, AI can help judge whether the drawings meet the reuse conditions, identify the chain effects that may be caused by changes in dimensions, materials or structures, and prompt the verification items that need to be supplemented.

Therefore, at the current stage, the most reasonable role of AI in the design department is not to replace engineers to make final decisions, but to help engineers collect, associate and organize the information required for decision-making. AI is responsible for expanding the scope of information, and humans are responsible for completing engineering judgments.

AI will promote design review from "spotting errors on site" to "risk-based decision-making"

Generative AI also has important value in design review, namely the DR stage. The design review of many enterprises still adopts the traditional way: the designer introduces the solution on site, and other departments temporarily check the drawings and raise questions. Many problems are found in the meeting, but due to the lack of background materials and preliminary analysis, it is difficult to form clear measures on the spot. In the end, the review easily becomes a "problem-raising meeting" rather than a "problem-solving meeting".

A truly effective design review should identify risks as much as possible before the meeting. The focus of the on-site meeting is not to spot errors temporarily, but to judge whether existing measures can control risks, whether additional tests are needed, which parts and processes will be affected by the changes, whether the remaining risks are acceptable, and who will complete the corresponding measures at what time.

If AI can associate historical defects, similar designs, design standards, verification records and non-changeable areas before DR, it can form risk prompts and review questions in advance. Participants do not need to spend a lot of time looking up basic information, but can focus on discussing response measures, resource input and final decisions.

This division of labor is clearer: AI assists in identifying problems, and humans evaluate risks and determine solutions. As a result, design review has gradually transformed from on-site judgment relying on personal experience to collective decision-making based on enterprise knowledge and historical evidence.

In the AI era, designers who can explain design logic are more needed

As drawing search, document compilation, general risk prompting and standard solution generation are gradually completed by AI, designers who only know how to operate CAD, partially modify the original drawings or directly adopt AI solutions will find it increasingly difficult to reflect their unique value.

The truly important capability in the future is to be able to answer the "why" behind the design:

Why is this function needed?

Why is this implementation method adopted?

Why are such materials, dimensions and tolerances selected?

Why cannot this part be modified?

If a change occurs, which performances and verification items will be affected?

This requires designers not to stay at the stage of "drawing the picture", but to be able to convert customer requirements into product functions, and then convert functions into structures, materials, dimensions, tolerances and verification requirements. At the same time, they need to integrate information provided by sales, quality, manufacturing, procurement and supply chain to make judgments on the final technical solution.

The role of senior designers will also change. In the past, they mostly told young people directly "what to do". In the future, they also need to constantly ask:

"Why is it designed this way?"

"What is the judgment basis?"

"Why are other solutions not applicable?"

"Does this measure really eliminate the cause of failure?"

Through such inquiries, personal experience can be gradually transformed into knowledge that the organization can record, reuse and pass on.

In this sense, the design knowledge base is not just a database, but more like the "design operating system" of the enterprise: the CAD system stores design results, the knowledge base stores decision-making logic, AI is responsible for searching and associating these logics, and finally the designer completes the judgment.

Before introducing design AI, enterprises should first check their knowledge foundation

Before introducing generative AI, enterprises may wish to check several questions first:

Does the existing system only store drawings and CAD data?

Do the bases for selecting dimensions, tolerances and materials still exist?

Are historical defects associated with the corresponding parts and design decisions?

Are the areas that cannot be modified clearly defined?

Are there clear rules for which items need to be re-verified after the design changes?

If these questions cannot be answered, what the enterprise most needs to do at present may not be to immediately deploy an AI tool, but to sort out the design knowledge and decision-making logic first.

Designers in the future need to complete three transformations:

From being able to operate CAD to being able to understand and build design concepts;

From being able to find and copy historical drawings to being able to judge whether a design has reuse conditions;

From relying on AI to provide answers to using AI to collect information and make independent engineering decisions.

Generative AI will not lower the capability requirements for designers, on the contrary, it will make basic design capabilities, risk awareness and engineering judgment more important. AI can find past design results, but only humans can judge whether these results are applicable to the future.