As the growth of AI continues to impact daily life, many engineers and technical professionals are seeking ways to implement AI into workflows. For mechanical engineers, potential benefits abound in numerous areas, ranging from design to manufacturing and facility maintenance. But even with potential benefits on the radar, some are hesitant to dive in, wondering how to get started with AI. Others wonder how to maintain control of projects and workflows as an engineer — guiding input and verifying output while addressing data security and data ownership issues.
Note: This is the second in a series of articles on artificial intelligence for mechanical engineers. In the first article, we explored how AI can be a partner, rather than a threat, for mechanical engineers seeking to reap the benefits of AI. In this article, we explore practical ways to implement AI in mechanical engineering workflows.
Image source: sdecoret/stock.adobe.com.
Exploring Benefits
In contemplating how to implement AI, consider the potential benefits that might apply to your organization. AI is already proving beneficial in areas such as product design, process automation, simulation, and testing. New applications are continually appearing, as AI matures and engineers discover potential areas in which to employ it.
For product design, AI can help make design processes more efficient, guiding designers through previously tedious processes that relied on trial and error. For example, the SOLIDWORKS Design Assistant includes AI-powered features that guide entity selection and other tasks.
The Selection Helper predicts and offers suggestions on what other edges in a design can be selected based on the single edge or multiple edges already selected by the user. If the user agrees with the prediction, one tap gives the go-ahead, and the Design Assistant automatically picks the appropriate edges.
The Selection Helper isn’t only used for edge selection. It can also identify and select locations in the model where other features can be created, such as end-plates in a frame structure.
The SOLIDWORKS Selection Helper suggests a range of selections based on the size, shape, and orientation of the geometry, along with the initial selection. Image source: SOLIDWORKS. Click image to enlarge.
Proteus Motion, a manufacturer of physical strength measurement equipment, has realized huge time savings using the Selection Helper. The Proteus® hardware consists of integrated software, hardware, and resistance training tools used by athletes for performance testing. With numerous edges in the equipment that need to be smoothed for safety, the edge selection tool has drastically improved the efficiency of filleting sharp edges.
“We do a lot of injection molded parts and smoothing of parts to prevent injury from sharp edges to both our internal manufacturing teams and externally to our athletes,” said Paul Vizzio, director of hardware at Proteus. “Some parts have hundreds of edges that need to be manually clicked. I would previously spend an hour just clicking edges. Being able to click one button and have all the edges smoothed saves me so much time.”
The Design Assistant also includes several other AI-powered features. The Mate Helper automatically inserts multiple instances of components into an assembly. It does this by recognizing and suggesting locations to replicate not only the new components, but also the mates with which they attach to parent components in the assembly. The Sketch Helper duplicates existing or newly created sketch geometry to multiple locations with similar surrounding geometry. Smart Mate helps create constrained mates to surrounding components as the user drags and holds a component in the desired position.
In addition to design assistance, AI can also aid in generative design, an emerging field that uses AI to explore multiple design solutions based on user-defined goals and constraints. MakeByMe, a free personal 3D furniture creation tool from Dassault Systèmes, employs generative design tools for bookshelves, stools, desks, coffee tables, and doors. By specifying the bounding box dimensions and furniture type, a user can generate manufacturable furniture models and functional 3D shapes.
MakeByMe uses generative design for creation of furniture such as bookshelves, coffee tables, desks, and chairs. Image source: SOLIDWORKS.
Generative design and other AI tools can also help automate redundant tasks during design, production, and other product lifecycle stages. For example, designers of rail-based conveyance systems may need to design rails to carry different payload capacities. Instead of designing separate systems for different capacities, the designers could use AI to generate multiple designs that maintain certain constraints and design requirements, such as wheelbase dimensions. The AI-aided process can reduce months of work to days, and possibly even hours, according to Dr. Shrikant Savant, Data Analysis & Science Director, with SOLIDWORKS.
AI can also help with predictive maintenance of equipment, such as manufacturing systems and components. Traditionally, manufacturing teams have relied heavily on reactive maintenance approaches, repairing or replacing equipment when problems occur. With predictive maintenance, issues can be anticipated prior to occurrence, a difficult task with numerous factors to consider, such as design life, operational history, lubrication levels, temperature, and other environmental conditions.
AI can help solve complex problems that do not have precise rules or formulas, a common scenario in predictive maintenance, according to Savant. “Multiple things go into the failure of a part. It is hard to develop an equation that will determine when a part is going to fail. This becomes a very important problem that can potentially be tackled with AI. It can look at all these factors and come up with a prediction.”
A key to successful predictive maintenance is gathering sufficient data. “AI is only as good as the quality of the data that you input,” noted Savant. To build robust datasets, firms need to gather specifications, records, and other maintenance-related data from shop supervisors and other resources to train the AI model.
AI in predictive maintenance uses information from monitoring sensors, machine learning, and analyzes data to predict and prevent equipment failures. Image source Phanuwhat/stock.adobe.com.
Product simulation and testing can also be aided by AI. For example, AI can help simulate product behavior during and after manufacturing, guiding cost-benefit analyses and design decisions, such as material selection. When different material options are available, one material may offer certain benefits, but may increase shipping costs. “Each decision has an implication, but with a framework in place, AI can help you make the right decisions at the right time,” said Savant. While still in relative infancy, AI-guided simulation is “a tremendously powerful tool,” he noted.
Getting Started
To get started with AI, many have found publicly available tools helpful in learning AI fundamentals. OpenAI’s ChatGPT enables users to ask questions in conversational text and receive responses in near real time. Textual responses can be tailored to specified parameters such as length, format, style, level of detail, and language. OpenAI’s DALL·E creates images from text prompts using ChatGPT.
Google’s Gemini is an AI assistant that can generate both text and images using natural language prompts. As an interface to a multimodal large language model (LLM), Gemini can be used to compose text, create images, and answer questions.
Other publicly available AI tools include Adobe’s Firefly for image generation, and Microsoft and OpenAI’s GitHub Copilot for code generation and other tasks. Virtual assistants such as Apple’s Siri and Amazon’s Alexa, also employ AI.
AI-powered tools integrated in commercial software, such as the previously mentioned SOLIDWORKS Design Assistant tools, can also be helpful in learning AI principles. While the AI technology is primarily working behind the scenes, these tools can demonstrate how AI helps make decisions and reduce human effort by handling redundant tasks.
For development of custom AI tools, engineers can use tools developed in various programming languages such as Python, C++, or Java. AI-powered code assistant products such as Continue, Windsurf and Cursor can accelerate code development. A wealth of training and educational tools on these and other AI tools is available on the Internet.
Whichever path is chosen for AI, new users should set manageable goals and evaluate results with understandable metrics, incorporating schedule and budget considerations. Pilot tests may be helpful in testing AI on hypothetical projects before using on actual projects. Current time-consuming processes might be candidates for AI to handle tedious tasks and reduce human effort.
Maintaining Control
When implementing AI, engineers may want to maintain control of processes. Much of this can be addressed by controlling input, verifying output, and establishing appropriate data security.
To control input, users should become familiar with AI prompting, the process of developing input queries (prompts) to guide the output of an AI model, like the LLM model. Prompts should be specific enough to avoid bias — incomplete or unbalanced information that skews outputs — and broad enough to enable AI to consider an appropriate range of possibilities. For example, in solving a complex problem that has many steps, AI might provide one answer to the general public and a different answer if the prompter identifies himself or herself as an engineering professor conducting research with explicit instructions to go into the details of each step in the response.
Input data should also be drawn from appropriate sources. In some cases, internal data based on an organization’s previous projects may be needed, while in other cases, external data may be appropriate. If using external data, users should use data from appropriate, reliable sources. For example, in seeking a cost estimate to build a new facility, the input should specify what other projects to use as basis (e.g., projects of similar size, function, and geographic location).
Organizations should also understand whether they are using AI locally or on a third-party server. Public tools such as ChatGPT and Gemini run on third-party servers and are not run locally. Prompts, data, and other information provided by the user are uploaded to the server and potentially accessible to others. Other AI tools can be run locally, but may not be as powerful, as the results may be based on smaller datasets. This risk can usually be avoided if users subscribe to the LLM and do not use the free versions, or if they have a corporate agreement with the LLM provider about data sharing. Companies should set up AI policies for their employees to make sure they use approved tools and secure company data.
“You have to know what kind of a tool you are using,” said Savant, noting that organizations using public servers should be aware that their data may be shared with others and should be careful about how much data they are willing to share. “Generate queries in a generic way, so that you know no particular information can be deduced from that prompt.”
In developing SOLIDWORKS’ AI-driven tools, no customer data is used without customers’ explicit permission, according to Savant.
AI has also generated numerous ethical and legal questions. The data sharing concept raises questions such as: “Who owns the data and the results?” Organizations should seek to control how much data is shared and be aware of using data from other sources.
Regarding the use of AI-generated results, Savant believes human involvement and engineering judgment play critical roles. “In the field of engineering, ultimately the engineer is still responsible,” he said. “Often, what AI is doing is giving you an estimated result — a prediction based on results that it has seen. Everything you get from AI you have to take with a grain of salt.” He recommends independent verification of all AI results.
Agentic AI is an AI system that acts autonomously to make decisions or actions without constant human oversight. Image source Deemerwha studio/stock.adobe.com.
Looking Ahead
Along with currently available opportunities to apply AI, new and improved prospects are continually being explored. Fields like generative design and agentic AI — where an AI “agent” uses natural-language prompts and sophisticated reasoning to solve complex problems in a human-like manner — continue to gain acceptance and find practical use. Agentic AI can typically gather information, plan or develop a strategy or sequence of actions, make decisions based on goals and feedback, and adjust its behavior over time based on experience or new data. Other examples of agentic AI are AI copilots in software development and design, robotics for warehouses, and autonomous agents for simulations or real-world operations in finance. Future AI applications will likely continue to integrate multiple technologies such as convolutional neural networks (CNNs) that learn features via filter optimization and recurrent neural networks (RNNs) designed for processing sequential data such as text, speech, and time series.
“Tools become obsolete very quickly,” said Savant. “But the fundamentals are still the same. If you understand the basics of how AI works — what it can do and what it cannot do — you are better off.”
In our next article, we’ll review how to become forward-thinking with AI technology to build confidence, creativity, and competitive advantages.
This article was sponsored by SOLIDWORKS.
###
Searching for more information about Product Design & Manufacturing?
Click here!
Share This Post