Onshape

Just where will artificial intelligence fit in with CAD software? Here’s a look at where developments stand now, and a preview of what might be coming.

Jean Thilmany, Contributing Editor

Artificial intelligence has a place in the future of computer-aided design technology, but right now, the role AI will play isn’t clear. That’s the view of Jon Hirschtick, chief executive officer of Onshape, which makes cloud-based CAD software. While some CAD makers are delving into AI functionality, the marriage of AI and design software is in the early stages, he says.

“AI has great potential, but so far no one has illustrated how it will unfold,”

Hirschtick says in reference to CAD vendors. “I’m not saying developers are not working on ideas.”

CAD makers would be wise to consider how AI may fit into their software’s growth and expansion. AI should be a $16 billion industry by 2022, according to a projection from research firm Markets and Markets.

AI across industries

First, definitions are in order. Many terms have been bandied about of late, particularly in reference to Industry 4.0, which goes by a number of names, including smart factory and connected factory. Technologies like artificial intelligence, machine learning, big data, Internet of things (IOT), and deep learning will come together to help realize Industry 4.0.

All these technologies are related in that they build upon each other, says Will Ramsey, director of marketing at Nvidia, which designs graphics processing units. The company developed GPU-based deep learning, which uses artificial intelligence to approach problems like cancer detection, weather prediction, and self-driving vehicles. Here’s how Ramsey defines pertinent terms:

“AI is a broad field focused on using computers to do things that require human-level intelligence. It’s been around since the1950s but was little used because it was limited in practical applications.”

“Machine learning enables AI by providing the algorithms that make the machines smarter and thus give AI a way to actually become more intelligent as time goes on.”

Machine learning is what Ramsey calls an approach to AI, meaning a way to use AI for practical applications. The approach uses statistical techniques to construct a model from observed data. It relies on inputs, or what Ramsey calls “extractors” set by the humans programming these machines.

“It’s like the bag-of-word analysis that made spam filters possible,” Ramsey says.

The filters could search for certain words (determined by humans) within messages, then flag those messages as unwanted spam.

Machine learning algorithms can sift through and find insights in large data sets. Combine AI and machine learning and the algorithms become more able to recognize patterns and specific issues, such as—when it comes to something like speech recognition software—accents.

But where does the data used by machine-learning algorithms come from?

“Now with social media, sensors, the internet of things, we have all these data,” Ramsey says. “And we have the challenge of understanding and extracting insights from it.”

His company uses what it calls deep learning, a method that automatically extracts and makes sense of all that information, and continues to learn from it or “learns to think.”

“Using deep learning, the fastest growing segment of AI, computers are now able to learn and recognize patterns from data that were considered too complex for expert written software,” Ramsey says.

What about design?

Hirschtick believes CAD programs will make use of AI, but in a more limited way in the near future, by using information the designer has entered to offer suggestions about design parameters and inputs.

Future programs might offer to the design default values for a shape based on the objects that person has designed in the past. AI would essentially learn what types of products the designer mostly works on and the inputs he or she has regularly used for those products. The suggested values may appear on the user’s screen in a dialog box, Hirschtick says.

“Or AI could offer something like: ‘Gee, I noticed you’ve done pattern of activity several times in a row, do you want that or was that a mistake?’”

And AI could make engineers’ search for needed and necessary parts easier. Hirschtick envisions a program, much like that which appears for Amazon shoppers, in which engineers could type in information about a part they’re searching for “and the program says ‘a lot of times people looking for that part also look at this one,’” he says.

In the future, CAD software users may also ask speech technology software, rather than the CAD-company’s tech-support operators, questions about the software and instantly receive a pertinent, helpful response. The natural-language-processing programs that drive these responses learn how best to answer user questions thanks to machine learning technology.

Such speech technology software could aid fast-growing CAD companies that would otherwise need to train a slew of customer-support employees quickly.

Currently, those AI possibilities remain unrealized, says Hirschtick. “Right now no one has demonstrated any particularly compelling idea with AI.”

Making manufacturing inroads

Other CAD vendors may beg to disagree.

Autodesk is already moving to use AI for customer support, teaming with IBM to create Otto, a digital concierge that uses IBM Watson technology to manage customer and partner inquiries, says Gregg Spratto, vice president of operations at Autodesk. “Watson’s natural-language-processing and deep-learning technologies help Otto understand the intent of customer questions.”

To offer Otto initial “training,” as it were, the Autodesk team fed historical data from chat logs, use cases, and forum posts into the program to ensure it could understand and respond to a wide range of customer queries.

Then, as the project expands, Otto will use machine learning to handle increasingly complex customer requests and will scale up as call-volume grows.

Also, last October, Autodesk announced plans to embed an AI modeling engine into its IoT cloud platform, Fusion Connect. The Eureqa engine is from Nutonian, recently acquired by DataRobot.

With the AI engine on board, the IoT platform will be able to predict product failures or design flaws based on how a product or device is presently functioning in the real world, says Bryan Kester, director of IoT at Autodesk

The pairing is natural, as IoT offers continual feedback on how products are performing in the field, in real time. IoT makes use of sensors and actuators attached to a product that send back continuous information on how the product is operating, moment-by-moment, in the field.

Fusion Connect helps gather information from that network of sensors and actuators as well as upon RFID, Wi-Fi, and a range of other communications and monitoring technologies, Kester says.

The information is then analyzed and output in a format useful to engineers, who can use it to find where improvements can be made to existing product designs and to determine how new products could be designed better designed. All this based on present, real-world operation, he adds.

Similarly, PTC plans this year to link its Creo computer-aided design system, to the company’s ThingWorx IoT development platform. Developers use the platform to build and deploy enterprise-level IoT applications, says Paul Sagar, vice president of product management at PTC.

Though it’s not an AI application, after the ThingWorx and Creo interface is complete, engineers will be able to instrument their CAD model with virtual sensors that act in the same manner as the real-world counterparts do; that is, they monitor and report back about particular features of part or system operation.

These virtual sensors can offer more insight into model behavior than the what-if questioning and virtual experimentation engineers now use to explore model performance, Sagar says. The sensors can help answer questions like: is the virtual system running hot in a certain area? Is airflow too high or too low?

With those questions answered, designers can redesign and repeat the process until they’ve optimized the model to meet—perhaps even exceed–specifications, Sagar continues.

AI aids 3D lookup

Introduced last summer, Autodesk’s Design Graph is another machine learning system that helps users manage 3D content, offering Google search-like functionality for 3D models, says Mike Haley, who leads the machine intelligence group at Autodesk.

“Machine learning and artificial intelligence are starting to make the first inroads into daily life, but to our knowledge this is its very first application for industrial design and mechanical engineering,” Haley says.

Design Graph algorithms extract large amounts of 3D design data from an engineering company’s designs. It then creates a catalog by categorizing each component and design using a classification and relationship system. Designers and engineers search across all of their files for a part type, such as a bolt or a bike seat, with the tool returning dozens or hundreds of pertinent options.

So how does machine learning come into play?

The system teaches computers to identify and understand designs based on their inherent characteristics–their shape and structure–rather than by tags or metadata, Haley says.

After all, whoever designed the part originally could label it any of dozens of ways, using full words or abbreviations. Metadata created by people, unless carefully managed, tends to be unreliable, Haley says. With Design Graph, the computer uses its own observations about the 3D geometry contained in every 3D model.

So while some AI capabilities already exist within CAD systems, look for more to come. After all, product design plays a key role in the connected factory and the IoT systems of the future. Without design, there’d be no need for a factory—no matter how connected–to make the products and nothing for IoT to monitor.

Autodesk
Autodesk.com

Nvidia
Nvidia.com

PTC
Ptc.com

Onshape
Onshape.com