4/3/2019 | 5 MINUTE READ

AI Advances into Production

Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

Adopting new technology in design and manufacturing is usually a series of baby steps. Artificial intelligence and machine learning are no different.

Share

Facebook Share Icon LinkedIn Share Icon Twitter Share Icon Share by EMail icon Print Icon

Artificial intelligence (AI) and machine learning (ML) are still so very new to manufacturing, despite the recent buzz about them—on top of the now-ancient buzz over expert systems in the 1980s and 1990s. Two recent announcements show again the value of AI and ML in complex products like vehicles. One announcement, from Siemens PLM Software, is about making the user interface (UI) to computer-aided everything (CAx) easier; in this case, NX (plm.automation.siemens.com/global/en/products/nx). The other announcement, from FACTON Inc. (facton.com/en-us), focuses on product costing.

The adaptive interface

For years, computer users have been dealing with the ALT-keys, mouse clicks, pull-down menus, and ribbons that comprise static, data-driven UI. None of these approaches have been particularly pleasant (read: simple, fast, or efficient). True, software vendors have mitigated the pain somewhat by letting users manually customize the UI to some extent, and by offering preset, out-of-the-box roles.

 UI tweaking has a couple of problems, though. First, it takes time. Second, users have to maintain the UI as work habits and workflows change. Says Paul Brown, senior marketing director for Product Engineering Software Group at Siemens PLM Software, the UI “is really only based on a point in time. A week later, life evolves, and the user wants something different. No matter what we do with roles and try to guess what you would want to use based on”—and Brown lists a slew of criteria: experience, knowledge, what the user has done in the past, what software systems the user has worked on in the past—“you cannot predict user behavior. What you really want to do is concentrate on solving your design problem, not worry about the UI.” What’s needed is a dynamic interface that adapts to all users.

The latest release of NX software does exactly that. AI and ML capabilities embedded in NX monitor how the user works with the software, predicts the user’s next steps in the workflow at hand, and updates the UI accordingly. For example, a designer might start with some 2D curves and then extrude, revolve, even sweep them to create 3D objects and surfaces. The adaptive UI will revamp the following set of icons in the UI based on what the designer has chosen to do. The displayed icons are not just the last 10 commands used. Instead, NX learns the steps in a particular workflow. This becomes apparent when the user changes the type of parts being work upon. For instance, the workflow for designing a casting is different than designing a piece of sheet metal, explains Brown. “If I draw a profile while designing a casting, I probably would use an extrude or a revolve to get a solid body. If I’m doing sheet metal, probably the next thing I would do is create a bend. NX understands this context.”

This all happens dynamically; the UI is constantly being refreshed and reevaluated. Granted, some commands are persistently used, so their icons never seem to be refreshed. However, there are other commands, other icons, that come and go based on the task at hand.

Along with eliminating the user’s hassle of customizing and maintaining the UI, the adaptive UI helps the user efficiently work with the software, thereby increasing user productivity. Plus, given that most engineers use a small subset of the entire functionality available in NX, the AI and ML capabilities also “automatically adapt the user interface to the needs of different types of users across multiple departments, can resulting in higher adoption rates,” says Brown. Moreover, “organizations can share their best practices generated from experienced users to new users.” Not so incidentally, the adaptive UI can also smooth migrations from third-party software products to NX.

Tightening cost savings

“AI is like the next frontier for us,” says Alexander Swoboda, CEO of FACTON GmbH. FACTON EPC Suite (“EPC” being enterprise product costing) helps companies quickly and accurately assess the effects of their cost decisions. The software system, available both cloud and on-premise, provides an overview of costs at every stage of product development—from initial idea through production—so that companies can evaluate alternative designs, materials, locations, manufacturing processes, and suppliers.

A stumbling block is that cost accounting is not standardized, while financial accounting is (e.g., financial accounting is covered by U.S. Generally Accepted Accounting Principles, U.S. GAAP; German GAAP; International Financial Reporting Standard, IFRS; and so on.) Costing data can be treated by each company and each division within that company in very different ways. For example, contribution margin can have different meanings, or the definitions of direct versus indirect costs can contain different elements. “People can call different things with the same names,” says Swoboda. Plus, many large companies still use Microsoft Excel to calculate product costing. “It’s hardly possible to ever standardize what they do,” adds Swoboda.

Implementing EPC Suite helps standardize product costing so that, even within companies, people can compare apples with apples—or the price of a car seat assembly from one supplier with that from another. EPC Suite also helps standardize the data structure and attributes required for product costing. Without this, says Swoboda, companies “can’t even use their own internal data to train an AI system.”

About AI Swoboda says, “I deeply believe that once you have data, you can move to artificial intelligence.” He ticks off three applications of AI in product costing. First, “an automated cost estimate for the early stage of product development based on a similarity matrix using historical customer data,” In other words, an AI engine for estimating initial values. “When you have a certain component you want to estimate, most likely it’s similar to components you have worked with in the past. Depending on the size and certain attributes, you can then figure out what the costs are.” FACTON prototyped an AI engine that, given certain input parameters, such as material thickness and product size, provided a good starting point for estimating a product’s cost.

Interestingly, FACTON didn’t build the AI engine. “We are a Microsoft shop; our whole system is on .NET,” says Swoboda. “Within Microsoft Azure”—a cloud service—“there are certain AI algorithms you can incorporate into your product.” In this way, by using existing AI algorithms, Swoboda can “add AI services to different parts of the system over time.”

A second application for AI in product costing is in “outlier analysis based on customer data that enables businesses to highly standardize cost estimates.” Here, the system finds anomalies in the data and flags them so they can be analyzed.

Third is an “automated costing of parts using algorithms—building on features of computer-aided design models and components that have already been manufactured.”

 There’s a fourth application, which Swoboda admits is “fancy”: linking costing information to recall and insurance information. Through AI, automotive OEMs and their suppliers can discover when certain parts get, for example, too cheap. “I think that’s intellectually very challenging,” Swoboda says. “People really want to use the system to automate the easy parts of product costing.”

Fancy or not, here comes AI and ML.

 

 


Resources