For a long time, artificial intelligence was something people spoke of in the abstract: a futuristic concept that seemed more aligned with movies and novels than actual computer science. But recently, a notable shift has taken place as examples of products generating real revenue with fully implemented AI appear across industries, impacting things we expected, such as stock market predictions, and those we didn’t, like identifying piles of debris in flood areas to find people who are trapped.
As AI continues its transition from science fiction into real-life, it’s instructive to take a deeper look at specific industries, examining how they have already been impacted and where they are likely headed in the future. In a two-part series, we will explore how AI is revolutionizing manufacturing and how software is re-shaping the core IP of the industry.
Globalization: Doing More with Less
A rancorous debate is being waged across the developed world today over the pros and cons of globalization, threatening to supplant the widely held “Davos Man” belief that the march towards integration is both inherently positive and inevitable.
As the gap between the haves and have-nots has grown, the dominating argument from the anti-globalization crowd follows a tidy narrative that the migration of production from developed to emerging economies has gutted the west of manufacturing capacity and the precious jobs that accompany it.
The problem with this argument, however, is that the numbers don’t entirely agree.
Examining the US market, which in this respect is indicative of most developed economies, manufacturing production has more than doubled in the past 45 years as globalization has spread. But during the same period, the number of manufacturing jobs has decreased:
How has this happened? In a word, technology.
As more tasks are automated, productivity gains are realized and the number of humans required to manufacture a widget goes down. As we’ll explore in this post, the automotive industry has seen this dynamic play out first hand, driven at a high level by robotics but powered by several AI-enabled building blocks, raising questions about what firms need to be thinking about as they continue forward in this new era.
If we take a short look back at history, robotics began humbly in the auto industry. At first, basic functions were automated, like the pulling handles to operate presses. As time progressed, the scope of automated operations steadily grew and automotive assembly lines grew accustomed to the site of robots. But at some juncture, the trend began bumping up against a stubborn reality: machines couldn’t think like humans, limiting their ability to intelligently react to changing environmental criteria.
With the arrival of AI, however, these limits are being redefined. Fueling this change are dramatic leaps forward in some of the core building blocks of robotics: machine vision, force-sensing (i.e. touch) and motion guidance. These advances are translating into continued growth in the number of robots being deployed in the auto industry, a sector that has grown the robotics installed base 20% a year over the past 5 years.
Programming Human Senses: Vision and Touch
There is a massive difference between programming a stationary machine to do discrete tasks and empowering robots to solve complex, dynamic problems. To enable this leap, AI is drawing on advancements in visual computing.
AI does best when fed massive volumes of data. Conveniently, many of the visual problems that need to be solved on a shop floor are similar to scenarios faced during regular life.
This means the vast caches of images and video stored on servers can be leveraged by machine-learning algorithms to solve seemingly mundane, but vital, problems such as the ability to walk around a mess that’s been spilled on the floor or find a tool sitting on a counter required to complete a job.
And while robots with intelligent vision are a big step forward, they still possess limited utility if they can’t apply a delicate touch once they arrive at their destination. To solve this, robotic algorithms are employing next-generation applications of “fuzzy logic” paired with deep learning.
As a bit of background, fuzzy logic is a form of many-value logic that’s used to solve computer science problems like filling in incomplete addresses. Historically, a small disruption in one part of the production line risked a cascading set of issues. But when fuzzy logic is integrated into robotics, machines can make small alterations to compensate for evolving environmental conditions, getting things back on track.
This is certainly helpful, but to know how much of an adjustment is appropriate to a given situation, this capability has been paired with deep learning neural networks. Essentially, each time a robot faces a certain situation and makes an adjustment, the outcome is fed into a neural network.
Over time, robots learn what reaction yields the optimal result, such as precisely how much pressure to exert when working with different materials. In other words, an intelligent sense of touch to go with intelligent vision, yielding a much more effective robot.
Next Generation Production Lines: Data-Driven Optimization and Modularization
Beyond robotics, production-line modeling is also being impacted by AI. Being well acquainted with the fundamental relationship between time and money, factories have for many years put a large degree of focus on optimizing production schedules. Today, however, the explosion of available data means the number of inputs that can be incorporated in a production model has exploded.
To handle this, today’s generation of simulation companies is using learning algorithms to gather these inputs to optimize production lines while fully respecting material, machine and workforce constraints. These programs set an initial production baseline based on factory guidelines and then test millions of iterations to model different combinations of inputs (material availability, resource requirements, etc.).
What began as simple Darwinian “survival of the fittest” algorithms now mimic how Deep Blue beat Garry Kasporov in chess. By running AI programs to calculate all permutations on material availability, machine requirements and customer orders to generate the “winning” combination: increased production speed, reduced space requirements, reduced cost and increased flexibility.
While it’s certainly interesting to optimize the sequencing of a production line, innovative manufacturers have realized a traditional linear assembly line – even one that is optimized – inherently possesses waste. This problem is being exacerbated in today’s era of ever-increasing customization, where cars with optional add-ons require incremental work that slows down the entire line. The answer, which prior to AI remained frustratingly elusive, was large-scale production line modularization: allowing products to move in unique patterns and at different speeds through assembly.
As drivingline.com recently highlighted, a good example of this can be found at Audi. When the German auto leader produces their limited production plug-in hybrid A3 Sportback at their Ingolstadt factory, the specialized electrical equipment required by the hybrid A3 requires additional stations, causing other cars in the line to sit idle on a conveyer belt suspended below the ceiling.
To solve this, Audi has leveraged modern AI-powered robotics to implement modularization. Vehicles requiring custom parts are moved off the main line via AI-powered driverless forklifts and transport systems, allowing the vehicles behind them to continue down the line unabated.
Supporting Cars in the Age Of AI
As Toyota executives could tell you when news of their rogue breaks became public in 2009, problems sometimes are not discovered until after vehicles have been sold and delivered to the public. Addressing these issues via recalls is very expensive, both in direct cost and potential damage to a manufacturer’s brand.
AI is helping address this challenge in two key ways. First, by leveraging advanced data mining techniques, firms can more effectively trace problems identified in the field back to root causes. This enables automotive companies to more rapidly identify and correct issues and manage communication to the public, thus reducing the scope of the issue.
Second, and perhaps more importantly, by leveraging massive amounts of data, neural-networks can gather inputs during the production process and predict end-state quality before the product has been finished, enabling quality control alarms that can instigate corrective measures before production has been completed. As the saying goes, “a stitch in time saves nine”.
Successful Companies Will Embrace Data and Robotics, But Not Entirely at The Expense of Employees
Powered by AI-enabled advancements in vision, touch and motion control, the capabilities and reach of robots in automotive manufacturing continues to grow. Looking forward, the global robot population is projected to grow between 6% – 20% per year, depending on geography.
It is therefore tempting to draw a simplistic conclusion: “I should buy more robots”. But looking back to the Audi example, when the company leveraged AI-powered robotics to efficiently modularize their production line, they strategically kept humans involved in the process.
Touting “human-robot co-operation”, Audi designed a new set of systems to leverage the strengths of humans alongside the emergent capabilities of robots. To give a simple example, a new robotic arm that can grip and hold heavy objects is coached by a human using a touchpad interface.
Indeed, between 2010 – 2015, the US automotive industry added 135,000 new robots while concurrently creating 230,000 human jobs.
Looking to the future, innovative companies will continue to aggressively introduce robots into their production lines. But instead of taking the path of simply replacing human jobs with robots, companies will use their imaginations to re-deploy workers into higher value-add roles. These firms will look for tasks humans can do uniquely well: highly variable operations in dynamic environments that require social IQ.
Successful companies will also examine ways they are using data to optimize production, both in terms of efficiency as well as predictability. With powerful new algorithms capable of identifying manufacturing variability in near-real time, releasing defective products in the future will be viewed not only as a defect in physical production, but also as a failure in data management.
Click here to read Part 2
Kishor is a managing partner at Infosys Consulting, based out of our Plano, Texas hub. He focuses on driving operational effectiveness in organizations through digital transformation and automation. He is an industry expert in design thinking and innovation and has spoken at a number of client and industry conferences on the subject. He has also published several articles in leading journals.