Skip To The Main Content

Deciphering Machine Learning Dilemmas


Friday November 30, 2018

The essence of machine learning is teaching computer programs to act like humans. Yet, as more supply chain professionals realize the value of mirroring company assets with artificial intelligence, it’s also clear that considerable value lies in teaching computer programs to act like other machines.

Advantages include quality improvements such as greater durability, extended performance and fewer defects; more efficient operations from better engineering change execution, equipment performance and less variability; service savings through greater efficiency and understanding equipment configuration in the field; faster product realization; and heightened global supply chain planning and execution. It’s an impressive list of potential supply chain benefits. Unfortunately, there are also some profound questions that have surfaced that need answering.

A recent Wired article describes the concerns of industry experts related to the long-term viability of machine learning and artificial intelligence – which author Clive Thompson notes have become the dominant ways to help computer programs sense and perceive the world around them. “For years, it seemed as though [machine] learning would only keep getting better, leading inexorably to a machine with the fluid, supple intelligence of a person,” he writes. “But some heretics argue that deep learning is hitting a wall. They say that, on its own, it’ll never produce generalized intelligence.”

Thompson goes on to suggest that, if we can’t figure out how to infuse computer programs with humanlike common sense, we will keep “bumping up against the limits” of machine learning. He says the field needs a boost in the form of rules that can help computer programs reason about the world. The core of the dilemma is the fact that humans possess a fundamental knowledge of how things work and interact with one another, combined with a unique ability to interpret and rationalize. Computers lack such inherent skills related to understanding, relativity and reasoning.

Facing the problem

Whether your organization is investing in machine learning to create algorithms that forecast product demand or to extend the life of machinery and equipment through digital twins, the issues Thompson writes about are critical. Now is the time to ensure your supply chain won’t “bump up against the limits” of machine learning.

Start by reading the recent APICS magazine cover story “Virtual Manufacturing Enhances Reality.” Author Elizabeth Rennie explains that machine learning enables manufacturers to build virtual replicas of systems, processes and products from design and development to the end of their life cycles. These digital twins model the way an asset interacts with its environment, enabling users to foresee potential outcomes. In so doing, machine learning can make it possible to not only address, but actually prevent, potential disruptions and keep supply chains as productive as possible.

“Digital twins rely on thousands of sensors, which are distributed throughout the physical manufacturing process,” she writes. “They deliver masses of cumulative measurements in a wide range of dimensions — everything from the behavior of machinery to environmental conditions in the factory itself. The information is continually communicated, aggregated and analyzed, with the ultimate goals being to optimize processes, detect current and potential physical issues, predict results, and build better products.”

The idea is that if you observe a digital twin, you’ll know exactly how its physical counterpart is functioning out in the real world – assuming the machine was taught properly in the first place, of course. In fact, Rennie echoes many of the concerns in the Wired article: “Verification and validation … are vital phases for ensuring successful application of a digital twin and increasing its reliability and acceptance. Validation requires assessing whether the digital twin serves the purpose for which it was intended; verification involves confirming if model components and functions are working properly.”

The good news is that both can be tested by comparing the outcome of the actual asset with that of its digital twin when given the same inputs. This kind of systematic approach is essential to achieving machine learning goals.

Read the full article to learn more about the right way to go about using machine learning in your supply chain. Then, check out the Innovation and Global Trends compilation on the APICS magazine website to discover even more expert content designed to advance your company and career.