AI, Cloud, and the Metaverse are fundamentally disrupting industrial engineering practice
Posted by Zeeshan Zia, PhD
The Toyota Production System introduced the notion of “lean” to manufacturing, and soon the success of lean principles encouraged several other disciplines to explore their application, from software engineering to entrepreneurship, and from healthcare to government. Amongst these disciplines, Software Engineering has been particularly successful in implementing and advancing lean methodologies. In particular the ideas of abstracting away details in software modules to replicate functionality across projects, and tools for automatic optimization of code have helped the software industry advance so fast that it has become the driving force behind innovations across far flung disciplines, leading some to proclaim that “Software is eating the world”.
The next generation of improvements to lean methodology are borrowing back ideas from software.
Borrowing two key ideas from Lean Software Engineering
Continuous Process Improvement requires continuous experimentation, which is expensive. The Toyota Production System discovered a shortcut through this process of experimentation by leveraging insights from workers.
Yet, Kaizen is essentially accomplished by brute force physical experimentation – trying different ways of configuring a process on the floor – and using time and motion studies to pick the best experimental configuration. Active worker participation accelerates the process, but it may still take years to discover the optimal process parameters. In the meanwhile, resources are being wasted on a sub-optimal process. A new factory operated by the world leading manufacturer may operate at first time yields of 60% in its first year, and take 2-3 years to hit 90%, assuming requirements e.g. takt times stay constant. This eats away margins in an already low margin business.
Secondly, lean methodology doesn’t provide a recipe to transfer discovered improvements across processes within even the same facility, much less across far away factories of a parent organization.
Lean teams typically document Kaizen events in Excel sheets. However, Excel is an inefficient representation which isn’t well suited to improve other processes with similar components. As a result, process knowledge transfer happens at the slow speed of human communication and collaboration. Corporate industrial engineering teams organize conferences to share learnings, which slowly permeate through the organization over a time frame of months.
In contrast, software engineering took lean to a different level, by solving both of these challenges.
Intelligent Code Completion tools proactively suggest better programming practices to engineers in real-time as they are writing code. Compiler tools are able to autonomously simulate alternative arrangements of instructions in a computer program to obtain highly optimized versions of those programs, without requiring engineers to implement those alternative programs, i.e. without running manual experiments. These tools makes the process of software optimization extremely efficient.
At the same time, the practice of abstracting away details in software modules enables sharing of functionality across widely disparate programs. For instance, a once optimized software module to sort a list of numbers into ascending or descending order can be copy/pasted across several programs that need such sorting functionality.
These two ideas can now be applied to the world of manufacturing thanks to advances in AI, the metaverse i.e. physics-based 3D discrete event simulation, and, massive computational, storage, and collaboration resources provided by Cloud computing.
“On-the-fly” digital twins for humans and work cells
The first step towards enabling computers to autonomously simulate Kaizen events and to propagate knowledge across processes is to bring human activities and workstation cells into the digital realm.
Mature technologies such as Xbox Kinect already allow capturing 3D human poses in digital form, however, it is the “semantic” interaction of humans with equipment, tools, and products that really needs to be digitized to establish commonalities across processes.
Our computer vision solutions focused on human activities achieve exactly these objectives, in real-time, as illustrated in the following video.
Similarly, attempts at building digital twins for work cells and equipment have been limited to a handful of sensors that provide limited insights into their operation. These tools look at a complicated assembly process through a tiny peeping hole, which has limited value.
Recent advances in artificial intelligence and machine vision are enabling physical spaces to be scanned in high-resolution 3D using commodity cameras and computers several times each second. This capability combined with other sensors radically increases the coverage of a “spatial” digital twin.
We link a few academic examples in the following, that are maturing rapidly.
Together, the digital twin of human activities and work cells will provide a universally grounded representation upon which automatic comparisons between processes can be performed, as well as counterfactual hypotheses can be generated and tested in physics based simulation environments.
High-fidelity 3D representation of an entire organization in the Cloud
These next generation lean systems will host work cells, assembly and product variations, and process improvement events across a manufacturer’s factories on the Cloud.
Since the bottleneck of representing processes and Kaizen events in Excel sheets has been solved, it now becomes possible for software to automatically perform comparison across processes within an organization, and depending upon permissions, across organizations.
A key strength of the software industry has been Open Source Software (OSS), which allows individuals and organizations to share software freely, and radically improve everyone’s software systems.
Such universal representations living in the cloud will inject software-like agility into continuous process improvement, where prescriptive analytics software will be able to propose improvements to a process, based on Kaizen events observed on other processes.
However, faster propagation of process improvement is not the only benefit of such high-fidelity 3D digitization.
Simulating counterfactuals in the Metaverse
Discrete-event simulation technology leverages computer game-like 3D simulations to answer counterfactual questions about processes.
Together, the digital twin of human activities and work cells will provide a universally grounded representation upon which automatic comparisons between processes can be performed, as well as counterfactual hypotheses can be generated and tested in physics based simulation environments.
High-fidelity 3D representation of an entire organization in the Cloud
These next generation lean systems will host work cells, assembly and product variations, and process improvement events across a manufacturer’s factories on the Cloud.
Since the bottleneck of representing processes and Kaizen events in Excel sheets has been solved, it now becomes possible for software to automatically perform comparison across processes within an organization, and depending upon permissions, across organizations.
A key strength of the software industry has been Open Source Software (OSS), which allows individuals and organizations to share software freely, and radically improve everyone’s software systems.
Such universal representations living in the cloud will inject software-like agility into continuous process improvement, where prescriptive analytics software will be able to propose improvements to a process, based on Kaizen events observed on other processes.
However, faster propagation of process improvement is not the only benefit of such high-fidelity 3D digitization.
Simulating counterfactuals in the Metaverse
Discrete-event simulation technology leverages computer game-like 3D simulations to answer counterfactual questions about processes.
Here’s a video from a 3rd party vendor:
So why doesn’t everybody use such simulators to perform continuous process improvement experiments?
There are two reasons:
1. They need professionals with knowledge of both mechanical engineering and computer programming to build these simulations. Such professionals are rare and expensive to hire. In addition, a lot of upfront investment goes into acquiring CAD models of work cells and encoding realistic worker motion and interaction in these systems.
2. Even after a process is modelled in complete fidelity, the engineer still needs to manually hypothesize improvements that get evaluated by the simulator.
The innovations we described earlier solve both these problems. Thus, we have reached the tipping point where it becomes possible to conveniently model human activities and equipment, and thus set up elaborate simulations automatically.
Secondly, once these huge Cloud-based datasets of process improvement ideas become available, simulators can autonomously generate counterfactual hypotheses to run Kaizen event experiments.
Autonomous Lean
Despite the past 120 years of innovation in industrial engineering, process engineers and workers still discover improvement opportunities by trial and error.
The ability to compare digital twins of worker activities and work cells across processes allows rapid propagation of Kaizen events throughout an organization, whereas the ability to autonomously generate and virtually test improvement hypotheses prescribes novel ways of process improvement.
These abilities will completely disrupt the way processes are designed and improved in the next few years, and usher industrial engineering into an era where it becomes possible to reach gold standard processes within days instead of years, generating hundreds of billions of dollars in value for manufacturers globally.
Huy takes our product roadmap and figures out what’s achievable with today’s AI advances, and what we need to invent to solve our customers’ pain points.
Huy led teams of scientists and engineers at NEC and invented technologies for driver assistance systems that are part of 500,000+ Honda cars. He is an inventor on more than 20 US patents, and his research papers are cited by every lab from MIT, Stanford, and CMU to Apple, Google, and Baidu.
Michael works closely with our marquis customers to understand their needs, and figures out the best solution that immediately adds dollars to their bottom line.
Michael has almost 20 years of enterprise sales experience. He has held sales leadership roles at pioneering AI and computer vision startups.
Andrey takes inventions and converts them into innovations. He leads our product engineering team to turn our scientific prototypes into robust customer-facing products.
Andrey co-founded and successfully exited a startup to immigrate to the US. He spent his next eight years building the core data science infrastructure for Amazon’s supply chain planning team, and the CV/ML infrastructure for Microsoft’s HoloLens organization.
Zeeshan focuses on delighting our customers by understanding their needs and defining product and strategy to meet and exceed these needs. He works closely with our sales leadership to grow our customer base.
Zeeshan was a scientific leader at Microsoft’s HoloLens division, where he invented AI technologies to assist frontline workers across industries. Earlier, he shipped IoT platforms and visual defect detection software for manufacturers at Siemens and MVTec GmbH.
