Methods and Systems for an Autonomous Supply Chain

A Robotic Internet of “Maker Things”

  • This project is a research undertaking that seeks to develop a way to network robots & robotic manufacturing sites and coordinate their interaction via a machine learning system in such a way as to significantly increase productivity — much like the internet has “exponentiated” the dissemination and production of information.
  • We might think of other names such as “Distributed Agent Manufacturing” or an “Internet of Maker Things.
  • Broadly speaking the project seeks to implement the vision described here.
  • There are other, existing efforts in this space such as the German government Industrie 4.0. There is also Australia’s CSIRO Advanced Manufacturing Roadmap. Industrie 4.0 and Advanced Manufacturing are umbrella concepts seeking to incrementally enhance supply chains and their processes using digital technologies. Our project does not aim to enhance existing supply chains, it aims to replace them. We might imagine replacing the existing fabric of tier-1, tier-2 and tier-n manufacturers with an intelligent network and just tier-n manufacturers. Towards this far reaching vision, we have designed a theoretical, mathematical framework to facilitate the scalable interaction of manufacturing agents. This framework is patent pending in the United States, the United Kingdom, Australia and internationally.
  • At present, we have a set of software and hardware prototypes, accompanied by simulations that implement parts of this framework.

Start with Why?

  • Computers were useful on their own. They only truly became disruptive when networked.
  • We aim to apply the same principles to robotics & in consequence to manufacturing.

So what will this do?

  • The end goal is to be able to ask:  I have a thing with the following specification. Who and what can make that for me? What is the cost of the job? How long will it take? Can the job be scheduled given available capacities? What is most efficient way to complete the job given a set of constraints? Now do it: Execute, monitor, deliver & feedback!
  • It will do this by providing the framework in the “cloud” that ordinarily big business provides to its business units or that tier-1 manufacturers provide to tier-2 manufacturers.
  • Central to our framework will be standardising the interfaces between workflows.
  • This standardisation will create the platform for composition of  function and in doing so boost productivity.

 

  • It seeks to create jobs in tertiary industry to replace jobs lost in conventional manufacturing:

  • Tertiary sector jobs created will range from builders to mechatronics specialists.

  • It seeks to enable “per product” customization of output:

  • For the first time since the industrial revolution, customers will be able to customize at no extra cost.
  • This follows from the autonomous supply chain being dynamic – like the internet – whereas traditional supply chains are static.
  • It seeks to re-empower artisan labour:

  • For the first time since the industrial revolution, artisans will be on-boarded again in an economy of scale.
  • This follows from the standardization of interfaces between workflows that crucially does not distinguish between robotic participants and traditional participants.
  • Mass manufacture & “handmade” can be mixed and matched.

Manufacturing through History

Before elaborating, it will help to consider the path manufacturing has taken through human history:

Slide 1: Background – Manufacturing Through History

 

Briefly, the artisan age was marked by high human labour input and generally lacked economies of scale. Bespoke was the norm, because standardisation was impossible. The industrial age changed all that. During the American Civil War, the Northern States were able to win the war in large part because they managed to industrialise the manufacture of rifles which proved superior over Southern smooth bore muskets. The age of the automobile was enabled by standardised and therefore interchangeable parts. We attained economies of scale, yet suddenly “handmade” became special. The machines that facilitated all this were dumb but required skilled labour to operate them.  What propelled the industrial societies of Britain, Germany, the United States and Japan to the forefront of history and left behind the non-adopters: Democracy walked in lock-step with an educated middle class that provided this skilled labour. The information age altered this equation by enabling smart, but crucially non-autonomous machines to be operated by unskilled operators. This heralded a tectonic shift to unskilled labour, outsourcing and enabled nations that had hitherto been “non-adopters” to catch up. The cost savings of this shift also enabled global supply chains: For instance, it became cheaper to assemble end-user goods ,such as a razor, by shipping blades made in Germany half way around the world and combining them with handles made in China for sale in stores in the United States – all by way of super sized container ships around the world’s oceans.

Yet everything is due to change profoundly once more. So far we have extended supply chains around the world and accelerated the flow of goods along these supply chains in line with cost equations – but crucially we have largely persisted with push strategy supply chains. As we move from smart but non-autonomous machines that rely on human operators to smart machines that operate autonomously (robotics), the optima presently found in push strategy supply chains will shift in favour of optima found in pull strategy supply chains. This transition will prove to be more profound than all the transitions that have gone before. There is precedent to support this assertion and it is to be found in how the Internet of today has transformed how modern society processes information. This precedent proved to be nothing short of a revolution.

Precedent: Information Revolution

Slide 2: Precedent: Information Revolution

The information revolution has proven so profound that reports indicate that 90% of the world’s data has been generated in the last two years (as of 2016).  It was not to be the invention of the printing press and publishing houses coupled with road and rail networks that achieved this. It was to be the Internet. What is interesting is to consider how we might apply the tenets of the information revolution to manufacturing and supply chains.

Applying Information Revolution Tenets to Supply Chains…

Slide 3: Applying Lessons from the Information Revolution

In thinking about how to apply the tenets of the information revolution to supply chains, we ask ourselves the following question:

What made the internet scalable?

Slide 4: What made the Internet scalable?

What most users of the modern internet don’t know when they herald a taxi on their smart phone or order swimwear off Ebay is that what seems to scale effortlessly to billions of users around the world, does so on the shoulders of an architecture that has been deliberately designed to effect the vastly distributed network we know today. At the core of the internet lies a theoretical model that permits a “telecommunication fabric” to accommodate many, many interacting & distributed agents.

What our research has focused on is to elaborate a parallel model for manufacturing supply chains. We have also defined an ontology and associated reasoning systems that not only allow individual agents to be autonomous but crucially seeks to make the network itself autonomous.  There is a certain meta reasoning aspect to the latter. Readers of this blog will be familiar with my love for the higher order logic that lies beneath.

To re-iterate: the end goal is to be able to ask the network: I have a thing with the following specification. Who and what can make that for me? What is the cost of the job? How long will it take? Can the job be scheduled given available capacities? What is most efficient way to complete the job given a set of constraints? Now do it: Execute, monitor, deliver & feedback!

Patents

Slide 5: Patent Protection

The first patent defines the Supply Chain Interconnection Model. The second patent defines the Robotic Capability Model.

The Supply Chain Interconnection Model

The centre piece of our research defines the Supply Chain Interconnection Model as shown in the illustration below:

Slide 6: Supply Chain Interconnection Model (SCIM)

We call this Supply Chain Interconnection Model, or SCIM. Unlike the OSI model, SCIM is defined using a rigorous mathematical model that permits us to reason about its properties.

Slide 7: CSP Model, an Excerpt

Implementation is by way of CSP or Communicating Sequential Processes, a model used for “describing patterns of interaction in concurrent systems.” CSP was authored by the Oxford scholar Tony Hoare also noted for the invention of the Computer Science sorting algorithm Quicksort. What CSP allows us to do is to formulate assertions that can be verified through a  proof checker.

Guaranteed Correctness…

Slide 8: Machine Checked Proofs on Properties of the Model

CSP let us prove properties of our system – even before writing it.

Productivity Multiplier Tenets

Slide 9: Supply Chain Productivity Multiplier Tenets

Multiplier tenets are design goals of the system. Realising the above tenets leads to the Robotic Capability Model, or RCM. The RCM combines a robotic ontology and a manufacturing ontology with an execution plan engine capable both of discrete, qualitative reasoning as well as quantitative reasoning across both ontologies.

  • The “robotic ontology” describes the domain of capabilities – robotic and human alike. This includes constraints upon capabilities.
  • The “manufacturing ontology” describes the domain of product specification. Likewise, this includes constraints but upon products or the manner of their manufacture.
  • The job of the Robotic Capability Model is to manage capabilities. It matches product specifications, or requirements, to capabilities that can fulfil those requirements.

Artificial Intelligence Driven Manufacturing Plan Engine

Slide 10: Robotic Capability Model

The concept of a manufacturing plan engine leans on how database systems translate declarative queries into optimised execution plans, but applies this concept to manufacturing while applying machine learning to solve “fuzzy” optimisation problems.  If that sounds theoretical, this article shows how reasoning systems of this type work.

Slide 11: Robotic Capability Model

The robotic capability model is capable of actual reasoning. It can explain why something is or isn’t the case. The slide below shows querying the system as to whether  a particular part, here a box, is a composite. The system is able to deduce TRUE and presents proof based on modus ponens and universal quantification.

Slide 12: RCM Reasoning Capability Demonstrator

Besides being able to reason qualitatively, the Robotic Capability Model can reason quantitatively based on neural network models. This is imperative when dealing with incomplete data, inferred relationships and significantly unknown features. Please refer to the section “Marriage between Discrete and Continuous Logic” in the Article “The Reasoned Lisper.” One example is shown below:

Machine Learning Based Feature Prediction

In the above example we hypothesise that an as yet unknown part might be constructed from known components A,B & C. Firstly we assert that A,B & C are indeed parts of a whole, then we assert that all three are parts of an as yet unknown part “unknownPart.”  Finally, we are able to obtain an estimate of the “unknownParts’s” MBF ( meantime between failure ). Note that this is not merely a trivial average of constituent meantimes between failure but rather exploits any number of possible complex features and interrelationships, such as for example knowledge of how componentA behaves when it is part of something rather than used on its own or how it behaves when it is part of something alongside specific other parts. Managing complexity in this way becomes possible in a world where Big Data enables machine learning concepts in like never before. On their own, many machine learning models are reducible to statistical regression or much simpler analytical models. This perhaps more than anything explains why AI of the 1980s did not attain its goals. The explosion of big data will alter this status quo. Managing complexity as shown above through machine learning is not only becoming possible, increasingly, the human analyst will be unable to compete in this domain. It is unsurprising that while only now mainstream technology companies like Amazon, Google and Uber are waking up to this trend, the United States Department of Defence has been a practitioner for many years.

Putting it all together… Supply Chains Now

Slide 13: Supply Chain Now


The above slide illustrates the traditional supply chain model. Push strategy takes product components along an extended line of supply across continents and to market. All products are identical to warrant the efficiency of the model. Changes to the supply chain are slow. We note that material is brought to process much like data in a conventional computation model is brought to the computer, or the CPU.

Supply Chains Then…Reshoring

 

Slide 14: Supply Chain Then – Reshoring

The above slide illustrates our revised supply chain model. Like the internet, interaction is broken down according to “back bone” transport and local clusters. Each cluster produces bespoke variants of the product. Manufacturing takes place locally in “cells.” Feedback within each cluster or cell is rapid, facilitating both agility in general as well as pull strategy manufacturing. This concept is at the heart of efficiency gains of our model. We also note that processes are brought to the material, much like the key tent of “Big Data” where function is moved to data to warrant extreme scale. Carbon footprints are lowered because material transport is minimised.

Possible New Industries

Slide 15: Possible New Industries

Local Jobs Created

Slide 16: Possible New Jobs

What comes around, goes…

As suggested earlier, Defence has contributed greatly to this domain. It is unsurprising that the concepts elaborated here likewise find potential application on the realm of Defence. What civilians call a supply chain, defence planners call a “line of supply.” In times of conflict, lines of supply are a determining factor in simple considerations such as the advantage a defender enjoys over an aggressor. See also Carl von Clausewitz’s classic work  “On War” (Vom Kriege): “Modern perception of war are based on the concepts Clausewitz put forth in On War, though these have been very diversely interpreted by various leaders (e.g., MoltkeVladimir LeninDwight D. EisenhowerMao Zedong, etc.), thinkers, armies, and peoples. Modern military doctrine, organization, and norms are all based on Napoleonic premises, even to this day”  Yet Clausewitz’s grasp on tomorrow is about to slip – as is evident in the British ARMY’s “Autonomous Last Mile Resupply” project embedded in the British ARMY Strike Concept.

Slide 17: Autonomous Resupply on the Battlefield

To date the British ARMY Strike Concept – Autonomous Resupply focuses largely on what our model terms the “Transport Layer.”

Larger benefits are to be realized from the model as a whole.  Coming Soon: Refactoring Classic Clausewitz Military Strategy: An Essay. 

What is next?

I am presently researching both software simulations of networked manufacturing capabilities as well as producing actual, usable wares based on my model which will act as demonstrators. See also Autonomous Supply Chain – A  Simulation. The slide below demonstrates geospatial capability filtering for the state of Tasmania. The simulation is created using the Unity Game Engine.

Slide 18: Tasmania & Geospatial Capability Filtering

Apologetics

I call this section “apologetics” rather than “references.” References are often furnished in papers as supporting arguments, but also suggest that the author was referring to his or her sources  “a priori.” Realising that what we propose is both unusual and far reaching, we seek corroboration but would like to avoid any suggestion of an “a priori” role of cited sources. Specifically, the publication dates of all sources are preceded by our patent priority dates and in some cases are preceded by communication with relevant government departments.

United States Army “Robots in Manufacturing Environments” 

From the Notice of Intent…

Over the last few decades, there have been numerous advancements in the use of robots in manufacturing, mainly in the automotive and electronics sectors. However, the manufacturing use of robotics has traditionally been restricted to one repetitive task…This technology has the potential to level the manufacturing playing field… 

This agrees with our own assessment that robotics is useful in isolation, yet disruptive only through collaboration — or networking — identical themes. Also note the shared theme of integration of “handmade” or “with humans.” The potential to level the manufacturing playing field was first cited by a representative of the Australian government. 

3dprintingindustry.com/…3d-printing-affect-supply-chain

This article gives the following insights:

The concept of mass producing goods half way around the world and then shipping them is inherently inefficient.

Now we face a new industrial revolution that brings its own unique problems. We know that there will be upheaval, but still nobody quite knows…

A flexible supply chain:

The way a product is produced will depend upon the location of the customer and the complexity of the task at hand. So the supply chain will not be rigid, it will be a fluid and ever changing organism depending on the specific variables that swiftly account for labor, transport and production costs at each separate facility…

Production will shift to an on-demand basis.

The above echoes all the tenets that we have proposed. Crucially, the article does not formulate a solution, but it does validate our insights.

Stay tuned for more…

 

Advertisements