IoT “Winter”: The Other Side

What is on the other side? Causal Digital Twin (CDT) that achieves Business Performance Improvement (BPI) measurable as topline revenue increase for IoT end customers. Before I define the terms, a brief look back . . .

Over five years ago, I drew the diagram below. In fact, IoT was not even in “autumn” then, let alone “winter”; there was lots of promise especially with 5G around the corner and billions upon billions of things to be connected to the Internet at the click of a button. Today in 2021, we do see some growth for IoT but nowhere near the predicted ubiquity.

Consider the arc of IoT ecosystem – I saw a kink in it. Back in 2000 when I developed and deployed Rockwell Automation’s Enwatch and Enlive (IoT and Digital Twin) in large paper, steel, automotive and other plants, I saw the initial excitement of plant managers and engineers sour in a few quarters – they loved it but their bosses, not so much. I discuss the business reasons later in this article.

My belief then and now stem from what I saw happening on the plant floor. Any condition monitoring event is invariably followed by a root-cause analysis phase where machine experts and engineers gather and try to figure out (a very cumbersome and often difficult task) why the event happened and how to fix it so that it does not repeat in the future. As the term root-“cause” implies, I knew that it had to do with “Causality” (it was all Correlation back then) but did not have the theory and tools accessible at that time. Being loath to raising alarms without suggesting a solution, I did not pursue Causality then.

Now in mid 2021, things are different – we have the theory, algorithm and potentially multiple solutions for Causality in IoT – “Causality & Counterfactuals – Role in IoT Digital Twin” (https://www.dhirubhai.net/pulse/causality-counterfactuals-role-iot-digital-twin-dr-pg-madhavan/). However, the all-important role of Root-Cause analysis and Causal Digital Twin solutions are still not fully grasped by the IT-OT community . . .

As @JoediPaolantonio has noted, “digital twin Instances and digital twin Aggregates are applied today for condition monitoring, predictive maintenance, warranty management, customer experience, and design-feedback, among others.”

These applications do not contribute to BPI *directly*. General performance improvement may happen from “condition monitoring” – perhaps the kind where there is peace of mind on the plant floor from less machinery breakdowns leading to employee positivity! That is not the kind of “performance improvement” that IoT end customer executives care about – they want *business performance improvement* (BPI) that enhances the topline revenue numbers. Anything less will get lost in the shuffle as I describe later.

There are good signs that IoT enablers (top row in Grey in the figure below) are developing fast with the help of massive interest from 5G purveyors and hyperscalers. BUT let us not confuse enablers with the prime mover! In the IoT case, it is Causal Digital Twins. There is consensus now on the definition of Digital Twin: Digital Twin is a Software Construct that is animated by real-time and historic Data of the corresponding Physical Entity. Unfortunately, this definition is so broad that much confusion has been generated by claims of almost any software service in the IoT value chain being a “digital twin”! To understand the importance of CDT, let us discuss the following expanded “definition” of Digital Twin.

Digital twin that *enhances BPI* has to perform THREE major functions: (1) Condition Monitor, (2) Root-cause Analysis and (3) Close the Loop. Each of these three functions are operationalized with “thresholding”, “simulation” and “visualization”, respectively. To execute these functions, each operationalization activity has multiple methods shown in the light Blue boxes.

(1)???Condition of the system being monitored.

Condition monitoring is well developed. There are simple as well as somewhat sophisticated methods based on ML. In general, it can flag unusual events and in some cases, predict events with certain probability.

(2)???Root Causes of alarm condition.

This is the step that is missing in the current IoT state-of-the-art. As rightly noted by Galloway, van Schalkwyk and Marian, “Causal Analytics in IIoT – AI That Knows What Causes What, and When”, Industrial Internet Consortium Journal of Innovation, 2018, “Knowing the real root causes of events is critical to resolving problems rather than continuously dealing with the symptoms”! Without knowing causes and effects among the underlying sub-system, NO performance improvement *prescriptions* are possible. Why? Correlation-based models (any ML including Deep Learning) do not tell us what caused what!

Just like plant floor specialists performed root-case analysis by doing experiments, root cause analysis using digital twins also needs experiments; but unlike the old days where the experiments were conducted on the machines themselves, here we perform the necessary experiments on SIMULATIONS instead of new data. ALL simulations are based on MODELS – the Simulate block mentions three.

·??????Causal model of the type mentioned above statistically estimates what causes what and by how much. Simulated scenarios based on Causal Models are the ONLY ones that can provide reliable prescriptions on what can be done to improve performance. They are called “counterfactual” experiments (loosely known as “what-if” analysis).

·??????Physics-based models came of age in the CAD/CAM era for product design (machinery, aircraft wings, chip manufacturing, etc.). Simplified versions (called Reduced Order Models or “ROMS”) are being repurposed as digital twins. Here causality structure is *imposed from outside* by the equations and assumptions therein as chosen by the Physicist. This has its uses when there is a dearth of data. But in IoT, that is hardly the case! Also, equations give us a false sense of precision – equations are gross approximations of reality (we still cannot solve n-body problem when n > 2!) and the assumptions made in its applications are rarely met in real life . . .

·??????We have met “If-then-else” before – they are the 80’s “expert systems” which we don’t hear about anymore but are still useful in simple cases. Pearl, Dagum, Horwitz and others wrapped Bayesian network around it to give it a probabilistic basis.

(3)???Closing the Loop.

Currently, visualization is the only way to affect the real world. Humans see the visualizations and act on the physical world. If the digital twin information has to be fed back to a mechanic on the plant floor, an AR device with visual overlays may be the most effective “action”. As the overall digital twin technology and its acceptance accelerates, human will move “above-the-loop” and digital twin will directly interact with machinery on the plant floor through control channels.

This expanded definition of Digital Twin is useful in general in that it will reduce confusion as to what can be called a digital twin. It is also clear that all functions - Condition Monitor, Root-cause Analysis and Close the Loop – may not be necessary in all digital twins. However, NOTE that Root-cause Analysis function is *essential* for digital twins used in operations that enhance BPI.

What is missing in today’s (mid 2021) digital twin is the middle box circled in dashed Red line – AUTOMATED Root-Cause Analysis. We have the theory, algorithm and potentially multiple solutions for Causality in IoT – CAUSAL Digital Twin (CDT) in hand. There is a compelling need for corporate technologists to get involved in multiple PoCs and generate new CDT methods and algorithms.

BPI – Business Performance Improvement:

The value proposition that the CURRENT IoT offers to END customers who are large corporations that run industrial plans and manufacturing machinery (and other IoT applications) is just not that compelling . . .?Condition Monitoring’s business value prop is Preventive Maintenance. Preventive Maintenance don't show up on end customer CFO's financial statement (it is buried in an Opex line-item!). Also, the dollar savings from Preventive Maintenance (a machine failure that did NOT happen!) is nearly impossible to quantify – customer CxOs cannot visualize the value that IoT brings!

CDT transforms the underlying production system into a "mechanism" – a system of causally interacting parts that produce effects. On a plant floor with interconnected machinery, engineers can figure out how to improve system performance by counterfactual experiments on CDT simulations using already collected data (super-important since you won’t be able to perform new “experiments” on production floors running at full capacity!). They can figure out how to increase production volume (Gross Revenue), reduce waste (reduce COGs) and increase quality (higher per-unit Price). ALL three go to make CFO's Gross Margin higher - that is when IoT deployment will surge.

BPI enhancement with CDT is possible in every IoT application. Take Smart City for example: Increased smooth traffic flow (“production volume”), reduced HVAC power consumption across city high-rises (“reduce waste”) and improved air quality by reducing CO2 emission (“increase quality”).

Quoting @JoediPaolantonio again, “Multichannel IoT Causal (MIC) digital twins (Instances and Aggregates) will be the “killer app” for #iot #iiot in continual strategy, process and product improvement that will bring unprecedented ROI to all types of organizations — industrial, healthcare, educational and governmental.”

Causal digital twin elevates the value proposition of IoT from preventive maintenance to business performance improvement. When that happens, IoT will flourish!

Acknowledgement: I am grateful to @JoediPaolantonio for his crisp and concise comments which I have included in this article.

Dr. PG Madhavan

https://www.dhirubhai.net/in/pgmad/

#IoT #Digitaltwin #Causality #Causaldigitaltwin #Simulation #grossmargin #iiot #smartcity?