Digital Twins: Debunking the Myths

This post originally appeared on LinkedIn here.

Having been at GE Digital now for almost two years, I’ve learned that there are a fair number of misunderstandings about the industrial IoT space.  I want to address two myths that I hear most often:

Myth #1: Companies can optimize their assets by hiring data scientists and equipping them with an analytics platform; no industrial domain knowledge necessary

Myth #2: Digital Twins are OEM-specific and only work well if produced by the OEM

To help clear up these misunderstandings, I think it helps to share a couple examples of tools and practices we deploy at GE Digital that help our customers advance their IIoT journey, regardless of the types of systems or assets they may be running.

Understanding the value of Digital Twins

There are many definitions of a Digital Twin in the industry. My simple definition is that a “Digital Twin is a software representation of a physical thing or system”, which admittedly leaves room for interpretation. Hence, let me focus on what makes a good Digital Twin for industry:

• Understand the past: It tracks historical context and performance data. This is useful for correlating variables or running machine learning algorithms

• See present conditions: It is regularly updated with sensor data, often through IIoT connectivity to detect anomalies and improve model accuracy. 

• Predict the future: It synthesizes and contextualizes that historical and real-time data to give insights into potential future states (e.g., whether an asset is heading towards a failure along with the consequences of that failure and mitigating activities)

One industry analyst indicated that there are over 500 Industrial IoT analytics / platform vendors now in the market. Looking at some of their offers, most can deliver some level of capabilities in (1) and (2) and will claim (3), but I would argue not at the level necessary to be reliable or usable for industry.

 Why is that? While most vendors have data models and analytics, they lack the domain knowledge to understand:

• What are the failure modes (how does an asset or system break or degrade in performance)
• What are the consequences (how much does it matter if this asset fails)
• What are the predictive indicators (so you’re monitoring the right things)
• What are the mitigating recommendations (to prevent it from failing)

To give one example, GE Digital has a transformer Digital Twin template (reference design) that includes:

• 24 failure modes
• 55 consequences
• 60 indicators
• 45 recommendations

Developing a rich template can take months and requires domain knowledge about the asset or process – domain knowledge about both the physical properties of the assets as well as the typical usage or operating patterns. This industrial knowledge comes from both GE subject matter experts and in working with customers who will typically have a wide variety of asset types from different OEMs. We have a catalogue of over 300 Digital Twin templates from both GE and non-GE manufacturers.

My advice to industrial companies is to investigate whether your vendors have this domain knowledge already codified within their Digital Twin templates. If not, you’ll either (A) get marginally interesting KPIs and correlations rather than compelling, actionable, insights, or (B) have to allocate time and project resources to train your vendor to become a subject matter expert.

Supporting companies using multiple equipment types

As a mentioned above, practically every company purchases equipment from multiple vendors to maintain healthy competition and push those vendors for the best capabilities at the best price. In the industrial sector, companies will also diversify their equipment to prevent total system failure. For example, utilities will source hardware from multiple vendors so that if there is a design fault within a particular equipment model, that failure doesn’t take the entire network down.

GE Digital serves thousands of industrial companies and they need our software to work across heterogenous environments. So naturally, we supported our customers and developed Digital Twin templates so that our software works on their equipment, whether manufactured by GE, Siemens, Pratt & Whitney, Caterpillar, or other OEMs.

One of the biggest myths out there is that GE Digital Twin templates are only for GE equipment. To correct that, let me share a few facts on our Digital Twin templates:

• 59% are OEM Agnostic (e.g., Electric Drive Motor)
• 24% are for other OEMs’ equipment (e.g., Siemens Combustion Turbine)
• 17% are for GE equipment (e.g., GE Wind Turbine Generator)

So only 17% of GE Digital’s templates are explicitly for GE equipment – fewer than the number of templates available for other OEM’s equipment.

Additionally, GE Digital has reliability engineering veterans who have worked for decades in mining, O&G, Power, chemicals and other industrial firms and have made the transformation from break-fix and time-based maintenance to predictive maintenance. They help our customers with subject matter expertise and, perhaps more importantly, credibility when undertaking a major change to work process and culture. Of our customers’ assets that this team works with, only 15% represent GE hardware. Our Digital Twin templates are as diverse and robust as even our most complex customers’ assets or systems.

Hopefully this perspective and these examples clarify the two most popular misperceptions that I see in the industry. The deep domain knowledge of assets and operations should be embedded in the content of all industrial Digital Twins, regardless of asset origin, which is an important consideration whenever thinking about how twins could improve your operations or management regimes. I’d welcome your feedback and comments.