Predicting the failure of assets is the holy grail of maintenance. And, there has never been a better time to achieve it than now. At stake is millions of dollars in savings through uptime improvements and downtime avoidance. Two things that are critical for investments in capital costing huge sums of money and revenue generation.

Over the past few years, technology advancements in the Internet of Things (IoT) sensors, analytics and simulation have emerged as possible panaceas to solve the puzzle of predicting failures with increased accuracy. These technologies take information generated from individual assets or systems over time and couple them with complex algorithms to predict failures.

The good news is there have been some promising results, such as reducing equipment downtime by five percent by predicting a valve failure. However, such stories are mostly about pilot projects that haven’t yet moved to full deployment with enterprise level monitoring, analysis and maintenance execution.

More recently, there have been increasing references to digital twins: the idea of creating an exact replica of a physical asset by combining computer-aided design (CAD) and simulation models, IoT sensors, time series data and maintenance records to build a picture of an asset and its current operating condition.

Now the hype has reached its peak. It’s time for organizations to take a step back and understand, in depth, what these technologies are capable of achieving. For example, is the approach you are taking with these new capabilities really providing the picture of asset health you require to make important decisions on proactive maintenance activities?

In reality, the top-down approach that has taken shape in the market will fail. Why? Organizations are addressing the underlying problems: what is the asset I am focusing on in the first place, what is the history, does it provide the ability to follow information easily, what is its current makeup and how is this different than the rest of the assets I need to manage?

A Digital Picture, but Not the Whole Story

In recent years, IoT sensors have emerged as a powerful tool, monitoring things like torque, temperature, corrosion, and start and stops, to name a few. The information from these sensors is then coupled with other historical data sources and predictive analytics to provide a picture of an asset’s health and forecast when components might fail. When you look at an asset in isolation, which is the definition of a pilot project, you will likely see some good results. However, the algorithms are based on one individual asset and its related sensor data.

The problem is when you try to scale hundreds or thousands of similar assets. The predictions you created are specific to the asset in the original pilot project. Applying these same predictions to other assets could lead to maintenance issues, as parts with useful life remaining are replaced or assets taken out of service due to an unforecasted failure. Why? Assets are not manufactured equally and over their time in operation, in some cases 10 to 40 years, they diverge further, even if sitting right next to each other.

For example, two similar assets might have different electric motors, each from a different manufacturer. One of the manufacturer’s components may be designed to last longer. Is your prediction based on that? Now, factor in the many other ways in which your assets differ. Will the predictions reflect the differences in thousands of similar, yet different assets?

Digital Models Will Never be Digital Twins

There is an emerging trend to use simulation models created during the engineering phase of the product lifecycle as the digital twin of an asset. The concept is that comparing these digital models with operational data may result in the identification of failures while running the many different simulations. After all, the simulation models are tested for many types of possible operational scenarios and the related failures that would occur if they persisted.

At first look, this is a vast, rich resource of information that can be used by maintenance to monitor the signals for failure in the field. The problem is that these simulation models may not necessarily reflect the final as-built configuration of the asset that went to the customer. As assets go hrough manufacturing, much can change. For example, as suppliers change, modifications are incorporated and defects are rectified. The original simulation models will not reflect these differences and the actual performance profile will differ from that predicted. Fast forward to the asset operating in the field that, over a few years, has undergone maintenance and upgrades to the point that its configuration is now significantly different from that assumed in the original simulations.

Flip the Process on Its Head – Make Your Digital Twin Built to Last

What is the maintenance group to do? Promising technologies are available; IoT data, predictive analytics and simulations all have value, but only when used in context. This means building and maintaining a digital record of the configuration of products as they are manufactured, maintained and upgraded. This is the key to keeping the asset in the field and its digital twin synchronized.

This first viable, contextual digital twin is created during the as-built phase of manufacturing. This is the first view into the exact makeup of the asset in context. This involves recording the exact product configuration, including any special features or options used, as well as capturing serial numbers. The digital twin is subsequently updated whenever a significant change happens to the asset. For example, if electric motor serial number #001 is replaced with electric motor serial number #002, the corresponding change is made to the digital twin.

Now, using the digital twin configuration, simulation models can be built specific to the characteristics of a particular asset and coupled with IoT data generated from the asset to predict potential failures.

Conclusion and Recommendations

Technology advancements in sensors, analytics and simulation can be the solution to predicting maintenance problems, but only if you take a digital twin configuration approach first. An individual asset’s context is king. Use it as the baseline to predict failure from data generated from IoT sensors and validate it with purpose-built simulation models.

A few key points to remember as you pursue a digital twin configuration strategy:

• Develop the business processes and technology to support tracking and changing an asset’s configuration first. Without being good at this, there will be no value in applying other technologies.
• Use IoT sensors and data compared against an individual asset’s configuration, not the generalization of all “like” assets. Doing the later will result in weak results.
• Use the power of simulation to build digital models of individual configurations of assets. The digital model from the manufacturer will have a short life or no life at all.