Unraveling Mystery of Machine Learning in IoT

There is a lot of buzz around predictive analytics for industrial machinery, particularly in manufacturing and for heavy-duty vehicles in trucking, construction, and farming. Machine learning is often cited as the secret sauce of predictive analytics, but few people really understand what it is and how it works.

The idea behind predictive analytics is that as complex machinery becomes connected to private industrial networks and ultimately the internet, the information it generates can be used to predict and fix failures before they happen, thus avoiding expensive downtime and productivity disruptions. Machine learning is important to be able to make sense out of the huge quantities of data generated by this equipment and drive this knowledge back into the business. Here’s what you need to know about it.

Groundwork is required

Beyond data generated by machinery, contextual information is often necessary to understand why and how a failure happens. These varying sources of data are frequently in different formats, units of measurement, and sampling rates. This becomes even more intricate as you factor in unstructured human-generated data such work orders, maintenance records and job notes. So incoming data requires clean-up and normalization before any analysis can take place. Next, the data must be examined for value. The vast majority is just noise, so analytics are necessary to recognize patterns that can signal a change in the machine’s condition.

Then it’s time to apply machine learning. Different techniques and algorithms for identifying patterns and behaviors across large population of machines must be evaluated depending on the type of data and desired outcome. Each machine must also be looked at as a state machine, from its optimal operation on the first day in service through changes in performance that occur until its final day of useful life.

The information derived from mapping patterns, machine states, and their complex associations is used to build a digital analytics model, or twin, that provides not just a representation of the physical machine, but a behavioral understanding of these intricate interrelations. The model can be queried to determine the probability of a state change or changes that may portend a failure. Time values assigned to the state changes enable calculation of a predictive failure situation on an individual machine.

Subject matter experts (SMEs) are essential

While powerful data analytics and machine learning can provide enormous insight into changes in equipment operation and behavior, SMEs are crucial to interpret the results. They can quickly determine whether an identified pattern is expected and guide the development of rules to recognize it as normal. On the other hand, the results can often spur SMEs to examine areas they may have overlooked when trying to troubleshoot an issue. They can also offer insight on how to weigh the risk of a particular condition in the context of an optimized maintenance strategy.

Taking the long view will provide the greatest value

Before embarking on an IIoT initiative, take a step back and look at your equipment from a complete lifecycle perspective. Machine learning – or IIoT for that matter – is not a once-and-done exercise, and machine learning itself is just one important building block of a complete IIoT system.

To get the greatest business value, the insight gained from machine learning and analytics must be operationalized back into the real-time operating environment in the form of a rules-based monitoring system that can identify and predict upcoming anomalies. Because machine states are evolving throughout their lifecycles, those changes must be continually incorporated back into the digital models and rules driving the system.

The addition of automation –complex actions based on rules created through analytics and machine learning – provides scalability across hundreds to thousands of machines. Edge computing is the final step to reap full benefit by enabling many of the functions outlined above to take place near or directly on the machine. An important example is the ability to shut down a machine immediately to avoid worker injury.

A holistic IIoT system designed with the long view in mind allows digital models to be built that can be reused for numerous other beneficial use cases in addition to predictive failure, including condition-based maintenance, streamlined repair processes, and asset optimization. The technologies that make up IIoT are complex and require the support and participation of many stakeholders across the organization for a system that delivers the greatest benefit and return on investment.

Dave McCarthy is a leading authority on industrial IoT. As senior director of products at Bsquare Corporation, he advises Fortune 1000 customers on how to integrate device and sensor data with their enterprise systems to improve business outcomes. Dave regularly speaks at technology conferences around the globe and recently delivered the keynote presentation at Internet of Things North America. Dave earned an MBA with honors from Northeastern University.