TDWI Upside - Where Data Means Business

3 Keys to Maximizing Machine Learning in Your Enterprise

What you need to know about machine learning to leverage its full potential in your enterprise.

Industrial organizations can lose significant revenue (into the millions of dollars every hour) from downed assets and lost productivity. Thankfully, predictive analytics and Industrial Internet of Things (IIoT) technology are making it increasingly possible for businesses to anticipate and address equipment failure before it happens.

For Further Reading:

Protect Your Network: Use Caution with IoT Data Sources

The Importance of Being IoT

Machine Learning: The Foundation for New-Age Predictive Analytics

These advancements promise to forever revolutionize how fleets of expensive equipment, oil refineries, and manufacturing plants will be managed. Despite all the excitement around preventing unplanned downtime, most managers don't understand how predictive analytics (and its underlying machine-learning base), really works.

Part of the tantalizing potential of machine learning is that it can create valuable insights from massive amounts of data, but it's not as simple as flipping a switch. Establishing a specific business use case and tangible goals are essential to maximizing success with any solution. Understanding the fundamentals of the technology is key to laying that foundation. What do you need to know about machine learning to leverage its full potential in your enterprise?

Data Preparation Is Required

Clear contextual information is a prerequisite for deriving value from the data generated by connected equipment. Each business must also identify its own unique signs for determining why and how a failure happens. To establish this context, businesses must take several steps to prepare their data.

Normalizing data types, such as units of measurement and sample rates, is essential for understanding the data coming in from all your connected devices. This is also true for human-generated information, such as work orders and maintenance records. Because the vast majority of data is just noise, managers will need to pinpoint and prioritize the data most relevant to identifying patterns that illuminate changes in an asset's condition. For example, temperature data might be more critical than time of day.

Businesses can explore potential methods for detecting patterns and behaviors across a large population of machines by using a digital analytics model, a virtual replica of the equipment known as a digital twin. Far more than simply a digital representation of a physical piece of equipment, this type of model provides a behavioral understanding of its internal components, external influences, and theoretical thresholds. It can be queried to determine the probability of a state change or changes that may portend a failure. Businesses can then assign time values to these state changes to predict a time frame for failure.

Machine Learning Is Not Magic

Although its business potential may seem otherworldly, machine learning is not some mystical panacea. Alone, the deep operational insights provided aren't enough to effectively implement predictive analytics, so machine learning relies heavily on the work of subject matter experts (SMEs) to make sense of the results.

SMEs provide the acumen to interpret patterns and behaviors, define normal and abnormal conditions, and recognize and adapt to important changes. For example, SMEs can quickly determine whether an identified pattern is expected and guide the creation of rules to recognize it as normal. On the other hand, new data can lead SMEs to consider variables they may have overlooked when trying to troubleshoot an issue. Experts are also pivotal in mapping future maintenance strategies and technology requirements.

Sophistication Takes Effort

To maximize the benefits of IIoT in your enterprise, apply the insights gained from machine learning and your SMEs back into the physical operating environment. For example, you could use that information, combined with advanced analytics, to develop rules-based monitoring, creating a system to identify and predict irregularities. That said, because asset characteristics evolve constantly throughout their life cycles, it's critical to regularly update digital models and the rules driving the system to reflect these state changes.

By incorporating additional elements of a holistic IIoT solution, businesses can unlock even greater value. For example, automated actions based on rules created through analytics and machine learning help managers streamline workflows and scale operational enhancements across asset populations. Layering in edge computing can help businesses further maximize ROI potential by enabling much of this functionality to take place near (or directly on) the machine, making it possible to quickly shut down a piece of equipment to avoid worker injury, a hazardous spill, or a quality control issue.

Machine learning doesn't just happen. You cannot just copy and paste a deployment, nor can you simply flip a switch and walk away. Operating environments and machine states change all the time, so attentive system management is required to achieve higher levels of operational efficiency, bottom-line productivity, and safety.

 

About the Author

Dave McCarthy is a leading authority on industrial IoT. As senior director of products at Bsquare Corporation, he advises Fortune 1000 customers about 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. You can contact the author via LinkedIn.


TDWI Membership

Accelerate Your Projects,
and Your Career

TDWI Members have access to exclusive research reports, publications, communities and training.

Individual, Student, & Team memberships available.