Demystifying Machine Learning for IoT and Industrial Use

Demystifying Machine Learning for IoT and Industrial Use

An exclusive article by Dave McCarthy, Senior Director of Products at BSquare Corporation*.

In the utopian future, machines will know when they are about to break down, make adjustments to their operations to ward off permanent damage while they continue to work, order the parts needed to repair themselves, and schedule a qualified technician to perform the repairs.

They will be able to schedule maintenance so that unplanned downtime is eliminated, and generate service plans guaranteed to fix the problem right the first time around.

Such a vision isn’t just wishful thinking. The continuing evolution of Industrial IoT (IIoT) technology is ushering in a new era of industrial automation that is driving dramatic advancements in understanding equipment health, and enabling use cases that provide value across the business.

From predictive analytics to condition-based maintenance and asset optimization, machine learning is critical to an IIoT system that is able to scale across hundreds or thousands of different vendors’ equipment at multiple locations. But what exactly is machine learning and what do you need to know in order to make it work for your business?

Machine Learning: The Foundation of Large-Scale IIoT Success

Machine-generated data holds the key to improving business outcomes. Equipment in industries like manufacturing has been equipped with machine-to-machine (M2M) connectivity and intelligence for many years. Compared to the pre-connected days, these systems have provided better visibility into equipment operation and improved decision-making ability, but they often operate in isolation.

Logic dictates that being able to see all the information from all connected equipments would provide a much more comprehensive picture of the overall process. Machine learning is applied mathematics that can sort through the mountains of data created by all these connected machines and unearth the nuggets of information that drive beneficial use cases like predictive analytics.

But getting worthwhile results isn’t as simple as just applying machine learning to a giant pile of data, and it’s not a once-and-done exercise. Before investing time and money in an IIoT project, you should consider the project from an asset lifecycle perspective. Digital models created by and for machine learning can be reused to enable additional use cases for ongoing business benefit throughout an asset’s useful life.

What You Need to Know About Machine Learning

The first step for most companies is consolidating all machine data in a central location. However, with the rate of complex machinery connections taking place, the sheer volume of data generated quickly overwhelms systems that require human management.

On top of that, the data from different connected machines rarely comes in a common format that’s easy to organize and analyze. Machines collect data at different frequencies (every minute or every hour) or in different units of measure (Fahrenheit vs. Celsius). There may be offline data that provides important context around a current condition. Data must be normalized, cleaned up, and sometimes synthesized between different data sets before machine learning can be applied.

Data is also noisy. The vast majority of machine-generated data is irrelevant to the desired outcome. Analytics are required to parse high sample rates and large volumes of data to identify small nuances and patterns that can indicate a change in condition.

Depending on the type of data and the objective, machine learning uses different techniques and algorithms to identify patterns and behaviors across populations of like machines, so even after data is normalized, there is a process of understanding which statistical method is going to be the appropriate one. It also needs to look at the state of each individual machine from its first day in service to its last in order to pinpoint anomalies at the individual level.

The process of mapping patterns, machine states and their complex associations builds an overall analytics model, or digital twin. This model is far beyond a simple anatomical model made up of a machine’s physical components. It’s a behavioral understanding of how the components and their attributes work together, relate to each other, and react to internal influences in the machine itself or external factors like environmental conditions. The model can then be queried as to the state of the machine, the probability of state change or multiple changes that may lead to failure. Then time ratings can be assigned to the state changes, effectively calculating a predictive failure scenario.

At this point, it’s important to understand that machine learning by itself is not the silver bullet for predictive analytics. Subject matter experts (SMEs) and data scientists are still required to interpret the results. Machine learning may identify patterns or behaviors that SMEs know are normal. Conversely, SMEs may be surprised by findings that they did not expect, yet immediately recognize the validity of the results and their value in identifying a potential failure, for example.

Beyond Machine Learning

Machine learning as part of an IIoT system provides much greater insight, but that insight must be incorporated back into the real-time operating environment to fully realize business value. The results of machine learning with SME involvement must be tied in with analytics and translated into a rules-based monitoring system that can identify when anomalous events are about to occur.

However, the state of each machine is constantly changing as it progresses through its lifecycle. It’s important to keep track of those changes through reporting on rules drift and use the new data to retrain digital models for continued accuracy.

Automation is the next step towards realizing the utopian scenario described above. Automating how analytics are employed, how the results are translated into rules, how those rules are applied against data as it comes in, and the execution of any subsequent actions needed creates significant business value. In addition to streamlining workflows, this also allows the system to scale across hundreds or thousands of machines.

The application of edge computing in conjunction with predictive analytics is another part of an IIoT system worth considering. By performing some of the analytics and processing near or on the equipment itself, actions can be taken nearly instantaneously. This is of particular importance in scenarios such as unsafe situations where a machine can automatically shut itself down to avoid worker risk, and even the few milliseconds required for transmission of data to and from the cloud can mean the difference between injury and safety.

artificial intelligence


Machine learning is a critical component of an IIoT system constructed to provide predictive analytics, but it is not a magic wand. It is equally important to understand the business outcomes you are trying to achieve from a complete useful lifecycle perspective so that investments in IIoT can be reused to enable future use cases. Analytics, machine learning, automation, and other aspects of IIoT are complex technologies and require the concerted effort of many stakeholders, from IT to operations and engineering, to guide the project to successful outcomes now and in the future.

*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. He is also a frequent contributor to IT publications, including IoT Evolution and TechTarget. Dave earned an MBA with honors from Northeastern University.

Related posts