November 4, 2016
In the industrial internet of things, data and AI-based analytics transform complex systems into engines for growth. Here’s how.
Digital-related investment in industrial production is growing fast. Through 2020, enterprises expect to pump $907 billion annually into digital technologies on the industrial floor, according to PwC’s Industry 4.0 research. That investment is expected to increase revenues by $493 billion annually and reduce costs by $421 billion each year. But where and how those dividends will be unearthed is only now coming into view. PwC sees enterprises following a path that spans prediction, prescription, optimization, and new business models. And at the core of it all is how well they are able to mine the rich data streams that provide a new window into their operations.
Industrial systems are incredibly complex; numerous components, subsystems, and variables affect overall performance. Moreover, enterprises adapt the systems to their unique (and often undocumented) working behaviors. The results are systems with so many interdependencies that even the specialists who tend to them may have difficulty comprehending why something isn’t working as it should.
Consider a railway transit system: About 80 percent of the time, experienced locomotive engineers can pinpoint the reasons for train breakdowns, estimates Ruud Wetzels, data analytics manager for PwC Netherlands. But they’re vexed by the other 20 percent. “Even for very experienced engineers, it’s not possible to see the complex interactions across hundreds of machines or subsystems,” he says.
Now, however, with the instrumentation of industrial equipment (from trains and turbines to farm machinery and pumping systems in oil fields), getting at that elusive 20 percent is now possible. The emerging industrial internet of things (IIoT) is essentially changing the nature of the problem—from a complexity one to a data one—and making it solvable. Many correlations that were previously unobservable can now be detected in equipment data streams via advanced analytics using artificial intelligence (AI) and machine learning.
Industrial systems are highly complex and challenging to manage, but analyzing their rich sensor data enables companies to see and do more: predict behavior, prescribe actions, optimize operations, and create new business models.
In the case of those locomotives, this approach is paying off: A transit company that PwC has worked with now has visibility into its trains’ myriad systems. It has operating data on thousands of variables, as well as external information like real-time weather conditions. Using this data, the company developed a machine-learning algorithm that predicts if a train would fail, a half hour in advance—enough time for the operator to have a replacement engine dispatched. The algorithm is now able to predict about 40 percent of the potential failures.
From predicting failures to creating new business models
Like the transit company, most industrials are using IIoT to predict equipment or system failure. Any failure has the potential to significantly disrupt business and adversely affect revenue and profits, as well as create risks for loss of life or damage to assets. Failures also inconvenience customers and could drive them to competitors.
For example, PwC is working with a major airline that uses structured data from sensors as well as unstructured data that maintenance engineers generate when they work with aircraft. This data is mined and converted into key variables that are fed into analytical models to look for patterns that might indicate an impending problem. By using this approach, the airline can predict as much as 30 percent of potential delays and cancellations.
In another case, a PwC oil and gas client uses IIoT sensors in pumping systems that are part of oil well environments. These environments are tremendously hostile—hot, dirty, and under extreme pressure often a mile underground. The sensors on the pumps generate data that is used to predict if a catastrophic failure is imminent, and operators can shut down the pump before the failure occurs. The difference between the proactive repair and the post-disaster repair represents a significant safety improvement as well as significant cost savings.
Predicting failure can be challenging. If the ultimate algorithm is too aggressive, the result is a plethora of false positives. That leads to a tricky dilemma, says Nikunj Mehta, founder and CEO of Falkonry, maker of AI solutions for industrial operations and other environments. “If you want to catch the problem early, you set your thresholds low, but that causes a lot of false positives,” says Mehta. “If you don’t set the thresholds high enough, you’re taking a risk, but you might get higher productivity. Every company decides how it wants to operate. It becomes an issue of risk management.”
What’s needed is a balanced approach. “The challenge in predicting failure is to package an analytical regime that has predictive power and has a manageable number of false positives and false negatives. You must build a modeling framework that is robust, reliable, and economical,” suggests Bill Abbott, a principal in the PwC data analytics practice.
Fixing what’s broken
Once a failure is predicted, what’s the next step? Jenny Fielding, managing director of Techstars and an IIoT investor, says you next want to determine why a component or system will fail and what should be done about it. Monitoring isn’t enough. “Just because a system presents a dashboard of data to the end user doesn’t make it useful,” she says. “Don’t just tell me that something is wrong. Fix it. AI is really the key to generating actionable insights around any type of IoT product.”
What is needed is a prescriptive model to generate actions that the business can take. “The basis of a prescriptive model is formalizing the decision models,” says Wetzels. “When do you stop a train? When is the probability of failure high enough? Why do you want to do that? Is it because you want greater customer satisfaction, safety, or uptime? The reason you want to do something will give you a different prescriptive model.”
Enterprises will need to make their business objectives much more explicit up front. For instance, enterprises should clearly define which business objectives—customer satisfaction, cost, profit, safety, or uptime—are to be prioritized over others, and when.
Achieving higher performance
The next step is to optimize asset use as well as human effort, so enterprises can reduce cost, reduce waste, and create flexibility in operations. The complexity of industrial operations makes it difficult to ascertain whether equipment is always operating at optimal efficiency.
IIoT data can change that. Analysis of the data will allow machines to surface information about process efficiency, material waste, energy waste, and operational efficiency. Analytics also can help find opportunities to optimize human behavior and asset behavior.
The effects of optimizing operations will extend beyond improved efficiencies in day-to-day activities. The optimization of operations also can influence downstream activities such as repair processes and upstream activities such as the design and assembly of equipment.
Driving new business models
The final step on the IIoT journey can be the creation of or the transformation to a new business model, such as equipment as a service (EaaS). Thus far, EaaS has not been widely adopted because systems could not accurately predict equipment failures. Customers paying for equipment in this fashion quickly become disenchanted when they experience downtime due to product failure. Improvements in IIoT technology have helped the industry move toward the EaaS business model: Jet engines, for example, are already purchased on a service basis. The transparency provided by IIoT systems will likely allow this model to expand to many other industries.
Barriers to the business adoption of IIoT solutions
At the core, achieving success with IIoT is a data challenge. The successful use of IIoT spans the full lifecycle of data, from surfacing it, transmitting it, formatting it, analyzing it, storing it, and possibly deleting it. Not surprisingly, many of the challenges related to IIoT adoption are related to data.
- Data quality: Data comes from multiple sources, has multiple formats, can be structured or unstructured, and can be incomplete, spiky, or noisy, therefore requiring significant effort in cleanup before reliable analysis can be performed.
- Data security and privacy: Data might contain proprietary operational information that has competitive value, as well as private information related to employee use.
- Data standards: The lack of industry-wide standards for formatting and sharing data across systems and with the ecosystem makes it challenging to share data, integrate workflows, and create automation.
- Managing data volumes: IIoT adoption will significantly increase the data variety and volume that organizations must manage.
- Cost: Enterprises will incur costs to update the equipment with new sensors, update the connectivity (and pay for it), and purchase IIoT solutions to manage it all.
In addition to the data challenges, IIoT adoption will entail the cultural and operational challenges common to the introduction of new solutions and process changes.
IIoT is becoming strategic to all industrial enterprises. Here are some recommendations as you think about using IIoT in your operations:
- Identify and pilot IIoT with equipment already generating data: Identify equipment that is already instrumented, and pilot projects that use this untapped data.
- Understand data challenges and the need for new talent and skills: Invest in data integration and data management capabilities. Acquire talent that has skills in machine learning, artificial intelligence, and emerging analytical methods. You’ll also need skills in user interface design to present data in a way that is easy to understand and interact with.
- Rethink operations to improve predictive and prescriptive capabilities: As you have greater confidence in predicting key events, rethink your operations and scheduling for maintenance, downtime, and so on.
- Build an architecture for flexibility: Investments in industrial equipment and IIoT sensors and devices will likely have a life span of decades. Also, many aspects of the equipment are proprietary. Use modular architecture and application programming interfaces (APIs) to create flexibility in processes and capabilities, so operations can change without major disruptions.