Design, deliver, and run enterprise blockchain workloads quickly and easily.
Thanks to IoT, predictive maintenance gets an extreme makeover
It’s a beautiful day and you’re enjoying a drive in the country. Suddenly, the “maintenance required” light on your dashboard comes on. Should you keep driving or call for roadside assistance? For an answer, you pull over and begin poring over the owner’s manual. So much for your carefree outing.
It doesn’t have to be that way anymore. In the past, many vehicle warning lights were set at the factory to illuminate at regular intervals, like 10,000 miles. But the advent of IoT enables automobiles to think for themselves. Increasingly, when a warning light flashes, the car really means it. Sensors are telling the driver that a threshold of heat or wear is approaching and it’s time to do something about it.
What's new in predictive maintenance
The concept of predictive maintenance is probably as old as the first machines that humans invented. In the recent past, predictive maintenance solutions were focused on complex machines, such as jet engines, for which failure would be catastrophic. Often, those systems involved the compilation of large amounts of historical data, the application of machine learning, and the creation of a digital “twin,” a digital rendition of the physical machine that would perform virtually like its physical counterpart. Now, with the advent of the Internet of Things (IoT), the lowering of connectivity and storage costs, and the creation of vast amounts of data, predictive maintenance is transforming industries and machines that were previously out of reach.
“Moore’s law has changed the cost of sensors and analytics,” says Lin Nease, chief technologist for IoT at Hewlett Packard Enterprise. “Now, cars, dishwashers, and air-conditioning systems are all candidates for predictive maintenance.”
IoT-fueled predictive maintenance is riding the wave of low-cost sensor proliferation. In 2004, the average cost of an IoT sensor was $1.30. By 2020, the average cost will drop to 38 cents, according to Goldman Sachs and BI Intelligence estimates.
Lower costs mean wider deployment. According to Gartner, the installed base of IoT units will reach 8.4 billion worldwide in 2017, up 31 percent from 2016, and 20.4 billion by 2020. These trends are driving the growth of the IoT economy: Gartner predicts total spending on endpoints and services will reach almost $2 trillion in 2017.
Lower costs and more devices are not all that is changing. In the past, methods for predicting failure were dependent on domain knowledge—you had to have experts who intimately understood the ins and outs of particular machines to make an educated guess as to the likelihood of failure. Today, sophisticated algorithms can make rapid predictions, often taking into account factors that previously might have been overlooked or not fully understood.
How predictive maintenance works
A predictive maintenance system consists of five elements:
- The thing itself. A machine or component that performs an important function, with parts that wear and might eventually fail.
- Data acquisition. The sensors on a machine that collect and transmit time-series data to a data acquisition system, converting the data to a format that can be understood by the analytics system.
- Analytics engine. Takes data from the acquisition system and generates events. For example, a temperature change on a part such as a carburetor constitutes an event. The analytics engine calculates when the event occurred and creates patterns of events that can be studied to develop predictions.
- Historian system. A database that receives and stores the same data that goes into the analytics system. It provides points for comparison with current events in real time. It can be located on-premises or in the cloud.
- Human interface. The predictive maintenance system visualizes its information through an asset management, business process, field maintenance, or other application.
Learning to love predictive maintenance
There are two main benefits to knowing exactly when maintenance is or isn’t needed. First, it eliminates unnecessary maintenance carried out according to a timetable rather than actual need. That cuts down the time and expense of routine maintenance procedures.
Second, it increases the likelihood of avoiding failure. By knowing ahead of time when failure is impending, a system can be taken offline and fixed so the failure does not occur. Downtime is avoided, customers remain satisfied, and corporate reputation is upheld.
If the machine life extension that results from the use of IoT analytics is 15 percent or less, you probably would not make a change. Between 15 percent and 30 percent, you might consider adopting a new system. Above 30 percent, you would definitely change to the new method.
Consider this scenario: A consumer products company makes diapers on an assembly line. The machine that cuts the fabric must be adjusted regularly to make clean cuts. In the past, a foreman would closely observe the cuts for quality and tighten the mechanism as needed. Today, an IoT sensor can accomplish the same task at a lower cost and with greater precision. When a sensor on the cutting plate picks up vibrations that indicate erosion in quality, it sends an alert that it’s time to adjust the cutting mechanism.
Here’s another example from the automotive industry. One day, quality inspectors at a manufacturer of doors and exhaust systems discover an unacceptably high number of product defects. At a meeting the next day, the cause is identified: The robots that make the products have exceeded their maintenance threshold and are in need of repair. In response, the company deploys a real-time IoT sensor-based analytics system that monitors robot performance and enables repairs so that defects don't occur.
Predictive maintenance as a service
Some providers are reducing the cost of predictive maintenance by offering IoT analytics as a cloud-based service. One such provider is Falkonry, which equips the IoT applications of its customers with real-time pattern recognition capabilities. Another is SparkCognition, which supplies cognitive algorithms that learn from sensor data and identify imminent failures.
Trust and ROI
Despite the promise of significant benefits, two major obstacles to deploying IoT-based predictive maintenance systems remain. The first is trust. Many organizations are reluctant to override systems that may have taken years, possibly decades, to evolve to an established and trusted level.
The second is return on investment (ROI). “The biggest barrier is not knowing whether or not the analytics will provide the value people are hoping for," says Nease. "The ROI may not be obvious.” In large-scale industrial production, for example, it can be difficult to test a hypothesis regarding suspected maintenance issues without taking down a production line. That's a significant disruption.
Nease offers this rule of thumb: If the machine life extension that results from the use of IoT analytics is 15 percent or less, you probably would not make a change. Between 15 percent and 30 percent, you might consider adopting a new system. Above 30 percent, you would definitely change to the new method.
Nease recommends a common-sense approach to resolve both trust and ROI issues: Try a pilot project. By running a proof-of-concept initiative in which domain experts can compare the results of the new algorithms with those of the previous system, it is possible to make an informed choice as to whether the new system will deliver results that are worth the investment.
For complex factory operations, the costs and benefits may take some study and analysis. Millions, possibly billions of dollars and the future of a company might be at stake through more efficient operations. Higher quality and more efficiently produced products will lead to a legion of more satisfied customers. For consumers, the widespread adoption of IoT-based predictive maintenance may seem like a no-brainer. Knowing what’s really going on with your vehicle just might be the difference between an afternoon of anxiety on the side of the road and a relaxing drive in the country.
Predictive maintenance: Lessons for leaders
- Predictive maintenance is not just for jet engines. Thanks to lower sensor costs, there are new horizons to explore, particularly in consumer products.
- You don’t have to build it yourself. IoT analytics as a service could be the best way to get started.
- Payback won’t just be in the short term due to more efficient manufacturing. Fewer defective products will mean greater customer satisfaction.
This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.