The secret life of IoT sensors
Sensors are taking over the world. They are now embedded in nearly every aspect of modern life, including smartphones, vehicles, and home automation systems. In industrial settings, sensors run the gamut from simple gauges of temperature or light to specialized sensors that measure electromagnetic waves or flow rates.
"Sensors are basically taking the physical world—the mechanical processes of an industrial operation, for example—and digitizing them," says Lin Nease, chief technologist, IoT, at Hewlett Packard Enterprise.
Many IoT sensors will be focused on fundamental operational tasks, such as classifying raw materials, verifying production output, measuring operational efficiency, and keeping workers safe. Others will monitor the health of the equipment itself.
"A lot of industrial operations have compressors, pumps, boilers, cutting machines, CNC machines—you name it," Nease says. "And typically, people are using sensors to ascertain the state of those assets. Is it working properly? When might it fail in the future? When does it need maintenance, and why?"
Anatomy of a sensor
IoT sensors can vary greatly in terms of cost and complexity. For example, inexpensive photoelectric sensors are widely used in manufacturing to position objects, count products, or activate safety systems. A simple setup consists of a transmitter that emits a light beam and a receiver that detects the light. If the beam is interrupted, the sensor will register that an object has passed. At scale, such sensors can help manage production flow on a high-volume assembly line.
At the other end of the spectrum are advanced sensors capable of gathering rich data and turning it into actionable insights using edge processing and specialized applications. For instance, video streams from an array of high-resolution video cameras can be fed into machine learning applications at the edge for automated quality assurance, traffic management, or building security.
Modern IoT sensors are based on integrated circuits, the cost of which has been declining for years: The average sensor cost dropped 200 percent between 2004 and 2018, per a Microsoft report on manufacturing.
Even higher end sensors are experiencing a "race to the bottom" on costs, according to John Schultz, who was the chief innovation officer at Allied Reliability at the time of the interview. He says triaxial accelerometers that also measure temperature are headed from current price levels of about $250 to less than $200, and he predicts the cost will eventually drop below $50.
IoT sensors may be built into industrial machinery or retrofitted to legacy equipment. Stand-alone sensors may also be attached at various points along a production line or deployed at remote sites to monitor unattended processes. At a factory, sensors are often hard-wired to industrial control systems. Wireless sensors powered by batteries or low-voltage connections have proliferated in recent years, not only greatly lowering the cost of installation but enabling businesses to gather data that may have been difficult to obtain in the past.
IoT sensors identify potential failure modes
Consider operations at a batch manufacturing plant. For a particular workflow—say, a mixer handles step 1, a human worker is responsible for steps 2 and 3, QA takes place at step 4, and a container is filled in step 5—sensors will help manage the process and the equipment along the entire production line. The sensors can determine what's happening when and where from a digital perspective.
"For monitoring assets, you'll have sensors for vibration, temperature, humidity, and viscosity of a fluid," Nease says. "You may have sensors to tell you whether or not a bin is full or empty."
The bugaboo of any industrial process is downtime. At a petrochemical plant, a failing compressor not only needs to be taken offline to be fixed, but forces downstream processes such as fillers to become idle and slows or halts upstream processes such as mixers.
On a production line with hundreds of components, multiple points of failure can stop production. IoT sensors' role in identifying potential failure modes cannot be understated, Nease says.
"For the vibration and temperature and humidity-type sensors, typically they're used to statistically determine the state of an asset and predict when it will fail and need maintenance," he says.
Allied Reliability, which specializes in digital transformation of maintenance and reliability practices, has built a business based on the behavior of machinery and constituent parts according to elementary physical properties and tolerances.
"There's an interesting misconception out there that an ANSI centrifugal pump in the oil and gas industry behaves and has different failure modes than an ANSI centrifugal pump in the mining industry," says Chris Klosterman, vice president at Allied Reliability. "We have a couple of decades worth of data that suggests if it is properly maintained and running within its design specifications, they really don't."
Since its founding in 1997, Allied Reliability has built an extensive database of machinery, mechanical components, and sensors. For an ANSI centrifugal pump, each one of its maintainable parts—including the shaft, housing, propeller, and bearings—has a discrete number of engineered failure modes. It's possible to map each one of those failure modes to every test, measurement, and inspection that can be performed. This not only determines the types of sensors that can be used to monitor the performance of the equipment, but can also be used to identify potential problems.
Bad data and good data on the edge
According to Allied Reliability, many industrial sensors fall short when it comes to delivering accurate measurements. This is an issue even with well-known brands making big claims about accuracy and performance.
HPE's Nease concurs: "One example I saw involved a sensor that yielded the exact same vibration pattern when it was dropped from two meters high versus when it was dropped just one inch onto a desktop. That's a problem."
Nease adds that the mechanical aspects of how sensors are designed and packaged can cause data readings to be drastically off. "If you don't test that you've got accurate data, you're frequently going to take action that's worse than doing nothing," he warns.
Another issue relates to trust and security, which are obvious concerns for consumer IoT or privacy-sensitive businesses such as healthcare. But even managers at a factory, power plant, or transportation fleet business need to know that the data is "pure," says Leo McHugh, vice president of marketing and customer experience at Analog Devices.
"A trusted sensor is probably the No. 1 thing in terms of being a good sensor," McHugh says, reflecting on his interactions with customers as an engineer and technology leader for more than 25 years. "You have to feel that nobody has tampered with or interfered with the data, and also that it is repeatable and accurate."
He adds that the trust question has an additional dimension when it comes to the sheer volume of data now being gathered by sensors. IDC forecasts that data created by connected IoT devices will experience a compound annual growth rate of 28.7 percent over the next five years, creating 79.4 zettabytes of data by the year 2025.
Chris Ryan, director of ecosystems and digital transformation at Analog Devices, points to the importance of edge processing in maintaining high levels of trust, with additional benefits including reduced latency.
"As you deal more and more with the deluge of information from sensors, we need to know a) how do we make that information secure, and b) how do we turn that data into information that can be used at the appropriate time?" says Ryan. To this end, he believes that edge processing of sensor data can handle both concerns, keeping the data in a trusted state and delivering quick, smart decisions at the edge without having to transverse a wide-area network back to a cloud service.
IoT roadblocks and bottlenecks
Even when businesses have access to good sensors, good data, and up-to-date IoT and edge systems, managers also must consider how processes and people are impacted.
"We worked with a very large packaged textiles manufacturer," Nease recalls. "They had a cutting blade. We figured that by using an audio sensor, the settings of that cutting blade could be more precisely tuned, making it possible to get twice the life out of a blade." In the end, he says, the goal wasn't feasible because nine other processes would have to be redesigned.
Analog's McHugh says there is also a risk of an internal expert becoming a bottleneck in an organization.
"If everybody is sending all this data to the subject matter expert, they're just going to be overwhelmed," McHugh says, adding that edge processing highlighting unusual occurrences can take much of the load off of the human experts.
HPE's Nease thinks some businesses will need to take things a step further and enlist extreme sensor experts who know how to evaluate and optimize sensors and the processes and equipment they monitor.
"While we talk a lot about digitalization and automation and machine learning, there's still a need for the root of trust to somehow label data and to ascertain whether it's accurate or not," he says. This involves appraising the sensors and their outputs using specialized test equipment.
"The ability to test a sensor goes beyond just using it in an operation," Nease says. "It actually requires test scaffolding in its own right."
- A trusted sensor is key in terms of being a good sensor.
- The bugaboo of any industrial process is downtime. Sensors can help with that.
- A sensor's role in identifying potential failure modes cannot be emphasized enough.
A trusted sensor is probably the No. 1 thing in terms of being a good sensor. You have to feel that nobody has tampered with or interfered with the data, and also that it is repeatable and accurate.
This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.