When Smart Technology Meets Old Machinery

Shiny new factories outfitted with robots, the winking lights of computerized controls - and very few humans may be the popular notion of contemporary manufacturing. But as the following article explains, it is far from the case for most industrial plants.

Retrofitting with sensors and computerized monitors provides one answer to the demand for technological competitiveness.

Doing so successfully may require rethinking processes and people as well as machines. JL

Mary Catherine O'Connor reports in the Wall Street Journal:

The vast majority of factories today, are still running on decades-old machinery that isn’t outfitted with sensors. Getting from where we are now to the factory of the future can be done—has been done—but it isn’t as easy as strapping the industrial equivalent of a Fitbit onto each piece of old equipment in a plant and calling it a day.
It’s a tantalizing vision: Bright and shiny factories where robotic arms and conveyors never break down and production goals are never missed—all thanks to internet connected sensors that monitor machine health and respond to the slightest supply or logistics hiccup.
But for the vast majority of factories today, the reality could hardly be more different. They’re still running on decades-old machinery that isn’t outfitted with sensors. Getting from where we are now to the factory of the future can be done—has been done—but it isn’t as easy as strapping the industrial equivalent of a Fitbit onto each piece of old equipment in a plant and calling it a day. It’s costly. There are no ready-made solutions—each case is different. And it requires a deep understanding of each machine’s functions and the metrics to be tracked; trial and error to determine the right sensor to use and the best place to put it; and a plan for collecting, filtering and making sense of the collected data.
“Many shop floors are covered in machines from 10, 20, 30 or 40 years ago,” says Isaac Brown, an analyst at Lux Research. “Plugging them into the internet is totally not trivial—it’s not like plugging in a PC.”

It’s complicated

It’s an exercise that’s familiar to Mike Fisher, general manager of Harley-Davidson HOG -1.37 % ’s York, Pa., manufacturing plant.
In 2010, the plant began a massive restructuring and deployed a software system that collects data from manufacturing equipment to look for early signs of mechanical problems. Hundreds of machines were retrofitted with sensors. Replacing them with new machinery that had sensors and connectivity built in wasn’t considered an option, because much of the existing equipment, at around 10 years old, has decades of production left in it.
The reason manufacturers go to all this trouble is that installing sensors on equipment and connecting them to computer networks—what some call the Industrial Internet of Things—can enable plant managers to track such metrics as temperature and vibration to keep their machinery operating at peak efficiency, and alert managers to problems that could slow production or, worse, shut down a line or an entire plant. The data from the machines also powers analytics systems that can predict problems long before they’re likely to occur.
Sensors “make the equipment more complicated, and they are themselves complicated,” says Mr. Fisher. “But with the complexity comes opportunity.”
The difficulties start with choosing which sensors to install. “Making sure you have the right ones can be difficult,” Mr. Fisher says, because sensors aren’t made with the particulars of each machine in mind. Often plant managers can’t tell which sensor will most accurately collect the data they want from a machine without a series of test runs—a time-consuming process.
Installing the sensors is another challenge. Ensuring that they are placed on or integrated into the equipment so that they collect the intended data—not vibrations from an adjacent machine, or heat being generated around rather than inside a motor—requires calibration work by experienced engineers. General Electric Co. GE -0.70 % faced those challenges at its power and water plant in Schenectady, N.Y. To monitor the power usage of a massive milling machine, GE attached vibration sensors to the machine’s pumps and mounted a current transducer to its control box, to track voltage.
Out of the gate, the system failed to definitively detect when the machine was actively milling, says Shaopeng Liu, a cyber-physical systems engineer at GE Global Research. But after some tweaks, the right data was teased out of the behemoth, allowing it to be run more efficiently.
Even when everything is up and running, though, a plant manager’s worries aren’t necessarily over. “They fail sometimes,” Mr. Fisher says of the sensors.
Despite all that, the overhaul at the Harley-Davidson plant was a success. Accelerometers now send data to a vibration-analysis program that can help forecast mechanical glitches. Thermographic cameras alert operators when a machine is running hot. Ultrasound sensors look for air leaks. Sensors track the temperature and humidity of paint as it is applied to motorcycle components, as well as the speed at which it flows, to prevent clogs or other failures.
The system has allowed Harley-Davidson to do away with much of the redundant equipment it used to keep on hand in case any of its machines conked out. “In fact,” Mr. Fisher says, “the machines are lasting longer than the electronic components that control them,” because of the plant’s sensor-based predictive maintenance program.

Wireless or wired?

Cost, of course, is always a consideration. Chris LeBeau, global IT director for systems integrator Advanced Technology Services Inc., says sensor prices vary widely, from $150 to $500, depending on ruggedness, the sensor technology, processing capabilities and connectivity options. As many as 40 sensors might be added to a single, vital piece of machinery, and when the labor and engineering costs of planning, testing and executing the retrofit are included, costs could hit $100,000 for one machine.
Plant managers can choose wireless sensors or ones that are hard-wired into their machinery. A number of factors are at play in that choice, including cost considerations. Even though costs recently have plummeted for wireless sensors, they generally are still more expensive than wired sensors. And wireless connections aren’t always reliable enough, especially when continuous immediate data collection is critical to keeping a plant humming—even a brief loss of connectivity can cause problems. Plus, many plants don’t have Wi-Fi networks, and adding one in a large facility can be a big expense, says Mr. LeBeau. “It can be a $250,000 investment for Wi-Fi coverage,” he says.
It often makes more sense to hard-wire sensors into equipment, says Sundeep Oberoi, global head of niche technology delivery for Tata Consultancy Services. Cost is one reason. Also, with wired sensors plant managers don’t have to worry about delays in data collection because of connectivity issues, or about the potential failure of the radios that transmit the data collected by wireless sensors, he says. But costs can climb for wired sensors when machines require an auxiliary port to accommodate them, he adds.