Predictive Maintenance, The Smart Way to Cut Downtime

Sept. 1, 2005
A predict-and-prevent approach yields more fruit than a fail-and-fix mentality.

Toyota Motor Corp. is no different than any other company. Like the rest, the automaker wants maximum uptime at the least cost possible. But it differs from many manufacturers in that it has an aggressive program aimed at cutting costs by tens of percentage points. And intelligent predictive maintenance is an important element of its strategy for doing so.

Toyota management proved the value of investing in maintenance on a machining line in Japan a few years ago. A team of engineers, technicians and operators there was able to cut downtime due to maintenance problems in half by applying the fundamentals of predictive maintenance. The machining line had been running at about 82 percent of the time that it was supposed to operate. “About 12 percent of the downtime was due to maintenance problems,” says Mark Rucker, power systems specialist at Toyota Motor Manufacturing, in Georgetown, Ky. “That 12 percent was cut by more than half within a year of doing that kind of analysis.”

Attention grabber

As one might imagine, this kind of success can attract quite a bit of attention. Toyota’s management, for example, now wants to install more intelligent maintenance systems to repeat the success in its plants worldwide. Other manufacturers in a variety of industries also have taken note, and are trying to implement the basic principles in their factories.

Toyota’s plant in Georgetown is complying with corporate continuous-improvement mandates by working with researchers at the National Science Foundation Industry-University Cooperative Research Center for Intelligent Maintenance Systems (IMS Center). The research is a multi-campus endeavor involving the University of Cincinnati and the University of Michigan, in Ann Arbor, as well as more than 40 global companies. The goal is to develop technologies and tools to help factories to reduce the number of breakdowns on their machines and systems to nearly zero.

The researchers plan to achieve their goal by learning to predict failures, rather than simply monitoring the status of machinery and reacting to problems as they develop. “Today, machine field services depend on sensor-driven management systems that provide alerts, alarms and indicators,” explains Jay Lee, Ph.D., founding director of the IMS Center. “The moment the alarm sounds, it’s already too late to prevent the failure.” So he advocates monitoring the important sources of degradation over time, using trends in feedback to forecast problems before they develop, and scheduling maintenance when it becomes necessary—that is, neither too early nor too late.

These intelligent maintenance systems would make predictions based both on real-time data from sensors on the machine and on quality and historical information already resident in enterprise-wide computer systems. “The goal is to predict product and machine health in the same way that the weather is forecasted,” says Lee. “We really don’t care about how precise the temperature prediction is. We care about the trend—cold to hot or clear to rainy.” Such a forecast would allow users to establish priorities and create a plan for maximizing asset utilization.

Interpretation is key

Research at the IMS Center includes finding the appropriate metrics and sensors for measuring them. Although Rucker expects Toyota to benefit from this research, he believes that his company will benefit even more from the efforts to transform the collected data into useful information in real time. The reason is that the controllers on today’s automation already collect tremendous volumes of data. “There’s an incredible amount of data coming off the lines in terms of machine and line performance,” says Rucker. “It’s just sitting there.”

In many cases, the problem is not the lack of technology for gathering the data, but the algorithms for interpreting it. “We have a lot of people in the company that deal with statistical quality analysis, but that’s only half the story,” says Rucker. “There are other signal processing techniques that the IMS Center can bring to bear on the problem.”

For example, researchers at the IMS Center are helping Toyota’s facilities engineers in a yearlong project to look for ways of saving money in the operation of the plant’s air compressors. Because rebuilding these 6,000-cubic feet per minute centrifugal units costs tens of thousands of dollars, the team is developing two models, one to predict bearing wear given the varying loads that the compressors experience throughout the day, and the other to control surges and damaging back flow. The goal is to generate cost efficiencies by finding the right parameters to measure, and then developing software to monitor and control them.

Right now, the vibration monitors on the compressors detect problems only after they have already developed. More accurate predictions of bearing wear would let maintenance staff schedule an overhaul before the bearings begin to fail and acquire the potential to cause costly damage. The trick, of course, is not to change the bearings more often than necessary.

The other software under development will prevent the surges and back flows that can create damage that can cost hundreds of thousands of dollars to repair. Although manufacturers offer a standard operating curve to prevent surges, the curve is a conservative measure to cover most operating conditions in a variety of factories. “The IMS Center is working with us to predict where that surge line really should be, based on varying inputs such as air pressure, humidity, oil temperature and other machine operating parameters, so that we can run closer to the surge line without going over,” explains Rucker. Once the researchers determine the exact relationship, they will be able to encode it into the controller so it can make the necessary adjustments automatically to minimize operating costs without an increase in risk.

Besides specifying the right parameters to measure, and developing algorithms to interpret them, an even more fundamental part of intelligent predictive maintenance is to conduct root-cause analyses. “We spent about a year learning how to do root-cause analysis based on reliability-centered maintenance principles, and we have begun applying it to our facility systems,” says Rucker. “It’s based on understanding the real behavior of the machine in the field: How it really breaks down, and how it affects the plant.”

The right tools

The purpose of such analyses is to direct limited resources to where they will have the greatest effect. For example, Toyota’s Facilities Maintenance department decided to run the exhaust fans on the roof to failure, rather than greasing the bearings and replacing belts according to the manufacturer’s recommended maintenance schedule. “It’s cheaper to replace and fix a couple of dozen units a year than it was to service hundreds,” Rucker explains. “In one building, we had about twice as many exhaust fans as we needed anyway to maintain good pressure in the building.” His group decided to redirect resources to a much worthier endeavor—maintaining the chillers that lower the temperature of the outside air by 10 degrees Fahrenheit before the ventilation system delivers it to the line workers.

Although the research has several months to go, Rucker reports that the root cause analyses behind it has already paid handsome dividends here and elsewhere. And it promises to pay even more once the facilities maintenance group can use the results of the current research project, and as the concepts spread to the various production groups in the plant. Rucker and the IMS team have already begun passing along the concepts to other maintenance groups that service production equipment.

The IMS Center, of course, is not the only locus of research and development in intelligent predictive maintenance. Other consortia and many vendors have their own programs underway. The result is a growing number of controllers that have the ability to sift through the data they collect and convey useful information to operators.

“Preventive messages and warnings are definitely on the increase,” says Noel Nichols, director of engineering at CL Automation, an equipment builder based in Machesney Park, Ill., and formerly called Cincinnati Lamb Assembly and Test. “We have them on probably 70 percent of our equipment now.”

On the custom automation equipment it builds and services, CL Automation programs the human-machine interfaces (HMIs) to monitor and make sense of the signals that they receive. For example, an HMI can evaluate the quality of signals coming from various sensors and tell the operator whether, say, a particular photoelectric sensor is dirty and needs cleaning. It also can count the total cycles that a servo motor works and alert the operator when the time comes to service it. On a higher level, this ability to count cycles can keep track of automatic machinery, such as a robot running three shifts a day, and issue the appropriate maintenance requests when the time comes for a routine action.

Users can expect to have access to even more detail about the condition of their machinery as drives manufacturers continue to add intelligence to their products. Consider the information now available from the microprocessors that Bosch Rexroth Corp., headquartered in Hoffman Estates, Ill., puts in the latest generation of its IndraDrive servo drives. A machine tool builder can store in the drive the ideal torque and current signatures for standard testing routines for evaluating the performance of one of its machines.

When the machine is down for routine maintenance, one of the user’s maintenance technicians can run it through the same routine and compare the new signatures with original data to determine whether the axes are too stiff or have too much play in them. The results then can be uploaded into Microsoft Excel or other software on any personal computer (PC). The HMI on a PC-based computer numerical controller (CNC) running the machine can display the results locally, or the technician can send the file through the Ethernet network connection to the factory enterprise system so an engineer can retrieve and analyze it. If the discrepancies between the ideal and actual are large enough, the technician or engineer can submit a repair order.

In some applications, such requests go directly over the network to the maintenance office, using software available for this purpose. CL Automation’s monitoring software, for example, receives such alerts, schedules downtime, tracks corrective actions and generates reports. “We set up an intranet application that can be broadcast throughout the corporation,” says Nichols. “One company even sends a page to the maintenance department, so technicians actually get an audible sound alerting them that they need to address a problem.” Others that have service contracts with equipment builders or third parties transmit the messages over the Internet or other network.

To close the information loop, some users install identification-card readers on the machines. As maintenance technicians answer calls and make the necessary repairs, they swipe their cards through the readers to turn off the alert messages and to update the maintenance log.

Good for business

Users who are backing the research in intelligent predictive maintenance and buying the supporting technology tend to be those who look at the value of equipment to the enterprise over its lifetime, rather than just the up-front costs. No matter how long the company plans to keep the equipment—whether it is three or 20 years—there is a specific cost to own it. Predictive and preventive maintenance programs can cut the cost to own by between 15 percent and 30 percent over the equipment’s lifetime, according to CL Automation.

The lower costs come from greater mean times to repair and greater mean times between failures. “Once you establish a plan for predictive maintenance, you’re able to schedule maintenance at a planned time,” explains Nichols. “A planned scenario gives you the ability to prevent unplanned downtime.” Moreover, when an expected problem does occur, the technology can give the maintenance supervisor a good idea of what is wrong before dispatching a technician, allowing the technician to arrive prepared to solve the problem and get the machine running sooner than would be possible otherwise.

Fewer unexpected repairs and less unscheduled downtime make manufacturing much more predictable, which offers businesses a number of other benefits. For example, manufacturers can set operating budgets and to stick to them. They can carry smaller inventories of spare parts, ordering many parts as they need them. Moreover, greater predictability helps factories to keep production schedules and make promised deliveries to their customers.

Reaping these benefits, however, requires more than simply having a predictive maintenance plan. It demands action. Users cannot be like a driver who ignores the change-oil light on the dashboard when it comes on. “When you get the alert, you have to do the required maintenance,” urges Jeff Anderson, service manager at CL Automation. “That calls for management to be involved and to enforce compliance.”

Only then will it be able to maximize uptime and cut cost at the same time. “Rather than reactive maintenance—fail-and-fix—companies can indeed move to predict-and-prevent maintenance,” says Lee at the IMS Center.

For more information, search keywords “predictive maintenance” and “preventive maintenance” at