Equipment Complaints as Diagnostic Clues

The latest generation of diagnostic and prognostic technology can monitor and analyze such clues so you can apply the right dose of preventive maintenance early enough to cut costs, extend service life and improve process stability.

Some may dismiss the condition-monitoring and diagnostic systems available today as an elective procedure, but not the Metropolitan Wastewater Reclamation District of Chicago. The MWRD operates the largest wastewater treatment plant in the world, the Stickney Water Reclamation Plant in Cicero, Illinois (in addition to six other plants and 23 pumping stations), and it cannot tolerate unexpected failures on its decanter centrifuges. Uptime must be nearly 100 percent to keep the lifecycle costs low enough for the units to pay their way.

Alfa Laval manufactures these centrifuge, which separate solids from liquid with a rotating drum. According to Peter Bloomberg, automation manager at Alfa Laval’s Copenhagen plant, “With rotational speeds of 3,000 rpm, anything but the most perfectly smooth motion can be detrimental to the machine’s life expectancy.”

To alert maintenance crews to developing problems before they can lead to failure and downtime, Alfa Laval puts a vibration-monitoring module on all of its Aldec G3 decanter centrifuges. By doing so, the company joins a growing number of machine builders fitting their equipment with new monitoring technologies for diagnostics and prognostics. Because these technologies are driving the cost of condition monitoring down, they have tilted the economics in favor of using more automation for preventive maintenance—enabling plants to extend machine service life and improve process stability.

Alfa Laval previously offered vibration monitoring on its decanters to boost their stability and service life, but the technology was just too expensive to offer as standard equipment. In addition to the accelerometers, the old monitoring scheme required adding data-acquisition modules and a computer dedicated to compressing data before sending it offsite for analysis. Not only can such monitoring equipment easily cost as much as the control package itself, but it also required the attention of an expert in vibration monitoring to assess the data. Consequently, Alfa Laval offered this form of preventive maintenance as a premium option.

The situation changed as the result of a continuous-improvement initiative. To reduce the decanter unit’s operating costs, Alfa Laval engineers upgraded its controls to the X20 programmable logic controller (PLC) from B&R Industrial Automation (www.br-automation.com). Besides offering functions that reduce power consumption by 40 percent, the PLC also uses a custom I/O module to convert data from the decanter unit’s accelerometers into suitable output files. The two-channel vibration-monitoring module then transfers those files to the central analysis unit.

This I/O module slides into the controller, giving machine builders a less expensive way to integrate vibration analysis into their equipment. “The cost of the module is a factor of ten less,” says Corey Morton, director of technology solutions at B&R. “The sensors are on top of that, but they are in a range that isn’t out of line with other devices that you’re already going to put on the machine.” Consequently, vibration analysis is now available for many applications that could not afford it in the past.

A condition-based approach
The increasing practicality of adding condition monitoring to a wider range of equipment is fostering an evolution in preventive maintenance, from a scheduled program to a condition-based approach. With a scheduled program, maintenance engineers replace particular components at intervals recommended by the equipment manufacturer, based on average failure rates and suitable safety factors.

“What that means, though, is often you’re replacing components that still had useable life in them and don’t need really to be replaced yet,” says Morton. “There is a cost associated with that.”

A condition-based approach, on the other hand, reclaims and uses that otherwise wasted life. The approach detects the distress that a component experiences before failure. “By looking at, say, bearing wear over time with vibration analysis, you can use the bearing to the point that you know failure imminent before scheduling downtime to replace it,” says Morton. “It’s a huge cost savings.”

Besides extending the useful life of many replacement components, vibration monitoring can also give technicians other useful diagnostic information, such as whether components are aligned properly or whether a load suddenly becomes unbalanced. Because the module contains the intelligence to glean this information from the data, hiring an expert to analyze and interpret vibration signatures is no longer necessary.

Now that Alfa Laval has integrated vibration monitoring into its decanter centrifuges with B&R’s X20 modules, Bloomberg has begun exploring ways to expand their use. “To limit condition monitoring to vibration alone would sell it far short of its potential,” he notes. “Temperature, pressure and a multitude of other criteria can be brought into the calculations for a more comprehensive result.”

Overcoming the obstacles
Condition monitoring and diagnostics are practical in a wider universe of applications in part because developers have overcome two important hurdles over the last decade. The first was obtaining the ability to collect the pertinent data. Sensors and data acquisition and communication hardware are now commonplace on machinery, and most controllers contain a wealth of information just waiting to be mined. So, the necessary data is usually plentiful these days.

The second, more recent hurdle to overcome has been the development of robust and accessible tools for analyzing that data to diagnose failures, determine their root causes, and detect problems while they are still in their infancy.

Analytical tools were a limiting factor in the past because the analyses usually relied on single-variable models, such as the ones behind conventional control charts. “Looking at just one variable—such as vibration, temperature, or pressure—might not be enough to infer the failure mode early,” explains David Siegel, PhD, post-doctoral fellow at the Center for Intelligent Maintenance Systems (IMS Center, www.imscenter.net) based at the University of Cincinnati.

Single-variable analysis also becomes increasingly difficult to manage as the number of tracked parameters increases. “If you have many variables, you have to set as many thresholds,” notes Siegel. “So, you might have a lot more false alarms.”

To generate more a robust method for monitoring and maintaining equipment, the IMS Center has been developing pattern-recognition techniques that use several variables or signals to find problems. Besides limiting the number of thresholds to just one or two, and thereby reducing the possibility of false alarms, these multivariable models also tend to both be more accurate and capable of detecting problems at an earlier stage. “The interaction between the parameters is sometimes just as important as the actual magnitude of any single variable that you’re looking at,” says Siegel.

Multivariable model for robots
Rather than studying large volumes of historical data to identify patterns associated with various problems, researchers at the IMS Center instead developed tools for identifying normal patterns and establishing a baseline. The tools then infer problems from deviations from the baseline. “Because you have an initial framework to detect problems, you can get your automated monitoring system up and running,” observes Siegel. “And as you gather more historical data, you can improve it over time.”
This multivariable technology is helping an automotive plant to streamline the maintenance of its multi-axis robots.

Because the plant had been managing a population of hundreds with single-variable models, it was encountering a fair number of false alarms. A human expert was spending a lot of time configuring torque signals on each axis after each repair and adjusting performance thresholds. The task was too time-consuming to continue doing manually.

“For each of the signals, you cannot set a master threshold that can be true for all robots,” explains Edzel Lapira, PhD, associate director of the Center. “They need automation to look at the signals.”

To provide the automation and reduce the number of false alarms, the IMS Center developed a multivariable model for the robots during a benchmarking study. “We combined current and torque from the axes to indicate which robot needs repair,” says Lapira. “We also provide some diagnosis on which of the axes needs the attention.”

The large population also makes this application a good candidate for remote monitoring. “Remote monitoring tends to make more sense when you have a large fleet of assets and when no one physically is near the assets,” observes Siegel.

The IMS Center is licensing the technology to the automaker through the spin-off company Predictronics Corp. of Milford, Ohio. Several researchers at the Center started the company to offer manufacturers condition-based monitoring, prognostics, and health-management solutions for robots, machine tools, and other industrial equipment.

Ethernet enables diagnostics
Besides computing power and software development, another important enabler for condition monitoring and diagnostics is the rise of Ethernet as the de facto standard communications network in manufacturing. One ramification has been an evolution of web-based diagnostics without separate software tools that run on a vendor’s software environment.

The earliest generations of these web-based services used Ethernet to access a web server and generate diagnostic data, but they were focused on specific devices in a machine. Since then, the services have evolved such that the latest generation offers visibility into an entire machine. “Through a web-based interface today, you can see the entire motion control system—the servos, I/O, and control,” says B&R’s Morton.

“A web-based solution can also serve up video-based instructions to the operator interface for diagnosing problems and, if appropriate, make the necessary repairs,” he adds. “It can even capture event logs in the machine and send them into experts like ours to look at for you.”

Such interfaces can often offer the same visibility into an entire line of machines if the line uses controls from the same vendor. Although this level of visibility may not be possible for cross-vendor solutions, Morton is hopeful that it will become commonplace as data standards continue to evolve and become entrenched.

Another result of the rise of Ethernet is consolidation of what used to be separate networks. Combining separate data, control and safety networks into one, for example, is giving companies like Beet LLC of Plymouth, Mich. (www.beetllc.com) unprecedented accessibility to data. “It allows our patented technology to gather and display massive amounts of data in a very efficient way,” says David Wang, president and CEO.

Beet’s Envision software relies on special drivers to tap into the data already available from controllers. “We want to leverage the existing sensors, controllers and automation infrastructure,” says Wang.

Because the collected data describes every event that occurs in a cycle, altogether it represents a kind of heartbeat for the machine. Exploiting this fact, an algorithm in the software assembles the data and displays it as a kind of EKG, highlighting any deviations from established baselines in color. Not only can users click on the deviations in the EKG to drill down into the data to look for its cause, but they also can access the data later for predictive analyses and planning.

Wang differentiates his company’s maintenance software from most others by describing his as a video and the others as snapshot. “Most diagnostic systems basically give static reports of conditions and faults as snapshots in time, rather than diagnostics,” he says. “Our software paints a living picture of the manufacturing process, displaying a live performance of what is happing on the floor. Every motion and event is accesible.”

Auto plant framer example
During a case study conducted at an automaker’s body assembly plant, Beet demonstrated the power of this ability on a framer, the heart of any body shop.

If the shop is going to make its production goals, clamps must close quickly to hold the underbody components in place for the welding process. The EKG generated from closing-speed data showed that one group of clamps was becoming sluggish, a sure sign that something was going wrong.

Looking at the detail revealed that Clamp 4 was closing at an unhealthy 0.8 sec, instead of the established baseline of 0.45 sec. So, not only was the software able to predict a developing problem that would eventually cause downtime if left alone, it also was able to pinpoint which clamp needed repair.

Consequently, the automaker has been able to find, fix and validate developing problems 75 percent faster, according to Wang. He also reports that the software helped the automaker to improve overall equipment effectiveness (OEE) by 2 percent in two months.

“We can predict failures as much as 10 days in advance, depending on the irregularity” claims Wang. “Mechanical devices don’t just have a catastrophic failure all of a sudden. They are always going to complain first.” You just need to listen to them.

More in OEE