Machine failure can be a tremendously costly problem in the manufacturing industry. Fortunately, new technologies, sensors and big data machine learning can now offer insight that may help better predict failures and reduce downtime.
According to data from the Bureau of Analysis, the average age of industrial equipment in the United States has been rising since the ’70s and is roughly 10 years old, the equivalent of what it was during World War II. Certainly, as assets get closer to end-of-life, the likelihood for repair and possible failure increase.
“Most manufacturers still use a run-to-failure approach to maintenance. Reactive maintenance is the most expensive as on top of the cost of repairing the failed machine, there always is collateral damage to the entire production line,” said Eitan Vesely, co-founder and CEO of Presenso, which offers predictive maintenance for manufacturers.
Because the age of manufacturing machinery is rising, so too are the costs and risks for many manufacturers. Many companies do not precisely quantify or track their downtime costs, but it can rise into the tens of thousands of dollars per hour even for a small factory.
“81 percent of manufacturers are aware of the potential for machine learning to enhance maintenance, but only 17 percent have implemented programs to put those principles into action.”
Industry 4.0 is offering new capabilities to measure and quantify downtime, and to better monitor machinery to predict sub-optimal performance or failure. Sensors in machines can push data through advanced algorithms to uncover trends and issues faster than ever. Whereas a human many have a checklist and set of variables and measurements based on historical data, machines have complex algorithms that can consider hundreds of thousands of variables in near real-time and detect even the most subtle changes in a machine’s behavior that would never be detected by traditional monitoring.
A global study by Infosys found that while 81 percent of manufacturers are aware of the potential for machine learning to enhance maintenance, only 17 percent have implemented programs to put those principles into action. Many mid-market manufacturers have operational systems that are dating back to the ’80s and are even using manual data analysis with rudimentary programs like Excel, Vesely said.
Presenso is currently sponsoring a research project by Emory University, and researchers already have found that mid-market manufacturers are delaying investments in predictive analytics until they decide on an IoT strategy.
Regardless of the age of the equipment, much of the sensor data needed for predictive analytics already is being captured by many SCADA (supervisory control and data acquisition) systems, Vesely said. Many third-party vendors also may have access to data. In other cases, a more collaborative effort between operational technology and IT can offer access to more information. “As long as sensor data can be extracted, machine learning can be applied to any machinery regardless of age, vendor or machine type,” Vesely said. “The challenges often are more organizational than operational.”
Manufacturers that have yet to implement predictive analytics may question the return on investment. Implementing predictive analytics can call for capital investments in sensors and SaaS solutions among other things. Manufacturers should measure the performance of their analytics by the accuracy of results (number of correct alerts, false alerts, missed failure, etc…) and by the impact it has on downtime.
“Start with a proof of concept. If it demonstrates that a solution provides early warnings of asset failure, management buy-in is more likely. A cross-functional team that includes both IT and OT organization also is critical to deployment success,” Vesely said.