Escaping the Reactive Maintenance Rut Through AnalyticsWinn Hardin | May 12, 2016
The ability to forecast when a machine might go down is the Holy Grail of industrial maintenance. Using data collected from sensors within the equipment, the plant engineer determines the best course of action to address the troublesome component or parameter before it leads to asset failure. But many plants hesitate to start the journey toward predictive maintenance.
The people responsible for making decisions about machine maintenance “spend much of their time firefighting or troubleshooting, or maybe they’re understaffed or being forced to cut expenses,” says Melissa Hammerle, business unit manager for Fluke Connect, a maintenance management system of software and logging devices. “The thought of setting up a preventive maintenance program seems like a bridge too far.”
Predictive analytics software helps plants and facilities break out of the reactive maintenance rut by monitoring the condition of equipment through sensors, analyzing the data and recommending the appropriate action. Despite the data-intensive nature of predictive analytics, software developers have designed programs that are generally easy to use and understand — meaning that decision-makers don’t have to rely on a data scientist to interpret the reams of information on equipment performance.
Preventive vs. Predictive Maintenance
The use of predictive analytics for industrial maintenance goes by many names: predictive asset management, asset performance management, condition-based maintenance and machine failure detection. In fact, confusion exists around the definitions of preventive and predictive maintenance.
Sometimes the two terms are used interchangeably. But more often, preventive maintenance constitutes calendar-based tasks typically managed in an enterprise asset management system. Although preventive maintenance is a necessary component of a comprehensive maintenance strategy, it doesn’t take into account how the equipment has been operated and under what conditions, says Sean Gregerson of Schneider Electric, which provides Avantis PRiSM predictive asset analytics software.
Predictive maintenance, on the other hand, actively monitors equipment conditions such as energy use, vibration and temperature. Predictive analytics software studies this recent and historical data to forecast future outcomes for assets. “It considers a lot of possibilities that could be responsible for failure,” says Olivier Jouve, director, IBM Watson IoT Industry Solutions and Predictive Maintenance and Quality.
The goal of predictive analytics is to identify pending asset degradation or failure as early as possible. Early warning gives plant engineers the time to evaluate risk and improve maintenance allocation. Instead of fixing things as they break, technicians “can plan corrective action without impacting production,” Jouve says.
This also allows companies “to realize less downtime due to surprises and shorter outages that are better planned,” says Chad Stoecker, leader of managed services for GE Digital. Additionally, predictive analytics can lead to more accurate spare-parts management, less waste due to unnecessary maintenance actions and improved production capacity.
Pulling the Trigger
Implementing a predictive maintenance program often constitutes a culture change, which Hammerle calls the “most significant barrier to adoption” as plant engineers move from a reactive to predictive maintenance mindset.
A common first step is prioritizing which assets get monitored and to what degree. “Oftentimes people think about putting sensors on their most expensive equipment, but they should also monitor the assets most critical to their output,” Hammerle says.
In many cases, facilities will launch a predictive maintenance program on industrial assets already equipped with sensors. “Taking data from the machine and sending it to cloud is quite easy and inexpensive if you have the right software to support it,” says IBM’s Jouve.
Before deploying its predictive maintenance software, IBM will send subject matter experts (SMEs) to walk the plant with personnel — including those responsible for fixing the machine when it goes down — to identify problem areas.
The SMEs review various processes, equipment, and types and frequency of failures on the production line. By understanding how machines fail, software developers can build analytic models specific to the equipment.
Once decision-makers start to see benefit from a predictive approach on their highest-value assets, they tend to add more advanced sensors and therefore expand diagnostic capability. As wireless sensor technology continues to improve and more options hit the market, “it is going to become even easier to deploy smart sensors with limited investment,” Schneider Electric’s Gregerson says.
For some plants, however, condition-monitoring tools still represent a significant expenditure. But not every piece of equipment needs a fixed sensor, either. Semi-permanent sensors and wireless test tools such as multimeters and infrared cameras enable facilities to conduct studies for a set period of time.
As Fluke’s Hammerle says, equipment monitoring should be accessible to all, “not something you only have on your most critical assets or well-funded production plants.”
Just as predictive analytics promises to increase maintenance efficiency, the software solution itself should be easily deployed and intuitive for users. The objective of the software is “to transform raw sensor data into actionable information so that reliability and maintenance personnel can easily understand what is causing the early warning alert condition,” Gregerson says.
Additionally, features within the software platform such as an asset health dashboard, graphing and key performance indicators (KPIs) help technicians and engineers visualize data in context. Despite the focus on ease-of-use, plants still need to take the time — and expense — to train personnel.
As more maintenance workers retire and take their tribal knowledge with them, plant engineers must find ways to educate younger personnel on operating and maintaining equipment. One way to do this is by establishing a case library in the software of early warning events.
Such a feature allows companies to capture information “so that people who are less knowledgeable about these assets can compare previous alert fault conditions to current ones and help guide them to the best decision,” Gregerson says.
Software designers are quick to point out that predictive analytics forecast what might happen in the future, not what will happen. Predictive maintenance also requires an investment in software, sensors and training for maintenance personnel. With the proper implementation, however, predictive analytics can prevent unplanned downtime, reduce maintenance costs and extend equipment life.
Says GE’s Stoecker: “Those who move to data-driven maintenance programs will realize significant performance improvements and financial savings.”