There’s a lot of buzz around Artificial Intelligence and Machine Learning, as supply chain leaders look for the killer app within their organizations. One promising area has been demand planning. Another has been MRO, as companies look to move from preventative to predictive maintenance. The idea is that by putting sensors on motors, gears and other critical components that measure conditions like heat, voltage output and vibration, a technician can better predict when that piece of equipment might fail. If it works, PM’s could become a thing of the past.
So where are we? That was one of the questions I put to Phil Jones, Target’s director of supply chain engineering, and Phil Gilkes, a regional maintenance manager with Dollar Tree, during a symposium put on by the National Center for Supply Chain Automation. The framework for that question was an asset management maturity model slide that Jones put together, illustrating the progression from an MRO organization at Level 0 – Survival Mode, where equipment runs to failure and repairs are done as problems occur to Level 4 – Predictive Maintenance, where AI and ML are utilized to analyze events to predict the timing of future issues and schedule maintenance.
According to Jones and Gilkes, Condition Based and Predictive Maintenance is the goal for both organizations, but realistically, both organizations are somewhere between Level 0, with equipment that runs to failure, Level 1 – Calendar Based Maintenance with time-based PMs and Level 2 – Usage Based Maintenance, where PMs are based on run times or the number of hours on a piece of equipment. Both organizations are investigating the first step to get to condition based maintenance, which is putting sensors on machines to monitor conditions, but neither was there yet – or not there beyond piloting. And remember that these are two large organizations with a network of distribution centers and experience with automation.
One of the challenges for the MRO industry to get to that Level 4 – Predictive Maintenance is going to be data. In order to produce reliable and actionable results, AI and ML need data and lots of it. Otherwise, the risk is that the maintenance system “will start flagging issues that aren’t really issues,” noted John Sorensen, senior vice president of lifecycle performance services at MHS. “You don’t want technicians and maintenance managers to think that the system cries wolf.”
Where then does a progressive maintenance team start. Sorensen and Rob Schmidt, MHS’s senior vice president of distribution and fulfillment, both recommended a crawl, walk, run, sprint approach similar to the adoption of any new technology.
Don’t try to put sensors on every motor and gear in a facility, which can number more than 1,000. Rather, start by categorizing components and equipment according to their criticality to the operation. Running until it breaks might be appropriate to some pieces of equipment, especially if spare parts are in inventory and the equipment is easy to fix. A limited number of sensors might be appropriate on items that are more critical, like a PLC. And finally, a broad array of sensors that can begin to gather important operating data in bulk can be deployed on mission critical items where reliability counts. At the same time, added Sorensen, “you might just need a data set of 200 sensors to begin the journey.”