Optimizing asset maintenance through the Internet of Things and machine learning
July 28, 2017Dr. Marc Sol, Principal Scientist, Infor Dynamic Science Labs
If your company owns a large fleet of expensive or mission-critical assets like trucks, pumps, or complex machinery, you are likely looking for ways to minimize the risk of negative business impacts due to asset failures. Companies like yours have, over the years, experimented with several asset management strategies. Expensive and critical assets are generally put under a preventive maintenance plan, in which the asset is inspected at a time- or usage-based schedule, often suggested by the asset's manufacturer. Enterprise asset management software supports you in executing such maintenance plans for your assets.
Preventive maintenance according to a fixed schedule has some foundational shortcomings. Assets of the same type, used under different circumstances, may require condition-based maintenance (CBM) to fully address their unique operating requirements. Condition-based maintenance presents a challenge in that it requires scheduled inspections to assess the asset's condition, which consumes time when the equipment could have been productive. Additionally, the act of stopping and opening an asset increases the risk of damaging an asset component, leading to future failure. In this way, assets still suffer from unanticipated downtime under either type of maintenance schedule, negatively impacting the operational cost and service level of the organization.
Wouldn't it be great if we could see when an asset was about to fail and optimize the maintenance schedule based on the condition of the equipment, while keeping the equipment in operation? What if maintenance engineers could focus on assets with a high failure probability before the next scheduled maintenance shutdown of the entire plant? When repairing an asset component, wouldn't engineers want to inspect other nearby at-risk components, thereby reducing travel time and cost? Building on top of a solid enterprise asset management foundation, two exciting recent developments may help achieve these goals: the Internet of Things (IoT), and machine learning.
Deeper data analysis
IoT brings sensor data to the cloud, ready to be analyzed. Sensors can measure temperature, pressure, flow rates, vibrations, voltage, or electric current, at various points on an asset. But sensors might also relay the geo-location of a moving asset, or exception events that were caught in the embedded software of the asset. Regardless of what is being measured, that information can be processed and stored in an environment that allows for deeper data analysis.
Assets may be huge complex structures (as large as an entire plant or railway) containing tens of thousands of sensors. Infor's Enterprise Asset Management solution already lets you model such assets as a hierarchical structure of components. In such situations, one is generally interested in optimizing the maintenance plan at a component level.
Infor Dynamic Science Labs is developing ways to add even greater sophistication to the asset management process. Our new Infor IoT platform can enhance Infor EAM's asset modeling capabilities by registering sensors to specific asset components. Not only does this make it easy to manage a large volume of sensors, it also brings together the readings of related sensors for predictive analytics.
The Infor IOT platform will also include machine learning capabilities, which enable pattern detection in data that can be used for classification and prediction. If you've seen face recognition software, you already have an idea of what machine learning can do: after training an algorithm with pictures of people's faces, together with the names of those people, the software can identify individuals in a new photo. The software has learned specific facial patterns and identifies which of those patterns best matches a new photo.
We can think of the set of sensor readings for an asset like a picture, in that the readings contain specific patterns that identify the asset. We can relate sensor readings to past work orders (from an enterprise asset management system) that tell us when the asset, or asset component, has failed. For each asset, we can now build a collection of "photos" consisting of the sensor readings for a point in time, annotated with the time until next failure. A machine learning algorithm can be trained on this data, and software can use this algorithm to make real-time predictions about future asset failure. These predictions can then be used to optimize the maintenance schedule. They'll also give you information that can help you improve the efficiency and reliability of critical assets.
Infor Dynamic Science Labs is currently working on training these algorithms for different types of assets and working closely with the Infor IoT and Enterprise Asset Management teams to create an even more powerful predictive maintenance solution.
- CloudSuite Distribution Enterprise
- CloudSuite EAM
- CloudSuite Facilities Management