Get started Bring yourself up to speed with our introductory content.

Building the predictive maintenance 4.0 era with cognitive-first models

Predictive maintenance has come a long way since the ’90s. Over the past decade, IIoT and the evolution of analyzing sensor information has pushed organizations to look at new ways to use data to understand machine health.

A new wave of data sets is being generated by and collected through a new generation of connected machines and sensors. With them, organizations are now able to make better and informed operational decisions; perform timely maintenance to minimize failures, unplanned downtime and associated risks; cut costs; reduce scraps and rework; improve productivity and quality; manage inventory and resources better; and deliver an overall superior customer experience.

Recent advancements of artificial intelligence and machine learning applied to IIoT enable organizations to track multiple components of heavy machines and identify highly localized and contextual signals that pinpoint potential causes for concern ahead of time to prevent a catastrophic event. With deeper insight into machine performance and minute fluctuations, changes and errors that were once undetectable by the human eye can now be observed, recorded and analyzed on a real-time basis to detect potential business threats.

We see more organizations tapping into the IIoT potential to predict equipment failure. In fact, IIoT-driven digital transformation is no longer considered just a strategic advantage, but an essential survival tactic for the highly competitive business environment of the future. But there are still many hurdles they have to overcome to truly enjoy the promises of Predictive maintenance.

Industrial IoT: The data challenges

Despite enterprises viewing IIoT as a game-changer, many still struggle to harness it effectively to get the optimal performance out of their machines.

First, legacy machines can hinder businesses from capitalizing on the IIoT boom. Intelligent sensors can be added to bridge the gap, but it is important to have the key components properly instrumented to capture critical signals and anomalies that indicate future problems.

The second challenge lies in how organizations analyze the information collected. IIoT gives businesses all sorts of unique information about how their machines function. This is great in theory, but that information has to be mined out from the raw data. Applying old and traditional approaches of data analysis in this new world of IIoT won’t tap into the true potential.

Many teams take a machine learning approach called supervised learning. It can build models trained on historical failure or fault data. Other teams simply write rules to generate alerts and notifications when past failures or faults are primed to reoccur. But these approaches only accentuate what many businesses are already good at: learning from historical problems and putting processes to prevent them from repeating. However, only about 20% of the problems that occur in the field are repeat problems. Eighty percent of those are actually new and unknowns for most industries. Unless the signs for these 80% of the problems are detected early on, predictive maintenance won’t yield the desired results.

Cognitive-first strategies to service the predictive maintenance 4.0 era

Predictive maintenance consists of collecting data on the condition of machinery so that engineers can foresee possible outages or failures before they even happen. Predictive models require large amounts of performance and condition history data to learn from before they can make forecasts for equipment accuracy.

Time may also be needed to analyze live data to be able to establish normal baselines for performance for each and every asset individually based on their operating and environmental conditions. In order to have predictive maintenance working as a practical tool, months of data analysis and investment in data scientists and a production data engineering team are often required to build and update models. But more often than not the scale of these analyses is such that it isn’t practical to do it manually, especially in a production environment.

To make the most of predictive maintenance, companies can combine connected technologies with cognitive machine learning. The cognitive approach will allow them to automatically find data patterns or groupings at scale that otherwise would be hidden and would have been humanly impossible to unearth in time to prevent major issues.

By using IIoT data through cognitive anomaly detection and prediction, businesses are better equipped to identify the unknown variables that are responsible for many of the defects that lead to recalls or failure events in the 80% of new events. This meta-learning approach increases prediction quality and accuracy by about 300% and requires 3% of the time and resources to get up and running compared to other approaches to data model development.

Promising use cases

Organizations in the manufacturing, oil and gas, automotive, aviation, transports and logistics as well as energy and utilities space are starting to lead the way in this field.

The work we are doing with leading global organizations in these industries show that the new generation of predictive maintenance boost overall equipment effectiveness by 90% and reduce unplanned downtime by 20%. They optimize inventory costs by 35% and boost overall equipment effectiveness by 35% and asset life by 25%.

We recently helped a Fortune 50 company to identify 66 times more anomalies in their commercial HVAC systems using the cognitive approach as compared to their traditional approaches and thereby prevent downtime. This translated into increased customer satisfaction.

Gartner forecasted there will be 20.8 billion connected things worldwide by 2020. Organizations that stick to an old preventive data methodology won’t be able to uncover the potential of all these connected things and will likely be left behind. Using cognitive-first models is what will make the difference.

All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.

Data Center
Data Management