Back to Operations Blog

How deep learning could supercharge preventive maintenance—and many other operational-improvement levers, too

Deep learning

by Mehdi Miremadi

Preventive maintenance has long been an important tool—not only for manufacturers, but also for companies across a wide range of sectors that use sensors and other Internet of Things (IoT) technologies to monitor equipment, improve uptime and detect anomalies, and increase the useful life of components. But recent artificial-intelligence (AI) breakthroughs, particularly in deep learning using artificial neural networks, are reimagining what preventive maintenance and many other traditional operational levers can achieve. Our research, published as the McKinsey Global Institute discussion paper Notes from the AI frontier: Insights from hundreds of use cases, confirms the potential.

Neural-network techniques loosely model the way neurons interact in the brain. Accordingly, to function well, neural networks particularly deep learning models, require massive amounts of data, from a variety of sources such as audio, video, or still images. For preventive maintenance, layering in such new data can enhance and possibly replace more traditional methods, leading to significant improvements in uptime.

Drawing on McKinsey Analytics expertise, we analyzed more than 400 AI use cases across 19 industries and nine business functions, from supply-chain management to HR, looking both at practical applications and potential value. Preventive maintenance emerged as one of the main ways these advanced AI techniques can be deployed across the economy.

In some of the industry examples, we found that using remote, AI-based onboard diagnostics to anticipate the need for service could generate significant value. In air cargo, for example, deep learning techniques integrated model data for a particular plane together with its maintenance history, images and video of its engine condition, and data generated from IoT sensors monitoring engine vibration anomalies. The resulting insights extended the plane’s lifespan well beyond what would be possible using traditional analysis.

There are some obvious caveats. Obtaining massive sets of well-labeled data is a challenge for many companies, and the organizational complexity involved in deployment of deep learning systems cannot be underestimated. As well as developing strategies to create or acquire training data and the computing power to process them, firms will need capabilities to ensure that the insights they harvest translate into action, by machines or by humans.

Yet the effort is worthwhile. Our use cases highlight the very real value that can already be created with these most advanced AI techniques—and we are only at the start. Successful adoption will require focus and the setting of priorities, but as the technologies themselves advance, the value they generate will inevitably grow. Identifying where and how that value can be captured looks likely to become one of the key business challenges of our technology-infused era.

Read the full MGI report “Notes from the AI frontier: Applications and value of deep learning