How Nvidia is harnessing AI to improve predictive maintenance

Follow along with VentureBeat’s coverage from Nvidia’s GTC 2022 event >>

The rapidly growing sectors of edge computing and the industrial metaverse were targeted by new technology developments, like sensor architecture, released by Nvidia last week at its GTC 2022 conference. Last week, the company also debuted the Isaac Nova Orin, its latest computing and sensor architecture powered by Nvidia Jetson AGX Orin hardware. 

Nvidia’s main focus is pursuing a tech-stack-based approach starting with new silicon to help manufacturers make sense of the massive amount of asset, machinery, and tools data they generate. In addition, predictive maintenance is core to many organizations’ Maintenance, Repair, and Overhaul (MRO) initiatives.

CEO Jensen Huang said during this keynote that  “AI [artificial intelligence] data centers process mountains of continuous data to train and refine AI models.” But, Huang continued, “raw data comes in, is refined, and intelligence goes out — companies are manufacturing intelligence and operating giant AI factories.” 

The complexities of predictive maintenance 

Accurately pursing predictive maintenance, repair, and overhaul (MRO) right is a complex, data-intensive challenge for any business that relies on assets to serve customers. MRO systems have proven effective in managing the life cycle of machinery, assets, tools, and equipment. However, they haven’t been able to decipher the massive amount of data in real-time that discrete and process manufacturers produce every day. 

As a result, IoT Analytics predicts that the global predictive maintenance market will expand from $6.9 billion in 2021 to $28.2 billion by 2026. Edge computing architectures, more contextually intelligent sensors, and advances in AI and machine learning (ML) architectures, including Nvidia’s Isaac Nova Orin, are combining to drive greater adoption across asset-intensive businesses. 

IoT Analytics advises that the key performance indicator to watch for is how effective predictive maintenance solutions are, how well they reduce unplanned operational equipment downtime

Not knowing what’s in that real-time data slows down how fast manufacturers and services companies can innovate and respond, further driving the demand for AI-based predictive maintenance solutions. Unlocking the insights hidden in real-time asset performance and maintenance data, whether from jet engines, multi-ton production equipment, or robots, isn’t possible for many enterprises today. 

Nvidia’s announcement of the Isaac Nova Orin architecture and enhanced edge computing support is noteworthy because it’s purpose-built for the many data challenges predictive maintenance has. The aircraft maintenance and MRO process is a perfect example, notable for its unpredictable process times and material requirements. As a result, airlines and their  services partners rely on  massive time and inventory buffers to alleviate risk, which further jeopardizes when a jet or any other asset will be available.   

Edge Computing is the future of predictive maintenance 

Nvidia has identified an opportunity in edge computing to update legacy tech stacks that have long lacked support for maintenance or asset performance management with a new AI-driven tech stack that expands their total available market. 

As a result, Nvidia is doubling down on edge computing efforts. Approximately one of every four sessions presented during the company’s GTC 2022 event mentioned the concept. CEO Jensen Huang’s keynote also underscored how edge computing is a core use case to the future of their architectures. 

IoT and IIoT sensors excel at capturing preventative maintenance data in real-time from machinery, production, and other large-scale assets. AL and ML-based modeling and analysis then happen in the cloud. 

For large-scale data sets and models, latency becomes a factor in how quickly the data delivers insights. That’s where edge computing comes in and why it’s predicted to see explosive growth in the near future. Gartner predicts that by 2023, more than 50% of all data analysis by deep neural networks (DNNs) will be at the point of capture in an edge computing network, soaring from less than 5% in 2019. And by year-end 2023, 50% of large enterprises will have a documented edge computing strategy, compared to less than 5% in 2020. As a result, the worldwide edge computing market will reach $250.6 billion in 2024, attaining a compound annual growth rate (CAGR) of 12.5% between 2019 and 2024.

Of the many sessions at GTC 2022 that included edge computing, one specifically grabbed attention in this area: Automating Industrial Inspection with Deep Learning and Computer Vision. The presentation provided an overview of how edge computing can improve manufacturing performance with real-time insights and alerts.  

Real-time production and process data interpreted at the edge is proving effective in predicting machinery repair and refurbishment rates already. 

Edge computing-based models successfully predicted yield rates for the resin class and machine combination. 

Streamlining predictive maintenance

Nvidia sees the opportunity to expand its total available market with an integrated platform aimed at streamlining predictive maintenance. Today, many manufacturers and service organizations struggle to gain the insights they need to reduce downtimes, further expanding the total available market. 

For many providers that sell the time their machinery and assets are available, predictive maintenance and MRO are central to their business models. 

As asset-heavy service industries, including airlines and others, face higher fuel costs and more challenges in operating profitably, AI-based predictive maintenance will become the new technology standard. 

Nvidia’s decision to concentrate architectural investments in edge computing to drive predictive maintenance is prescient of where the market is going.