Predictive Maintenance with AI
Industrial equipment is typically serviced on a fixed schedule, irrespective of actual operating condition, resulting in wasted labor and risk of unexpected and undiagnosed equipment failures. Once instrumented with sensors and networked with each other, devices can be monitored, analyzed, and modeled for improved performance and service. An industry leader in the space, GE enables manufacturers to create “Digital Twins”, or physics-based virtual models of large-scale machinery, on their industrial cloud platform, Predix.
“Twinning” a piece of equipment allows human operators to constantly monitor performance data and generate predictive analytics. According to Marc-Thomas Schmidt, Chief Architect of Predix, nearly 650,000 twins are currently deployed and range widely in complexity. Complex twins like those of gas turbines interpret data from hundreds of sensors, understand failure conditions, track anomalies, and can be used to regulate production based on real-time demand.
Even relatively simple twins yield clear business benefits. Schindler, an elevator manufacturer, makes the bulk of their revenue on servicing costs, not asset sales. Operating a crew of service engineers on a fixed schedule is an inefficient use of labor. Instrumenting elevators with simple sensors and twinning them with Digital Twins enables Schindler to send service on a need basis rather than a time basis.
Instrumentation and digitization is not entirely straightforward in manufacturing, as we’ve previously shown. Schmidt explains: “Instrumentation is hardest when equipment is very remote. On offshore drilling platforms, for example, the biggest challenge is getting the data back to a place where it can be analyzed.”
Automated Quality Control
Faster feedback loops enable factories to tackle unplanned downtimes, low yield (% of units that pass quality control), and low productivity (time it takes to make a product). “Issues with low yield are most acute around high complexity products – like a laptop where there are a ton of various systems that need to come together perfectly for the product to work,” explains Plethora founder Nick Pinkston.
Pinkston also points out that productivity often trades off against yield. The faster a manufacturer pushes a process, the more likely they’ll hit errors and low quality output. “Better monitoring and adaptive control can allow you to increase the productivity of a single machine, and likewise better overall system monitoring and planning can allow the overall system to produce more product on the same numbers of machines.”
Rather than rely on humans for in-process inspection and quality control, a task that’s increasingly more challenging due to exploding product variety, companies like Instrumental.ai leverage cameras powered by computer vision algorithms to triage defects immediately and identify root causes of failure. Performing anomaly detection on hundreds of units in seconds, rather than hours, enables manufacturers to identify and resolve production failures before expensive delays pile up.