Changing the Economics of Industrial Operations with Petuum Industrial AI
By Leena Joshi
Over the last couple of decades, there have been significant strides in what can be achieved with artificial intelligence, but realistic applications of advanced AI in industrial operations have been limited by a lack of practical approaches to marrying best-in-class technology with the in-depth domain expertise and data needed to solve real-world problems effectively.
This is among the things that make our announcement today meaningful. The Petuum Industrial AI Autopilot product incorporates a pragmatic standardized approach to solving a range of industrial use cases, helping industrial customers achieve substantially higher levels of automation, and in the process, change the economics of running complex manufacturing businesses.
While it is common to hear about the use of AI for point use cases such as anomaly detection or predictive maintenance in the industrial sector, it is noteworthy that our product does not just incorporate the common, well-known use cases but also autosteers entire end-to-end operations in a supervised autopilot mode. This was a milestone achievement both for our customer and us as it reduces operator training time and increases the operators’ understanding of their operations.
Why is this important? In industries and processes where there is a great deal of variability from operator-to-operator or day-to-day in yield, throughput, and energy consumption, traditional techniques often fail to capture non-linearities or temporal/long-range patterns. This is where Petuum’s sophisticated approach makes a real-world impact. We took an approach where the models in our product learned the true patterns of process parameters, staying robust to noisiness or outliers in the data. The effectiveness of this approach showed during implementation, where our customers increasingly began to rely on our product to comprehend their operation better and to increase the number of “golden days” for their site.
Reducing variability and increasing standardization causes throughput and yield metrics to rise as process stability increases and shutdowns are avoided. Costs go down as the system uses more alternative cheaper fuels, without compromising product quality or process stability. Sustainability improves as emissions get reduced. In effect, advanced artificial intelligence that used to be accessible to only a few starts to influence and solve practical, real-world problems of balancing convoluted tradeoffs, faced by several industries. The new levels of automation and intelligence result in a sea-change in the underlying economics of these industries — and the change has only just begun!