Break | AIChE


Industrial data and analytic needs are unique. First, digital teams need to build robust data pipelines that connect data sources such as plant historians or control systems across varying network topologies, communication protocols and scalability issues to analytic platforms. Then, data engineers need to persist these high volume and velocity data feeds, build workflows that integrate disparate data sources operating at different resolutions and address several data quality concerns that originate from root causes across the data ecosystem. Finally, typical industrial machine learning problems have significantly smaller number of trainable examples – for instance, the number of equipment catastrophic failures or safety incidents are usually a tiny fraction of normal events. This leads to the non-trivial problem of building models that generalize well using these biased datasets.

In this talk, we will focus on strategies that industrial companies can adopt to address the above challenges. Using real-world examples, we will elaborate on how industrial companies can simplify the entire analytic life cycle from data connectivity to model deployment with rinse and repeat design patterns, allowing delivery of value at scale,. We will also discuss some of the common pitfalls in industrial digital transformation initiatives and how to avoid them.