I recently presented Building a Scalable Telemetry Based Multiclass Predictive Maintenance Model in R at the ICDSE conference. This conference was inter-disciplinary where the attendees were primarily from academia and shared their scholarly research and innovation. Due to the nature of the conference, the focus was on the methodology used to solve their domain-specific problem rather than the tooling needed to solve a large-scale problem.
My talk at the conference was focused on outlining how a user or an organization would build a Scalable Telemetry based Predictive Maintenance Model. To set the context, I described how we routinely come across IoT devices with sensors embedded all around us, which collect a lot of telemetry data over time. Then the natural next question was on how this data can be used to address business questions like, "When is my device going to fail?" Some tips on how the raw sensor data can be enhanced with additional machine related data and how to formulate and build a reasonable ML model were briefly discussed during the talk.
Finally, typical scenarios for an on-premise and cloud based solution was outlined with focus on SQL Server R Services and Azure Machine Learning Studio, as well as jupyter notebooks as example tools to develop and operationalize these models. To accompany my oral presentation, I wrote a short paper which describes the methodology in more detail. The audience was intrigued with the solution and hoped to use such a similar technique for the healthcare domain.
Quelle: Azure
Published by