Abstract: To model complex industrial machineries and systems for failure prediction, identifying relevant data is often necessary. Mostly all features are not always equally valuable. Apart from equipment data, it’s sometimes possible to infer failure from seemingly extraneous data, such as downstream and peripheral equipment.
Often domain knowledge about data capture and transformation processes at the sensors can be acquired through thorough exploratory data analysis. Finally, domain expertise frequently becomes most relevant in the interpretation of insights gained using machine learning and deep learning techniques about the nuances of complex physical processes. In this session, we will discuss how data science techniques complemented with domain expertise can be effective in modeling industrial systems.
Next, we’ll discuss examples of how statistical significance does not guarantee practical significance, but to be practically significant a model must be statistically significant.
Bio: Anshuman has a background in equipment anomaly detection with more than six years experience in analyzing industrial machinery data and working on models to monitor equipment health and predict equipment failures. He has pursued his masters degree in computer science recently focusing on machine learning from John Hopkins University. More than six years ago Anshuman started working at a predictive analytics firm in Maryland specializing in industrial asset management industry as a 'domain expert.' But soon, he started getting involved in statistical analysis and mathematical modeling side of product development. Prior to that, Anshuman worked in the maritime and manufacturing industry as an engineer for a few years and has been actively involved in asset management projects.