Rethinking predictive maintenance
Who is this presentation for?
- A technical audience, data science managers, machine learning practitioners, data scientists, and researchers
Predictive maintenance is surely one of the most talked-about topics in maintenance and asset management. Uptime improvement, cost reductions, lifetime extension for aging assets, and the reduction of safety, health, environment, and quality risks are some of many reasons why asset-intensive industries have plans for machine learning-based predictive maintenance.
Traditional approaches to predictive maintenance fall short in today’s data-intensive and IoT-enabled assets. Zaid Tashman dives into a novel machine learning-based approach to predicting the time of occurrence of rare events using mixed membership hidden Markov models. He outlines how these models were used to learn complex stochastic degradation patterns from data by introducing a terminal state that represents the failure state of the asset, whereas other states represent “health” states of the asset as it progresses towards failure. The probability distribution of these nonterminal states and the transition probabilities between states are learned from data.
Zaid explains how this technology was implemented to predict the failure of 3,000+ water pumps installed in Africa with a two-week lead time, and how the models allowed for the optimization of the maintenance schedule using stochastic optimization to maximize uptime and reduce cost.
- A basic understanding of artificial intelligence
What you'll learn
- Learn a useful methodology you can apply in your next project
- Discover how to approach predictive maintenance problems in any domain
Zaid Tashman is a R&D data scientist at Accenture Labs exploring new research problems in the areas of probabilistic programming, casual inference, and stochastic optimization. Zaid has a progressive experience in recommendation systems, customer behavior analysis, survival modeling, failure time prediction, hierarchical Bayesian networks, and anomaly detection. Previously, Zaid was a senior data scientist at ABB where he led the analytics efforts within ABB’s IoT platform serving all of their business units and a senior data scientist at Spacetime Insights, a Silicon Valley IoT startup where he successfully led and completed many machine learning projects in areas of predictive maintenance, anomaly detection, fraud detection, and optimization. Zaid holds a MSc in electrical engineering from Washington State University.
Comments on this page are now closed.
For conference registration information and customer service
For more information on community discounts and trade opportunities with O’Reilly conferences
For information on exhibiting or sponsoring a conference
For media/analyst press inquires