Machine Learning Decision Support System for Event/ Object Detection

Misty Blowers, Datalytica ABSTRACT: Detection of specific events in real-time longitudinal data is often critical for industrial optimization. This lecture will focus on the development of a machine learning capability to minimize the downtime of a paper machine in a recycling box plant that results from paper breaks. This novel approach generated significant value and was tailored to a specific environment that assists a company in making consumer grade product from post-consumer waste. This value was realized despite considerable variability and noise inherent to the operational environment. By fusing historical operator logs that attributed the causality of the breaks in a way that help maximize the probability of detection while minimizing false alarm rates, a machine learning enabled decision support tool provided 25-30% improvement on paper-break detections over traditional threshold based alerting methods. Similar approaches can be used to improve production across a wide-variety of industry sectors.