Predictive Maintenance with Apache Flink

on YouTube ↗41m

" summary

Describes a deployed Predictive Maintenance product called metatron PdM that runs a DNN trained with Keras on TensorFlow/Theano against live sensor streams. Explains the Flink runtime built with Flink DataStream API using event time, sliding windows, timestamps/watermarks, custom sources/sinks and checkpointing for real‑time inference. Compares JVM integration strategies—embedded CPython, TensorFlow Java API, RPC to TensorFlow Serving—and covers delivery with the official Flink Docker image and Flink metrics for monitoring.

AI-written summary · about

" chapters (6)

1. 0:00Introduction to talk
2. 3:49Apache Flink overview
3. 12:25Predictive maintenance demo
4. 24:26Real-world use cases
5. 33:05Challenges and solutions
6. 39:16Q&A and closing

" related talks

• 15mData science at scale using Apache Flink: Dynamic model serving and real-time feature generationShashank Agarwal · Flink Forward Global Virtual 2020
• 35mIntroducing Tensorflow on FlinkXuannan Su · Flink Forward Global Virtual 2020
• 20mUsing Flink for Versatile Feature Engineering PipelinesMichal Ciesielczyk · Flink Forward Global Virtual 2020
• 53mApache Deep Learning 302Timothy J Spann · ApacheCon@Home 2021
• 21mReal-time Stream Analytics and Scoring Using Apache Flink, Druid & CassandraCiesielczyk, Zontek · Flink Forward Europe 2019
• 39mOperating Deep Learning Pipelines Anywhere Using KubeflowJörg Schad, Gilbert Song · KubeCon + CloudNativeCon China 2018
• 28mEnabling Machine Learning with Apache FlinkSherin Thomas · Flink Forward Europe 2019
• 25mReal-Time Fraud Detection at ScaleSatyanandam Kotha · Conf42 Machine Learning 2025