From Noisy Sensors to Events
summary
A production Python pipeline smooths noisy weight sensor streams with a Kalman filter (filterpy) then infers discrete phases with a Hidden Markov Model (hmmlearn), producing cycle counts, per-cycle tonnage and a dashboard. The metric claims median error ≈ 0 for the new method versus ~20% bias from manual thresholding (business assumed ~10%). The implementation runs online, handles missing samples by prediction, and yields near‑perfect per‑cycle payload estimates in the demos.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:29Raw sensor data challenges
- 5:53Weight sensor data processing
- 9:32Noise removal algorithm design
- 12:26Kalman filter for noise removal
- 18:55Hidden Markov models for state detection
- 25:30Kalman vs Hidden Markov models
- 27:15Results and business impact
- 29:04Q&A
related talks
50mClosing the Machine Learning Loop using Python
01h 21mBuilding a ML Model Observability pipeline
24mWhat's the Buzz with Machine Learning
17mFrom Sensor to Value: A journey
47mIPython and Sympy to Develop a Kalman Filter for Multisensor Data Fusion
28mReal Time Sensor Anomaly Detection with Sci Kit Learn and the Azure Stack
01h 23mBuild a production ML system with only Python on free serverless services
46mAI-Driven Techniques for Noise Filtration in Software Validation Logs