Predicting the Internet of Things with Tensorflow.js
summary
Uses vibration sensors and Tensorflow.js to predict machine failure by turning high-frequency XYZ accelerometer streams into image-like tensors and training convolutional neural networks for pattern detection. Requires lots of labeled running and breaking data, faces brittle models when hardware or mounting changes, and needs periodic retraining. Training starts near 17% accuracy, climbs through ~80% and can hit ~99% on a focused case. Deployment exports the model as a file for cloud serving or edge inference milliseconds to stop machines locally.
AI-written summary · about
chapters (28)
- 0:00Intro
- 0:50Life Cycle
- 2:03Predictive Maintenance
- 3:18Sounded Vibration
- 3:58Forecasting
- 5:41Factory Demo
- 6:42Machine Learning
- 7:11Google
- 7:35Wikipedia
- 7:50Google Deep
- 8:35Karis
- 9:21Carrots
- 10:11Training the model
- 11:23Good or bad
- 12:26Asking the right questions
- 13:18convolution networks
- 13:58handwriting examples
- 14:34code
- 15:24model
- 15:44training
- 16:14big data
- 16:24accuracy
- 17:02results
- 19:51tinderflow serving
- 20:22edge computing
- 21:30training and retraining
- 22:21developers to data scientists
- 22:58data transformations
related talks
54mSmarter Everything with ML & the IoT
18mPredictive Network: How AI Forecasts System Failures
38mReal-world Machine Learning with TensorFlow and Cloud AI
55mUsing TensorFlow for Performing Deep Learning at the Edge
21mA Journey into the World of Machine Learning with TensorFlow.js
32mUsing Machine Learning Models in JavaScript Application
20mTensorFlow.js Bringing Machine Learning to the Web and Beyond
08mMachine Learning in Node.js using Tensorflow.js