ThingWorx Analytics Training: Module 9 Part 1
This video begins Module 9: Anomaly Detection of the ThingWorx Analytics Training videos. It describes how Thingwatcher can be set up to monitor values streaming from connected assets, and send an alert if its behavior deviates from its 'normal' behavior.
This video does not contain any hands-on exercises, and you don't need access to a running ThingWorx Analytics environment.
This video has chapters — use the chapter menu in the player to jump to any section.
Chapter Summaries
0:00 – Introduction: ThingWorx Analytics Module 9 – Anomaly Detection, Thing Watcher & SPC
Kicks off Module 9 with a roadmap covering the ThingWorx Analytics Thing Watcher feature, a hands-on alert configuration exercise, statistical process control (SPC) concepts, and how the SPC accelerator jump-starts building an SPC application.
0:32 – ThingWorx Analytics Server: Builder, Manager & Codeless Model Building
Explains that the ThingWorx Analytics Server powers predictive signals, profiles, machine learning algorithms, prescriptive scoring, clustering, and time series analysis, while Analytics Builder and Manager provide a codeless, point-and-click way to build or import models.
1:14 – Thing Watcher & Anomaly Detection: Supervised vs. Unsupervised Outlier Methods
Introduces Thing Watcher as a pattern-detection machine learning service and defines anomaly (outlier) detection, contrasting unsupervised methods that flag dissimilar data points against supervised methods that require a labeled set of anomalies.
2:27 – Thing Watcher in ThingWorx 8: Real-Time, Self-Learning Anomaly Detection
Describes Thing Watcher as native to ThingWorx version 8, finding anomalies in real time, automatically learning an asset's normal patterns, and collecting its own data with no historical dataset or predefined rules required.
2:55 – Sensor Data & Why Value Thresholding Misses Anomalies
Uses a farm tractor's sensor readings—CO2 emissions, fuel level, oil pressure, and fan speed—to show how subtle CO2 fluctuations slip past simple value thresholding, and where Thing Watcher's predictive model catches them.
3:50 – Five Thing Watcher States & Training/Results Microservices
Walks through Thing Watcher's five states—initialized, calibration, training, buffering, and monitoring—and explains the training and results microservices that build a three-layer neural network and score streaming property data.
5:19 – Model Training Workflow: 80/20 Train-Validation Split & Look-Back Sizes
Details how Thing Watcher splits calibration data into an 80% training and 20% validation subset, trains multiple time series models with different look-back sizes, and picks the best-validated model to compare predictions against actual values.
6:28 – Thing Watcher Terminology: Priority, Look-Back Window, Temporal Features & Certainty
Defines key Thing Watcher settings: priority (alert severity), the automatically determined look-back window, temporal features derived from streaming data, and the certainty threshold that controls how frequently anomaly alerts are raised.
7:18 – Rate, Binning, Interpolation & the Failed State
Covers the rate and outbound anomaly rate settings, binning of streaming values into the closest expected timestamp bins, and interpolation that averages up to three consecutive missing values before Thing Watcher enters the failed state.
8:15 – Three-Layer Neural Network Algorithm & Thing Watcher Requirements
Explains that the model is a three-layer neural network using automatically engineered features and an optimal look-back, and notes limits: one sensor at a time, multiple alerts per application mashup, and a 580 property-value minimum.
9:37 – Flowserve Demo Setup & the IIoT Business Case
Introduces a live demonstration built from Flowserve, National Instruments, and Hewlett-Packard Enterprise equipment, framing the IIoT business case with Gartner's forecast of 26 billion connected devices and the value of actionable intelligence.
11:05 – Live Pump Rig Demo: Thing Predictor & 30,000 Readings/Second
Tours the Flowserve pump rig retrofitted with six National Instruments sensors and HPE computing, showing Thing Watcher's automated edge analytics, Thing Predictor's failure forecasts, and self-evaluation at 30,000 readings per second per sensor.
13:22 – Simulating Pump Failures: Real-Time Alerts & Automated Notifications
Demonstrates ThingWorx Analytics reacting in real time as a closed valve reduces flow and a pump misalignment is introduced, flagging dropping efficiency and impeller risk while triggering alerts and an automated technician phone call.
15:13 – Conclusion: PTC Predictive Service & ThingWorx Machine Learning Value
Closes by positioning Thing Watcher within PTC's predictive service solution, highlighting how the ThingWorx platform lets machines teach themselves to predict, detect, and even help fix production hangups with out-of-the-box simplicity.