Network Operations Center · Monday 06:00

TimeH2O watches every reading on your cell sites and finds faults before a site fails.

What TimeH2O doesIt forecasts the normal range of every reading. When a live reading leaves that range twice in a row, it raises an alert hours before the site fails.

Customers firstfind most faults, and your team last
1,200 linesto watch every 15 minutes
42,000 peopleoffline when one busy site fails

One fault, two ways

Site 117 · the same battery fault on the same day

The app uses sample data made for this example. The numbers show how the steps work. They are not results from a customer. A fixed limit is the simple baseline used here; compare it with your own alert rules before you rely on the gap.

Network fault detection · Network Operations Center

Find network faults hours before a cell site fails, not when customers call.

The problemA fixed alarm limit fires late, when the site is already down. A fault starts earlier, as a reading that leaves its normal range (anomaly detection).

What TimeH2O doesIt reads the past readings of every site and forecasts a normal range for each reading. It flags live readings that leave the range. There is no training step.

What you getAn alert hours before the failure. Your rules move traffic or send a crew, so the site stays up.

Set up the run

Choose the network TimeH2O watches and your response rules.

Network to watch

Readings to watch

Choose at least one.

History TimeH2O readsPast readings for every site

Normal range holdsMade wider from the 80% range TimeH2O gives

Automated responseMove traffic or send a crew

move traffic alone when a reading is outside the twice in a row; send a crew to other alerts, with free

Step 1 of 3

TimeH2O is forecasting the normal range of every reading

What TimeH2O doesIt reads the past readings of each site. It forecasts the normal range of each reading for the next 24 hours. You do not set alarm limits.

Lines to watch–

Past readings TimeH2O reads–

Ranges forecast–

Metro network

Normal Alert Crew sent or traffic moved Drag to turn the map. Click a site to see it.

Step 2 of 3

TimeH2O is checking live readings against the normal range

What TimeH2O doesIt compares each live reading, every 15 minutes, with its normal range. When a reading is outside the range twice in a row, it raises an alert with the people at risk.

Alerts0

People at risk0

Shortest warning–

Metro network

Normal Alert Crew sent or traffic moved Drag to turn the map. Click a site to see it.

Step 3 of 3

Your rules act on TimeH2O alerts before the site fails

What TimeH2O doesIt gives the normal range of each reading. Your rules use it to move traffic or send a crew. You decide on the rest, and an operator reviews each move afterwards.

Done by rules–

Waiting for you–

People offline0

Metro network

Move traffic if outside the · crews

Normal Alert Crew sent or traffic moved Failed Click a site to see it.

Results · next 24 hours

People who lost service–

Faults found before failure–

Shortest warning–From the alert to when the site would fail.

The sites you followed

Site	Reading	Alert	Action	What happened

The app uses sample data made for this example. The numbers show how the steps work. They are not results from a customer.

Source code · Python

Run this use case with TimeH2O in your own environment.

What the code doesIt makes sample data like this app: 14 days of readings every 15 minutes for 12 sites. TimeH2O forecasts a normal range for the next 24 hours. Then the code flags readings outside the range twice in a row. To use your own data, change step 1 in the code to read your CSV.

Get a free API key at tabh2o.h2oai.com, then set itexport TABH2O_API_KEY=<your key>

Install the packagespip install pandas numpy requests

Save the code below as a file and run itpython network_fault_detection.py

network_fault_detection.py115 lines

The API runs TimeH2O for you, so you do not need a GPU. The code makes its own sample data, so its numbers differ from the numbers in this app.