Exosite

AI / Analytics

AI and analytics on live equipment data

Functions and machine learning run on each signal as data arrives, and the results become signals, statuses, and conditions.

Talk to a solution engineer →Read the docs (opens in new tab)
Search ▾ Home ▸ Administration ▾ North America ▾ Midwest Service ▾ Great Lakes Molding ▾ Holland Plant HPU K-HX450-0013 HPU K-HX450-0023 Pump 11 Pump 12 Pump 14 ▸ Muskegon Plant ▸ Prairie Water District ▸ Ridgeline Foods ▸ Southeast Service ▸ Europe ▸ Direct Customers Pump 12 Realtime Summary Hydraulics Analytics Historical Home › North America › Midwest Service › Great Lakes Molding › Holland Plant › Pump 12 Anomaly Score 0.82 10/30/2026 4:55:42 PM Vibration RMS 0.42 in/s 10/30/2026 4:55:42 PM Vibration Std Dev · 1 h 0.071 in/s 10/30/2026 4:55:42 PM Bearing Condition ! Abnormal since 2:14 PM Bearing Vibration and Anomaly Score · 7 days Vibration RMS (in/s) Anomaly score Rule: score above 0.7 for 15 min Fri Sat Sun Mon Tue Wed now Conditions Warning Vibration anomaly on bearing Today 2:14 PM Cleared Vibration anomaly on bearing Oct 24 Cleared Discharge pressure above 85 psi Oct 22 Cleared Low oil level Oct 21 Operating Plot · Vibration vs Flow Flow (GPM) in/s

Rules, anomaly detection, prediction, and AI

Each approach builds on the one before. Most teams start with rules and condition monitoring, and add anomaly detection, prediction, and AI as their data and needs grow.

Rules and thresholds limit A fixed limit on a known signal Anomaly detection learned normal range Learns normal behavior, flags the unusual Prediction failure limit ~3 weeks Estimates time until a limit is reached AI assistants and agents Which pumps need service? Pump 12 and HPU-0023 have open conditions. Answers questions and acts on your data

Prediction is the basis of predictive maintenance: planning service before a failure instead of after it.

Predictive maintenance →

Offering these insights to your customers as a service, such as monitored uptime or performance reporting.

Service transformation →

Analytics on streaming data

Search ▾ Home ▸ Administration ▾ North America ▾ Midwest Service ▾ Great Lakes Molding ▾ Holland Plant HPU K-HX450-0013 HPU K-HX450-0023 Pump 11 Pump 12 Pump 14 ▸ Muskegon Plant ▸ Prairie Water District ▸ Ridgeline Foods ▸ Southeast Service ▸ Europe ▸ Direct Customers Pump 12 Signals Transforms Rules Dashboards Metadata Asset Configuration › Transforms Vibration Std Dev · 1 h Standard Deviation in Time Interval ⋮ Discharge Avg · 10 min Average in Time Interval ⋮ Pump Efficiency Kinetra Insights · Pump Efficiency ⋮ Run Hours Flow Transformations · Accumulate ⋮ Anomaly Score Machine Learning · K-Means Single ⋮ Fault Code Changes On Change Filters · On Change ⋮ New Transform Insight module Windowed Transformations Function Standard Deviation in Time Interval Standard Deviation of the values in a time interval. Outputs: Standard Deviation, Average, Count Input signal Bearing Vibration RMS (in/s) Interval size 1h Interval alignment 0 Outputs Standard Deviation Vibration Std Dev · 1 h Average Vibration Avg · 1 h Count Not used CANCEL SAVE

Built-in functions

Averages, minimums, maximums, sums, counts, and standard deviation over time windows, rolling aggregations, and change filters, added to any signal without code.

Your own functions

Turn your engineers' knowledge into functions, such as pump efficiency or wear rate, and reuse them in every asset template.

Results are signals

Each result is a new signal or status, so it can drive dashboards, rules, conditions, and notifications like any other.

Anomaly detection

Anomaly Score Machine Learning · K-Means Single Online k-means updates cluster centers as new data arrives. Points far from every center are flagged. Adapts to changes in how the asset runs over time. Input signal Bearing Vibration RMS (in/s) Number of clusters 3 Learning rate 0.05 Threshold (deviations) 2 Outputs Anomaly Vibration Anomaly Score Anomaly Score Centers Cluster Centers State Model State How the score develops Vibration RMS Anomaly score Rule threshold 0.7 Rule Anomaly Score above 0.7 for 15 minutes Condition Warning Vibration anomaly on bearing · Pump 12

Learns as data arrives

Online clustering updates with each new reading and adapts to how each asset normally runs, without a separate training step.

A score for every reading

A rule on the anomaly score raises a condition when behavior moves outside normal, with the same notifications as any other condition.

Set once in the template

Clusters, learning rate, and threshold are set in the asset template, so every unit of a product model is monitored the same way.

Data science with the data lake

pump-fleet-bearing-study.ipynb Python 3 [1] import pandas as pd lake = "s3://kinetra-exosense-datalake" signals = pd.read_csv(f"{lake}/tsdb/v=1.1/2026-*.csv") events = pd.read_csv(f"{lake}/events/2026-*.csv") [2] # label bearing failures from rule state changes in the event history failures = events[events.rule == "Bearing temperature high"] vib = signals[signals.signal == "Bearing Vibration RMS"] vib.groupby('asset').value.describe() asset count mean max failures Pump 07 8,712 0.31 0.48 0 Pump 11 8,698 0.33 0.51 0 Pump 12 8,705 0.38 0.71 1 HPU-0023 8,690 0.44 0.92 2 Vibration before failure vs. normal 30 days before failure [3] # training set: 7-day windows labeled by failures in the next 21 days train = build_windows(vib, failures, window='7D', horizon='21D') train.to_parquet(f"{lake}/studies/bearing-train.parquet") 1,284 labeled windows from 64 assets · 37 before a failure

Everything ExoSense records

Signal history, events and rule state changes, asset configuration changes, and the group hierarchy.

In your cloud storage

Files sync to AWS S3, Azure Storage, or Google Cloud, with signal history written about every hour.

Use your own tools

Work in Python notebooks, data platforms, and BI tools: compare units across the fleet, label failures from the event history, and build training sets.

Your own models

ExoSense · Pump 12 pipeline Signals → transforms → rules Bearing Vibration RMS signal Discharge Pressure signal Motor Current signal Remaining Useful Life from model Failure Risk from model Risk above 60% rule Your model service External insight · HTTPS + JSON Hosted on Murano, AWS, Azure, Google Cloud, or your servers predict_rul inlets: vibration, pressure, current signal data results ExoSense data lake Exported for your data tools tsdb/…/signals.csv Signal history, hourly assets/…/pump-12.json Asset configuration groups/…/hierarchy.json Groups and users Training Your data science tools Label failures from the event history Train and validate on your fleet's data Package the model as a web service deploy history synced to your cloud storage Published to Exchange Its functions appear in ExoSense for anyone building asset pipelines.

Train on your history

Build models from data lake history with the data science tools your team already uses.

Run models as insight services

Host a trained model as a web service on Murano or any cloud. ExoSense sends it signal data and the results come back as signals and statuses.

Use it on any asset

Once published, the model's functions can be added to asset pipelines and templates like the built-in ones.

Exosite Engineering Services

Analytics and AI usually need engineering for your equipment and data, and many organizations don't have the people or time to do it in-house. Your team can partner with Exosite's engineers and data scientists to do the work.

Analytics for your equipment

Turn what your engineers know into functions and rules, such as efficiency, wear, duty cycle, or remaining filter life, and add them to your asset templates.

Models and data science

Use data lake history to label failures, train and validate models, and run them as insight services on your assets.

AI prototypes

Try an assistant, agent, or reporting idea on your real equipment data before committing to it.

1 Assess Review equipment, signals, and failure history Agree on goals with your engineers Rank use cases by value and data readiness Result A short list of first use cases 2 Prototype Build functions or models on your history Check results against known failures and events Review false alarms and misses together Result Validated analytics, ready to deploy 3 Deploy Add them to your asset templates Connect rules, conditions, and notifications Add dashboards for your team and customers Result Running on every unit of the product model 4 Improve Track alerts, outcomes, and service actions Tune thresholds and retrain models Extend to more equipment and use cases Result Regular reviews with your team

Talk with us about analytics and AI

Tell us about your equipment and data, and we'll suggest where to start.

Talk to a solution engineer →Next: Integrations & APIs