About AI-Powered Anomaly Detection
| Platform: | AvailableWebNot availableMobile |
|---|---|
| Plan Type: | Not availableBasicNot availableEssentialNot availablePremiumAvailableEnterprise |
| User Type: | Not availableRequesterNot availableFull UserAvailableAdministrator |
AI-powered anomaly detection is available on the Enterprise plan as part of Asset Intelligence. For more information, contact your MaintainX® representative.
AI-powered anomaly detection uses machine learning to identify when an asset deviates from its normal behavior. It analyzes the machine signals that your connected equipment sends to MaintainX, such as temperature and current, and raises an alert when a signal moves outside the pattern that's normal for that asset.
It's the detection layer of Asset Intelligence, which uses your machine data to optimize maintenance and reliability. Asset Intelligence connects your data, detects problems in it, and helps your team act on them.
- Connected data sources bring your equipment's signals into MaintainX.
- AI-powered anomaly detection finds the anomalies in those signals.
- Alerts and work orders are how your team acts on them.
Equipment often gives warning signs before it fails. A motor runs a little hotter or a pump draws more current, and nobody notices until the asset goes down. Because the models learn what's normal for each asset, you don't have to define thresholds or write rules to catch these changes. Your team reviews each alert, and then turns it into a work order, keeps watching the asset, or dismisses it. The outcome of each review, and of each work order your team closes, gives MaintainX more context to sharpen detection over time.
AI-powered anomaly detection checks the signals that your connected equipment sends to MaintainX. It's different from Smart Input Checks, which check the values that people enter in procedure fields and manual meter readings.
How AI-Powered Anomaly Detection Works​
AI-powered anomaly detection works in four steps, from getting your equipment's data into MaintainX to your team acting on an alert.
- Your data flows into MaintainX: A connected data source, such as an OT data connector, streams your equipment's signals into MaintainX as meter readings.
- You turn on an anomaly monitor: Choose the assets to watch and confirm which tags carry the signals. For more information, see Create a Monitor.
- MaintainX detects and raises an alert: After a learning period, MaintainX compares each asset's signals with what's normal for that asset. When it finds an anomaly, it raises an alert and explains what it found.
- Your team reviews and acts: Your team decides whether the alert is worth acting on. The outcome of each review, and of each work order your team closes, helps MaintainX improve detection over time.
What It Detects​
MaintainX provides three kinds of anomaly detection. Each one is an anomaly monitor that you turn on from the Monitors page.
| Detection | What it looks for | Signals it needs |
|---|---|---|
| Temperature | Unusual behavior in an asset's temperature readings, such as spikes, gradual rises, and sustained shifts away from normal. | One temperature signal for each asset. |
| Current | Unusual behavior in an asset's current readings, such as spikes, gradual rises, and sustained shifts away from normal. | One current signal for each asset. |
| Pumps, fans, and compressors | Efficiency loss, which means the asset is working harder to do the same job. MaintainX compares the effort the asset uses, such as its current, with the output it produces, such as the discharge flow of a pump. | Two signals for each asset. You must connect both. |
Temperature and current detection works on any asset that has that kind of data. It looks for behavior that's unusual for the asset, so it can surface problems that no one thought to write a rule for. An anomaly shows that something changed, but it doesn't say what.
Detection for pumps, fans, and compressors looks for a specific developing condition and combines more than one signal to do it. This kind of detection is available only for those three asset types. Assets of any other type use temperature and current detection.
If more than one kind of detection flags the same asset, MaintainX groups the detections into a single alert, so you get more context instead of duplicate alerts. For more information, see How MaintainX Groups Detections Into Alerts.
An anomaly monitor evaluates all of an asset's signals that measure the same thing, such as every temperature signal on the asset, and not only the ones you selected during setup. As a result, a signal you didn't select can affect when MaintainX raises an alert and when the alert clears.
When Detection Starts​
MaintainX needs to learn what's normal for an asset before it can spot what isn't. Detection starts after a learning period of about two weeks.
- Temperature and current: The two weeks are counted from when data starts flowing into MaintainX. If an asset is already sending data when you turn on the monitor, it's likely ready right away. A new asset that comes online takes about two weeks.
- Pumps, fans, and compressors: The two weeks are counted from when you turn on the monitor, and the data must include at least 17 hours of asset runtime. Readings must arrive every 20 minutes or more frequently.
MaintainX also checks the quality of your data. If an asset sends unexpected values or stops sending data, detection waits until the data looks healthy again. For more information, see A Monitor Hasn't Raised an Alert.
What You See When MaintainX Detects an Anomaly​
MaintainX raises an alert on the Alerts page, and a counter appears next to Alerts in the sidebar. Each alert includes:
- Readings: A chart of the signal behind the alert, so you can see when the problem started and what changed. If the signal returns to normal, the chart shows when.
- AI interpretation: A written summary of what MaintainX found, and a recommended action. MaintainX bases the interpretation on both the machine data and the asset's history, such as its work orders, and links to the records it used.
- Evidence Timeline: A history of what happened to the alert, including comments and linked work.
To write the AI interpretation, MaintainX adds a service account named Alert Summary Bot to your organization when you turn on an anomaly monitor. The account is read-only, doesn't use a billing seat, and doesn't take action on your behalf.
Temperature and current detection runs on a daily interval, so MaintainX can raise an alert slightly before the data looks anomalous. The alert marks the start of the anomalous interval.
For more information about what an alert shows and how to review one, see About Alerts and Review an Alert.
How Detection Improves Over Time​
MaintainX uses the outcome of every alert your team reviews, and every work order your team closes, to sharpen detection. The more context you provide, the more information MaintainX has to learn from. For example:
- Describe what you found when you confirm an alert.
- Choose the reason that fits best when you dismiss an alert.
- Make sure to complete the work order's procedure.
For more information, see How Alert Detection Improves With More Data.
Dismissing an alert doesn't stop MaintainX from watching the asset. While the detection behind a dismissed alert is still active, MaintainX doesn't raise a new alert for the same ongoing issue. Once the detection clears, a new occurrence raises a new alert.
After your team closes the work order for a confirmed alert, MaintainX waits until the asset's readings have been normal for several days in a row before it closes the alert. While it waits, the alert's status is Verifying Fix, and when it closes the alert, the status changes to Resolved. This keeps alerts from closing and reopening repeatedly. For more information, see Alert Review Statuses.