AssetPulse User Guide

Step 1 — Asset

Describe the machine itself.

Machine group, power band and foundation together decide which ISO zone boundaries apply, so they are worth getting right rather than leaving at the defaults.

Step 2 — Drivetrain

A machine is not one object — it is a power path. Step 2 is where you list the parts of that path, driver first: the motor, then whatever it turns through whatever couples them. A pump skid might be motor → coupling → pump; a cooling tower might be motor → belt → fan.

This matters because rotating machines fail at predictable frequencies. A pump with 6 impeller vanes pushes energy at 6 times shaft speed; a 23-tooth gear meshes at 23 times its shaft speed; a 4-pole motor shows electrical faults near line frequency. Once AssetPulse knows the parts, it knows which frequencies to look at, and a peak in the spectrum stops being “energy at 174 Hz” and becomes “vane pass frequency”. Without the drivetrain it can still grade overall vibration severity — it just has fewer names to give what it finds.

Adding a component

Click + Add Component. Each component appears as its own card with:

• Component N — the type. Choosing the type is what reveals the parameter fields below it, so set this first.
• Name — free text, how the component reads in the app, for example 50 HP Motor or Gearbox (4:1).
• Remove — deletes that card.

Add them in power-path order. The position of each card is the drivetrain order, so a component added out of sequence should be removed and re-added rather than left in the wrong place.

Parameters by component type

Each type asks only for what it needs. These values build the kinematic model that is handed to the analysis engine with every capture; leave one blank and the frequency it would have defined is simply not part of the model.

Skipping it, and adding it later

Step 2 is optional — Next works with an empty list. Anything you skip can be added later from the asset’s Setup tab, and the next capture is analyzed with the detail you have added by then. Earlier captures keep the scores they were originally given; they are not re-analyzed.

One thing to come back for: in step 3 each monitoring point can be attached to a component through the On Component field. That attachment is how a finding gets pinned to a part of the machine rather than to the asset as a whole, and it can only point at components that exist — so components added here are what make that field useful.

Step 3 — Monitoring Points

A monitoring point is one sensor probe measuring one point on the machine. Click + Add Monitoring Point and fill in:

For bearings, entering a part number and clicking Look up gives exact defect frequencies. If you only supply ball count, pitch diameter, ball diameter and contact angle, the analysis uses a flagged approximation instead.

Step 4 — Review

The last step is a read-only summary of everything the wizard is about to create. Check it before saving — the machine description is what every future severity score is graded against, and a wrong ISO group or a missing vane count is far cheaper to fix here than to notice a month later in the trend.

← Back returns through the steps with everything you entered intact, so correcting something costs nothing. Create Asset saves the asset, its components and its monitoring points together in one operation — there is no half-saved state to clean up if something is rejected.

What happens after you save

The asset appears on the Assets list straight away, showing No assessment yet — it has no severity score because scores come from waveform captures, and it has not seen one yet. From here:

• The first score arrives with the next capture from one of its sensors. Every capture that lands on one of this asset’s monitoring points is analyzed automatically from now on.
• To get one immediately, open the asset, go to Telemetry and click Request Waveform. The sensor takes the capture on its next sync rather than instantly.
• Historical captures are not back-filled. Captures that arrived before the asset existed are not analyzed retroactively, and neither are earlier captures re-analyzed when you add drivetrain or bearing detail later. Trending starts now.
• The baseline takes a few captures. Until roughly five running captures have been folded in, the Health tab reports the baseline as learning… rather than a ratio.

Everything entered in the wizard remains editable from the asset’s Setup tab, so a rough first pass is a reasonable way to start — get the sensor attached and the ISO class right, then add drivetrain and bearing detail as you learn the machine.

1. Health

Health is where AssetPulse earns its place. Everything else on the gateway records measurements; this tab turns one waveform capture into a graded, explained verdict about a specific machine, and puts thirty days of those verdicts on one chart.

The tab has three parts: the AI Assessment panel on the left, the Measured panel on the right, and the Severity Trend chart underneath. The important thing to understand first is that the two panels always describe one single capture — the one currently selected on the trend chart. Click a different point on the trend and both panels change to that moment in the machine’s history.

The AI Assessment panel

The panel header names the capture it is describing — Captured <date and time> — followed by the actions that operate on that capture: Report Clauses, View Waveform, Prompt, and Regenerate (or Generate AI report where no report exists yet). Each is covered below.

The body of the panel holds, in order:

The panel reads differently depending on what exists for the selected capture:

Severity bands

The 1-10 scale is an AssetPulse scale, not a published standard. The standardized part of the assessment is the ISO zone in the Measured panel, described next.

The Measured panel

Where the assessment is an interpretation, this panel is the raw evidence behind it. Its heading names the monitoring point and the capture time, so it is always clear which probe on the machine produced these numbers.

The baseline is learned, not configured: the first several running captures from a monitoring point are averaged into its normal level, and once established that reference is frozen so a slowly worsening machine cannot quietly redefine “normal” for itself. Captures taken while the machine was off are never folded into it.

When the selected capture was taken with the machine stopped, this panel adds a note that the machine appears off and that the last running assessment still stands.

Severity Trend — and how to read the dots

The chart underneath plots one dot per capture over the past 30 days, with severity on the vertical axis. It is both the history of the machine and the navigation control for this whole tab.

Click anywhere on the chart to select the nearest capture. The selected dot grows and gains a ring, and the AI Assessment and Measured panels above jump to that capture — its assessment, its defect indications, its measured values, and its waveform. This is how you answer “what did this machine look like when the trend started to climb”: click the dot where it turned, and read the assessment from that day.

The dot colors are the important part:

Hovering any dot shows the exact time, the monitoring point, and whether the number is an AI Severity or an Engine Severity.

Report Clauses

AI reports are useful in proportion to how much they know about your plant. Report Clauses is where you decide which standing instructions are added to this particular machine’s reports — things the analysis cannot infer, such as “this pump runs intermittently, so short run periods are expected”, “refer to bearings by our stores code”, or “always state whether the reading justifies an outage”.

The modal has two lists:

• Applied — the clauses this asset uses, in order. The arrows move a clause up or down, and Remove takes it out. Order matters because it is the order they appear in the prompt.
• Available — every other enabled clause in the library. Add appends it to the applied list.

Save applies the selection to this asset only, from the next report onward. The clauses themselves — their names and wording — are written once in Settings under AI Reports, then reused across assets; a clause shown as disabled has been switched off there and will not be sent even while it is listed here.

Prompt — seeing exactly what is sent

Prompt opens the complete request for the selected capture, in two parts: the System message that sets the analyst role and output format, and the User message carrying this capture’s evidence. The header names the provider and model it would go to. Any report clauses applied to the asset are already included in what you see.

Two reasons to open it. The first is trust: this is the whole payload, so you can confirm for yourself that what leaves the gateway is the structured analysis evidence and nothing else — no raw waveform is ever sent.

The second is iteration. Both boxes are editable, and Generate with this prompt runs the report using your edited text, so you can try a different wording on a real capture and read the result immediately. The title marks the prompt · edited once you change anything, and Reset restores it. Edits here are deliberately not saved — only the report they produce is kept. When you find wording that works, put it in the clause library so it applies from then on.

The Waveform viewer

View Waveform opens the frequency spectrum of the selected capture — the evidence underneath everything else on this tab. The button appears only when the selected capture has a stored raw waveform behind it. The spectrum is computed on the gateway when you open the viewer, from the stored samples.

The strip along the top states the conditions of the capture: when it was taken, the run speed in RPM (marked configured when it came from the asset’s fixed speed and measured when the sensor reported it), the ODR (output data rate, in Hz — the sampling rate, which sets how high in frequency the spectrum can see), the number of samples, and the temperature.

Three metric views, selected with the buttons above the chart:

In the acceleration and velocity views, the Traces checkboxes turn the X, Y and Z axes on and off individually, plus Overall — the resultant of all three, off by default. Showing one axis at a time is the fastest way to confirm which direction a defect is acting in, which is exactly what the axis orientation you set on the monitoring point is there to make meaningful. The envelope view is Z-only, so it has no trace selection.

Two features make the spectrum readable rather than merely present:

• A dashed Run speed line marks 1x — one turn of the shaft. Nearly all vibration diagnosis is a question of where energy sits relative to this line.
• The tooltip reports every frequency both in Hz and as a multiple of run speed — “174.0 Hz (5.8× run speed)”. A peak at 2x suggests misalignment; a peak at the vane or blade count suggests the pumped or moved fluid; a peak at a non-integer multiple with sidebands suggests a bearing.

Drag the handles on the slider under the chart to zoom the frequency axis, and use Full screen when a crowded spectrum needs the room.

2. Telemetry

Health answers “what is wrong with this machine”. Telemetry answers “what have the numbers been doing”, and it is where you set the thresholds that make AssetPulse tell you about a change without being asked.

Everything on this tab is scoped to one monitoring point. Where an asset has more than one, the Monitoring point selector at the top switches between them; dual-probe sensors list each probe separately. Two buttons sit alongside it:

• Request Waveform queues a one-shot raw capture from this point’s sensor. The sensor is asleep between its scheduled reports, so it takes the capture on its next sync rather than immediately — then the capture is analyzed and appears on the Health tab as a new dot.
• Alert Thresholds opens the table of every threshold on this asset, described at the end of this section.

Refresh reloads the graphs for the current window.

Choosing a time range

The panel at the top of the tab sets the window every graph below it draws. Preset buttons cover Last Hour, Last 12 Hours, Last 24 Hours, Last Week and Last Month. Beside them, a Days box with Apply takes any number of days from 1 to 365 — useful for “the last 90 days” without hunting for dates.

Under that, Or select specific date/time range takes an absolute Start and End date and time and applies it with Apply Range; Reset returns to Last 24 Hours. This is the one to use when you are comparing against a known event — the day a coupling was replaced, or the shift when an operator reported a noise.

Wide windows are averaged into buckets before they are drawn, because a month of raw readings is neither readable nor quick to load. The bucket size is chosen for you:

The seven graphs

Each monitoring point produces seven graphs. Together they are the sensor’s routine reporting — the small numbers it sends on every cycle, as opposed to the full waveform captures behind the Health tab.

Working with a graph

Every graph carries the same controls.

Setting thresholds by dragging

Tick Thresholds on a graph and two dashed lines appear across it: an amber Warning line and a red Critical line, each with a labelled tab showing its value. Drag a tab up or down to set the level; the value follows your drag and is saved when you let go. The graph keeps its scale while you drag, and leaves headroom above the data so a threshold can be placed above anything recorded so far.

Setting a threshold against a real trend is the point of doing it here rather than in a form: you can see the machine’s normal band, its daily variation, and where a line would have to sit to catch a genuine change without firing on every start-up. A common first pass is Warning a little above the highest normal peak, and Critical where you would want someone woken up.

Thresholds are stored per monitoring point, per graph — so the drive end and the non-drive end of the same machine can have different limits, and temperature limits are independent of vibration limits.

Alert Thresholds — deciding what a breach does

Dragging a line sets a value. What should happen when a reading crosses it is set in the Alert Thresholds table, reachable from this tab or from Activity. It lists every monitoring point on the asset, and for each graph that has thresholds:

The checkboxes for a level stay disabled until that level has a value, and a point with no thresholds at all reads No thresholds set. Enable thresholds on this point’s Telemetry view first. — the values come from the graphs, the actions come from here.

3. Activity

What AssetPulse has already done about this machine, in two lists.

Threshold Alerts is every threshold breach this asset has raised, unacknowledged first, with the count of open ones in the heading. Each row names the metric and the monitoring point, states the reading against the limit it crossed (reading 0.126 vs warning 0.082), and timestamps it. Tags on the row record what the breach triggered — waveform requested, emailed. Acknowledge clears one row from the open count and Acknowledge all clears the backlog; the asset keeps its raised status until every open alert is acknowledged. A row marked timed out was never acknowledged within the re-arm window, so a continuing breach raised a fresh alert instead of staying silent behind the old one.

Notification History is the fault emails actually sent for this asset — time, severity, monitoring point, subject line and recipients. It answers “was anyone told?” without going through a mailbox. Send test sends a test notification using the current configuration and real report content, which is the quickest way to prove SMTP and the recipient list work before you need them; the gear button opens Notification Configuration for this asset.

4. Service

The Service tab is the machine’s maintenance record — and it does one thing no other tab does: adding a service report is what stops repeat fault emails after you have fixed the machine. If you read only one paragraph about this tab, read the next one.

Why notifications go quiet, in plain terms

A machine with a real fault produces a high severity on every capture, for days or weeks. Emailing each one would be useless, so AssetPulse remembers the highest severity it has already emailed you about for this asset, and does not email again for anything at or below that level. One fault, one email.

Only three things break that silence:

1. It gets worse. A capture that scores higher than the level already reported — and crosses another of the severity levels you asked to be told about — sends a new email.
2. You add a service report. The remembered level is reset, so the very next qualifying capture emails you again. This is the intended way to close the loop.
3. The re-arm timeout expires. If a fault is still there and nobody has recorded any work, AssetPulse re-sends after a set period — 48 hours by default — so an ignored fault resurfaces rather than being silenced forever. The period is set with the other notification defaults in Settings, and can be switched off.

So the practical rule: service report after every repair. It is the difference between “no emails because the machine is fine” and “no emails because the app already told us once”.

Adding a service report

Click + New service ticket. The Add Service Report dialog asks for two things, both required:

Add Service Report saves the entry and resets the notification level in the same action; the app confirms with Service confirmed — notifications reset. There is no separate “clear faults” button to remember.

Reading the history

Service History lists entries newest first: when, who, the note, and — where notifications had already fired — the level it reset from, shown as reset from Sev 8. That last detail makes the log a record of what the machine was doing when someone intervened, not just a list of jobs.

Read alongside the Severity Trend, the log is what turns a chart into a story: severity climbing to 8 over two weeks, a bearing replacement, severity back to 3, and a clean baseline for whatever happens next. It is also the honest answer to “did that repair work?” — the trend after the entry says so either way.

5. Setup

Everything from the Add Asset wizard, editable, plus two per-asset switches at the top.

Below the switches, Details holds the machine description — the same fields as wizard step 1 — with Reset and Save details. Then Drivetrain and Monitoring Points, each with add, edit and remove controls; a monitoring point row shows its sensor, device ID and mount location, and whether a bearing is set. This is where the detail you skipped in the wizard gets added.

Changes apply to the next capture. Correcting an ISO group or adding a vane count does not re-score captures already recorded, so a trend can legitimately step after an edit — worth a service-log note when it does.

The Danger zone holds Delete asset, which permanently removes the asset along with its components, monitoring points and history. The sensors themselves are untouched and stay available to assign elsewhere.