# 4 · Measurements, trends, and alarms

This chapter shows you how to read a point's trends, dig into a vibration reading's
waveform and spectrum, record what you conclude, set alarm limits, and trust the
single, reliable state Mechbase derives from a stream of readings — the part that
keeps the whole system worth paying attention to.

## Reading a trend

Every measurement point has a **Trend** — its readings plotted over time. It's the
single most useful view for spotting a problem early, because a value that looks
fine on its own often tells a clear story once you see where it's heading.

![A temperature point: headline, trend with alarm limits, and readings](../_images/04-readings-by-run.png)

For most points — temperature, pressure, ultrasound — the trend is one chart:

- **The line** is the reading over time. You're looking for direction and pattern —
  a steady climb, a step change, a new wobble — as much as the latest number.
- **The dashed limits** are the alarm thresholds: **amber** for warning, **red**
  for critical. How much headroom there is between the line and those limits tells
  you how hard the equipment is being pushed.

**Vibration points** show a **grid of small charts** instead — one per metric
(*Acc RMS*, *Peak*, *Vel 10Hz*, *Vel 2Hz*, *O-P*, *P-P*, *CF*, *Kurt*), each spanning
the week up to the latest reading. The metric the alarm watches carries its limits.
Under each chart, **normal** gives the band the value usually sits in — the middle
80% of running readings over the previous 30 days — so you can tell at a glance
whether today is unusual for *this* machine, not just whether it's under a limit.
Hover any chart and a cursor follows on all of them; click a point to inspect that
reading below.

When the machine has a tachometer, the grid adds **Speed**, **1X**, **1X phase**,
**2X**, and **2X phase**. Each of these has a **Time | RPM** toggle: against RPM it
becomes a run-up/coast-down (Bode) plot, where a resonance shows as a peak in
amplitude with a swing in phase.

Remember that the units on every axis are *your* units, set in your profile
([Getting started](getting-started.md)) — the same data reads in `Hz`/`mm/s` or
`CPM`/`in/s` depending on who's looking.

## Filters and time range

The controls above the trend let you focus on exactly the readings you care about:

- **From → To** — set a date range to zoom into a specific period, like the week a
  problem started.
- **Status** — show only readings of a given state (Nominal, Check, Warning, or
  Critical). Handy for jumping straight to every warning-level reading.
- **Waveforms only** (vibration) — only readings that carry a full waveform.
- **Plot stopped** (most other types) — by default the chart skips readings taken
  while the machine was stopped, and tells you how many it left out (*"N readings
  taken while the machine was stopped are not plotted. Plot them"*). Tick it to show
  them. On vibration charts, stopped readings are always shown, faintly.

## The readings table

Below the chart, **Measurements** lists the readings in the chosen range, newest
first, under a heading per day. Instead of hundreds of near-identical rows,
Mechbase folds each stretch of one machine state into a **run**: *"06:00 → 14:00 ·
Running · 48 readings"*, with the low, typical, and high value and the worst state.
Click a run to expand its readings. **List every reading** switches to a plain
list; **Group by run** switches back.

The columns follow the point's type: for vibration, **RMS**, **Peak**, **Vel**, and
**CF**; for ultrasound, the **Level** in dB; for thermography, a thumbnail, the
**Max** temperature, **ΔT**, and the number of **Hotspots**; for everything else,
the **Value**. Readings taken while the machine was stopped stay in the list,
marked **Stopped**. The search box (`F`) finds readings by their notes.

Click a vibration reading to inspect it in place; click any other reading to open
its own page.

## Inspecting a vibration reading

For vibration, a single number isn't enough to diagnose a fault. Mechbase keeps the
**full waveform** behind each reading, and the **Inspect** panel on the point page
lets you look at it without leaving the trend. It opens on the latest running
reading; click any point on the charts above to switch.

![Inspecting a vibration reading: time waveform and frequency spectrum](../_images/04-inspect.png)

The panel's header repeats the reading's key values (RMS, Peak, Vel) and its
state. If the reading came with a tachometer, it adds the shaft speed and the
**1X** and **2X** amplitude and phase (*"1X 2.1 mm/s ∠ 47°"*); a phase Mechbase
doesn't trust — because the speed wandered during the capture — is shown muted with
the reason.

- **Accel / Vel / Env** — switch both charts between acceleration, velocity, and the
  demodulated **envelope** (for bearing defects).
- **Time waveform** — opens on about four shaft revolutions (or 200 ms without a
  tachometer), with ±RMS and the peak marked and the largest excursion flagged.
  Tachometer pulses appear as tick marks. Drag the overview strip underneath to
  move along the capture. **One rev** folds the signal into a single averaged
  revolution with its 1X fit — the clearest view of unbalance and looseness.
- **Frequency spectrum** — opens zoomed on the band that holds almost all the
  energy, with the five largest peaks labelled. **Hz | Orders** switches the axis
  to multiples of shaft speed, with 1X–10X order lines drawn in. Double-click for
  the full span. If a bearing is assigned to the point, its defect frequencies are
  marked too ([Bearings](#bearings-and-defect-frequencies)).
- **Pin** — keep the reading for side-by-side comparison
  ([Comparing readings](#comparing-readings)).
- **Record finding** — write down your diagnosis ([Recording a finding](#recording-a-finding)).
- **Open** — go to the reading's own page.

If the spectrum shows a strong tone that looks like a bearing but the point has no
bearing assigned, Mechbase points it out — *"Possible bearing tone at 3.58×"* — with
an **Assign bearing** button that lists the library bearings matching that tone.
Nothing is assigned until you choose.

## A single reading's page

Every reading has its own page — open it from **Open** in the inspect panel, from
the readings table, or from a route report.

![A reading's page: each value in context](../_images/04-reading.png)

- The header shows the reading's state and, if it's in alarm, why (*"Vel 10Hz 3.98
  above warning limit 1.80 mm/s"*), with who recorded it and how (a MaintNode, a
  route, or by hand). **Previous** and **Next** step through the point's readings;
  **Point trend** goes back to the point.
- **Values** — every value in context: a small trend of the last 40 running
  readings, the normal band, the limit, and a line comparing it with the limit, the
  previous reading, and the normal band (*"limit 1.80 · prev 2.68 · normal
  0.79–1.43"*).
- **Attachments** — the files that came with the reading: the raw waveform
  (`.dr`), photos, and thermograms with their hotspots. Click to download.
- **Notes** — add a note to the record.
- For vibration, **Bench** gives you the waveform and spectrum with a little more
  control: choose the **Signal** (*Accel*, *Vel*, or *Env*), set a **High-pass** and
  **Low-pass** filter in Hz, and click **Apply** (or **Reset**).

**Delete reading** removes it straight away — no "Are you sure?" — and shows an
**Undo** button for a few seconds in case it was a slip. The point's state is
recalculated from the readings that remain.

## Comparing readings

To compare readings — before and after a repair, this month against last — click
**Pin** on each one in the inspect panel. The pinned readings collect in a tray;
click **Analyze** to open the **Compare** drawer along the bottom of the page.

In the drawer you can switch between **Velocity**, **Accel**, **Envelope**, and
**Waveform**, apply a high/low-pass filter, overlay the readings, show the spectrum
in **dB**, label the **Top** peaks, and add **harmonic** cursors at a frequency you
choose. **Record finding** works from here too.

## Recording a finding

A **finding** is an analyst's conclusion about a reading — what's wrong, how
serious it is, and what to do — written down where everyone can see it, and in a
form a maintenance system can pick up.

Click **Record finding** on the inspect panel (or in the Compare drawer). If you
clicked a peak in the spectrum first, its frequency, amplitude, and any matching
defect (*"148.3 Hz · 2.31 mm/s · 3.05× · BPFO"*) are attached as evidence.

- **Fault** — *Bearing outer race (BPFO)*, *inner race (BPFI)*, *rolling element
  (BSF)*, *cage (FTF)*, *Unbalance*, *Misalignment*, *Looseness*, or *Other*.
  Mechbase suggests one from the peak you clicked.
- **Severity** — **Check**, **Warning**, or **Critical**.
- **Action** — what to do (*"Replace DE bearing"*), and a **Due by** date.
- **Comment** — what you see, in your own words.

Click **Record finding**. It's added to the asset's notes as a card with the
severity, fault, action, due date, and evidence, and it appears in the activity
feed. A finding can be edited but not deleted — it's part of the record.

Findings also have a work-order lifecycle — **Open**, **Acknowledged**, **Closed** —
that your CMMS moves along through the API, attaching its work-order number
(see the [Analytics & BI](developers/analytics-bi.md) and integration cookbooks).

## Bearings and defect frequencies

When a vibration point knows which bearing it's watching, Mechbase can mark the
bearing's defect frequencies on every spectrum, trend each defect band, and alarm
on them.

![The bearing library, searched for 6309](../_images/04-bearings.png)

**The library.** Open **Bearings** from the menu (**Setup → Bearings**). Mechbase
carries a library of several thousand rolling-element bearings. Search by
designation or maker (*"6309"*, *"SKF 6205"*) — press `F` to jump to the search box.
Each result shows the number of rolling elements and the four defect frequencies,
in **orders** (multiples of shaft speed):

- **BPFO** — outer-race defect.
- **BPFI** — inner-race defect.
- **BSF** — ball spin.
- **FTF** — cage (fundamental train).

**Your own bearings.** If the library is missing one, click **Add bearing**. Give
its **Designation**, **Manufacturer**, and **Rolling elements**, then either the four
**Defect orders** from the maker's data, or its **Geometry** (ball diameter, pitch
diameter, contact angle) and let Mechbase work the orders out. Bearings you add are
visible only to your installation.

**Assigning one to a point.** Edit the point and, under **Shaft & bearing**, set
the **Nominal speed**, search for the **Bearing**, and set a **Speed ratio** if it
sits behind a gearbox or belt. From then on:

- Spectra show markers for shaft speed and **BPFO**, **BPFI**, **BSF** (with 2× and
  3× harmonics) and **FTF**, placed using each reading's actual measured speed.
- The point's trends gain **BPFO env**, **BPFI env**, **BSF env**, and **FTF env** —
  the envelope energy in each defect band, reading by reading.
- You can set **bearing band** alarms on them (next section).

## Setting alarm limits

Alarms are what make *Needs attention* mean something. They're defined by **alarm
rules**, on three levels:

- An **installation default** for each measurement type — for example, "for any
  temperature point, warn at 55 and go critical at 75".
- For vibration on rotating equipment, the **ISO 10816 default** that follows from
  the asset's class.
- A **point override** for the cases where one specific point should behave
  differently.

Open a point and click **Alarm rules** to see what's in effect.

![The alarm-rule page for a temperature point using the installation default](../_images/04-alarm-rule.png)

The line at the top tells you where the rule comes from — **Point override**,
**ISO … default**, **Installation default**, or **No rule configured** — and what it
says (*"Threshold · High (above threshold is bad) · Warning 55 · Critical 75"*).
When the point is inheriting a rule, the form below is already filled in with it:
*"Inherited. Editing creates an override for this point."*

An alarm rule has two parts:

- **Method** — how the status is worked out:
  - **Threshold** — a simple limit on the value. The common case.
  - **Narrowband Envelope** — watches a specific frequency band, for spotting a
    particular vibration fault.
  - **Maintenance due** — raises a status based on time or usage, for
    condition-plus-schedule maintenance.

  Plus the **Data key** (which value of the reading to watch — blank uses the usual
  one for the type) and the **Direction**: **High** (above the threshold is bad —
  temperature, vibration) or **Low** (below the threshold is bad — a flow or
  pressure that shouldn't drop).
- **Thresholds** — the **Warning** and **Critical** limits.

Adjust them and click **Create override** (or **Create rule** where no default
applies, or **Save rule** when you're editing an existing override). The new
limits appear immediately on the point's trend.

To go back to the inherited rule, use **Remove override** at the bottom. It takes
effect at once, with an **Undo** in case you didn't mean it.

**Bearing bands.** On a vibration point with a bearing assigned, the page adds a
**Bearing bands** section. It shows the current verdict for the defect bands and
any band rules in force. Each **Band rule** picks a **Band** (or every envelope
band), a **Mode** — an **Absolute value**, or a **Multiple of baseline** (3× for
warning and 6× for critical if you leave them blank) — and the **Warning** and
**Critical** limits; click **Save band rule**. Relative rules compare against a
baseline Mechbase learns from the first five waveform readings while running;
**Reset baseline** starts a fresh one — do it after a bearing change, a repair, or
a remount.

![Alarm rule for a vibration point, with the Bearing bands section](../_images/04-alarm-rule-vibration.png)

> **Tip:** Base your limits on history and expert judgement, not a guess. The
> **normal** band under each chart, and the *limit · prev · normal* line on a
> reading's page, show you where the value really sits. Set warning and critical
> limits that sit clearly above (or below) that band.

> **Note:** Installation-wide defaults are maintained centrally by your Mechbase
> administrator ([Setting up and managing your account](administration.md)).
> Point overrides are yours to set.

## Keeping false alarms down

A predictive-maintenance system is only worth as much as people trust it. If it
cries wolf — firing an alarm every time a sensor twitches or a reading blips — teams
stop paying attention, and the genuine faults get lost in the noise.

Mechbase is deliberately built to avoid that. Instead of tripping on a **single**
reading that crosses a limit, it looks at the **recent pattern**:

1. **Every reading gets a status** against the alarm limits — *normal*, *minor* (a
   little past the limit), or *major* (well past it).
2. **Mechbase averages the status over the last several readings**, not just the
   latest one.
3. **The state changes only when that average crosses a threshold** — so a sustained
   run of off readings moves the asset to Warning or Critical, while a lone spike
   does not.

On top of that, Mechbase knows when a machine is **stopped** and keeps those
readings apart from the running trend, and relative rules compare each machine
against **its own learned baseline** rather than a one-size-fits-all number.

That's why the states are **Check / Warning / Critical** rather than a blunt
on/off: **Check** is Mechbase telling you "a soft pattern is forming here" before
it ever escalates. The payoff is fewer false alarms, warnings you can act on with
confidence, and maintenance effort spent on real problems instead of chasing
ghosts.

> **Note:** Two things shape how responsive this is — the **alarm limits** (from
> history and expertise) and how many recent readings are averaged. A longer window
> is steadier but slower to react; a shorter one reacts faster but tolerates less
> noise. Treat the first weeks on a new point as a "learning" period and review
> both.

## How a reading becomes a state

Pulling the whole chapter together, here's the journey of a single reading — whether
it arrived automatically from a MaintNode device or was typed in on a route:

1. **It's recorded** against its measurement point and added to the trend.
2. **It's scored** against the point's alarm rule — normal, minor, or major.
3. **It's combined** with recent readings (the false-alarm logic above) to decide
   the point's state.
4. **The asset's state** rolls up from its points and its open non-conformities —
   the worst of them sets the asset's state, and feeds the **in alarm** count.
5. **Everything downstream updates**: the asset's colour, **Needs attention**, the
   section cards, any dashboards, a plain-language line in the **activity** feed
   (*"degraded to warning"*), and an email to everyone who asked for one at that
   level.

Running-state readings (*Production*) and on/off **State** readings take a
different path: they open and close periods rather than being scored and averaged.

So the numbers you act on across Mechbase aren't opinions or stale spreadsheet
entries — they're the live result of this pipeline, running on every reading.
