Once setup is done, you work across three main tabs — Monitoring · Detail Analysis · Report. The screens can be shown in Korean, English, or Japanese; a Korean label in parentheses is what the Korean screen shows.

Monitoring

The at-a-glance view of every piece of equipment.

Summary bar

The top summary bar has these cells.

Equipment cards

Each selected tag gets one card. Click a card to open Detail Analysis for that tag.

• The default sort is worst grade first. Drag cards to arrange them however you like; that order is saved and restored next time.
• The card size slider adjusts grid density.
• A card’s grade follows the same rule as the alarm — at every fault-pattern extraction interval (10 minutes by default) the fault pattern switches the grade on and the ISO zone sets its height. The RMS window in alarm settings is used for running/stopped, signal loss, instrumentation checks, and the versus-normal evidence.

Card badges

Besides stopped, signal lost, and instrumentation fault, On hold has these reasons:

• Cannot judge (판정 못 함) — no RPM or bearing geometry, so the fault pattern cannot be built (sleeve bearings included). Fill in Equipment.
• Waiting (판정 대기) — right after opening the app, before the first evaluation.
• Fault pattern awaiting first measurement / measurement failed — the fault pattern could not be measured this cycle, so the previous grade is held. It is measured again next cycle.

Table filter

With two or more data sources, a table filter (테이블 필터) appears next to the “Equipment monitoring” heading. Pick only the tables you want to see to narrow the cards and the summary bar.

• It changes the display only. Monitoring and alarms still cover every table (to remove one for good, deselect it in Settings → Data Source).
• Hiding everything shows “no table selected for display” with a Show all button.

Monitoring status alerts

If monitoring is not running properly, a warning appears at the very top of the screen — the timer stopped, it is running but has nothing to evaluate, or app-side evaluation stalled. Most of these offer a Re-register button that fixes them on the spot.

This exists so you can tell whether “no events” means the equipment is fine or monitoring is dead. Each message and its fix is listed in Alarms & events › Checking that monitoring is alive.

Event list

The event list below shows recorded events with time · tag · type · detail · grade. Click any row to open detail analysis at the moment that event occurred.

The detail cell shows, by type:

The fault pattern’s grid score, site, and cycle (e.g. 격자 3.2 · 외륜(BPFO) · 2회 연속 확정) are written in the “detail” column of the exported file and in notification text.

• Period chips (today · 7 days · all) and the tag and type dropdowns narrow the list. It shows the most recent 200 matching events, and the count is every matching event. Use export to see beyond that.
• The list is re-read every 30 seconds — events recorded by server monitoring or by another tab appear without a reload. Only events of the tags currently monitored are shown; events of deselected tables are available through export.
• There are eight types — grade: observation (first cycle, no alert) · threshold · sustained · improved · recovered; monitoring health: instrumentation fault · instrumentation restored · signal lost. What each means is in How detection works › Events.

Baseline and band produce no events. A breach never appears in this list — it is displayed on the card and in the evidence table. Grade events arrive once per fault-pattern extraction interval (10 minutes by default).

Exporting events

Press Save (저장) in the event list to open a dialog for choosing a period and file format (CSV · JSON), then download the events for that period. Unlike the list, the file includes events of deselected tables too.

The CSV columns are time · table · tag · alias · type · grade · value · limit · detail.

Column names and the type, grade, and detail values are written in the screen language at download time. JSON key names (time, grade, …) are always English.

How alarm evaluation decides when to create events is covered in Alarms & events.

Detail Analysis

A deep look at a single tag, entered from a monitoring card or an event row.

Detail analysis is a snapshot — it is frozen at the data as of when you entered, and does not keep updating live. To re-capture the latest state, press Refresh, or change the tag or period. If the instrumentation state seen by monitoring changes after you entered, a notice with a Refresh button appears at the top.

Tag and period selection

• Use the dropdown at the top to switch the tag being analyzed.
• Choose the query period from presets — 1 hour · 6 hours · 24 hours · All — or set it in the calendar next to them. Presets are anchored to that tag’s latest data.
• If the tag has less data than a preset’s length, that preset is disabled. Clicking it shows “not enough data”; use All instead.
• A custom range is set in whole days (start date 00:00 to end date 23:59). Switching from a preset pre-selects the dates of the range that preset covered.

Chart time axes span the whole selected period. Gaps with no data are left empty, so you can see exactly where data actually exists.

Measurement validity warning

If a measurement anomaly such as clipping (saturation), a flatline, DC drift, or an isolated spike is detected in the signal, a Signal anomaly — check the sensor (신호 이상 — 센서 점검 필요) warning appears at the top and diagnosis enters a diagnosis withheld state.

A saturated signal reads lower in amplitude than reality, and a single impulse in the time domain smears across the whole spectrum as broadband content. So while this warning is up, confidence drops for all evidence, not just one candidate. Check the sensor, cable, and mounting before suspecting the machine.

A single grey DC component (DC 성분) line without the warning box is not a fault. It notes a sensor with a large but steady mean (a MEMS accelerometer’s gravity axis, a proximity displacement probe, …), and judgment continues. Reports carry the same grey note.

Severity assessment

The top of the card shows the overall grade (the bar under it marks which of the four steps it is), the verdict, and one line on how the two axes produced the grade (for example “fault pattern present + ISO zone A → Warning”). Below them come what to do now (지금 할 일) and the supporting facts (evidence score, plus the judgment time and how many of the period’s segments gave the same verdict); without RPM or a bearing, “criteria n/N (… not applied)” is shown as well. A reason sentence is added only when that one line cannot explain the result, such as when the fault pattern could not be measured or the vibration is high with no pattern.

Two boxes underneath show how the grade was set. ① Fault pattern (① 결함 패턴) is the axis that switches the grade on (none / present), and ② Vibration magnitude (② 진동 크기) is the axis that sets its height — the ISO zone A–D bar fills up to the current value, and the row below lists the grade each zone gives when a fault pattern is present (A·B → Warning, C → Danger, D → Critical). The zone bar uses no grade colors, and the zones are named by their ISO meaning (A new-machine level · B long-term operation OK · C restricted operation · D damage likely) rather than by grade words.

That grade is not set by the vibration magnitude (ISO). Exactly one axis raises it: the fault pattern — whether evidence stands at a bearing defect frequency (BPFO / BPFI / BSF / FTF) in the envelope. Only then does the grade rise, and the ISO zone decides how high it rises.

So a machine that shakes hard but shows no fault pattern stays at Good, while one with ordinary amplitude but a clear pattern moves up — which is exactly where early bearing damage lives, before it reaches the amplitude. Baseline and band values never move the grade; they appear as supporting evidence only.

The period statistics table (ISO value · Crest Factor P95 · broadband RMS P95 · rate of change · worst sustained RMS · observation window) is computed over the whole selected period. Only the ISO value row is a basis for the grade; the other rows only state the fact against their limit (within · over). Hover over a note to see the full calculation conditions.

Monitoring cards use the same judgment rule, so the two screens do not disagree (detail analysis uses a longer window and checks several points across the period). How detection works has the full principle.

Switching the display unit

The same vibration signal reveals different faults depending on whether you view it as acceleration (g) · velocity (mm/s) · displacement (μm). Switch with the display quantity (표시 물리량) buttons in the “Spectrum analysis” heading; the unit the sensor actually measured is marked (원본) (original).

If you suspect a bearing fault but the velocity waveform shows little, switch to acceleration. The conversion applies to the waveform, FFT, spectrogram, cepstrum, and diagnostic indicator (time-domain) display, and the unit currently selected is also used when you generate a report. The envelope is a judgment input, so it is always computed in the sensor’s original unit; trends and grading are unaffected.

Kurtosis saturation warning

When the kurtosis trend falls back from a peak while RMS holds steady or rises, a warning appears under the chart.

Falling kurtosis normally reads as “getting better”, but this combination can mean the fault has progressed to where impacts have become continuous. An early bearing fault raises kurtosis with sharp impacts; as the fault spreads, the impacts run together and kurtosis drops again. Don’t read kurtosis alone as recovery.

Charts

Diagnosis and action details

Press Expand comprehensive analysis (종합 분석 펼치기) to see the rule-based diagnosis. It calls out faults such as imbalance, misalignment, bearing inner/outer race, looseness, and gear issues, and presents supporting evidence (with the indicator each piece came from) alongside action details (how to carry out what to do now, and what would strengthen the verdict, each with its reason).

Action details are not a fixed list: only the items that match the verdict (fault pattern · vibration magnitude · impulsiveness · trend · vibration of unknown origin · missing equipment information) are picked, up to five. When a measurement gate is active, only measurement checks appear and no machine inspection items.

This verdict is computed from a single-channel, single-direction measurement with no phase information, using nameplate rotational speed. Fault mechanisms are hypotheses; a confirmed diagnosis requires multi-axis and phase measurement plus review by a maintenance engineer.

LLM opinion (optional)

With an LLM connected, Generate LLM opinion (LLM 상세 소견 생성) below the comprehensive analysis produces a narrative expert opinion grounded only in the measured values shown in the charts. Pick the provider from the dropdown next to the button; the model used is noted beneath the opinion.

The opinion is written for the equipment owner rather than a vibration specialist, and answers three questions: current state (지금 상태) · why it looks that way (그렇게 보는 이유) · recommended actions (권장 행동). The LLM reads the measurement data with all its technical terms, but writes the opinion in plain words (for example, “BSF sidebands” becomes “a signal from the bearing balls”).

The verdict (grade · cause ranking · actions) has already been set by the engine; the LLM only explains what it means. It never recomputes values or reorders the ranking, closes on a provisional note when the measurement warning is up, never describes items that are undecidable in principle as “unlikely”, and does not inflate the grade. Recommended actions cover only the on-screen what to do now and action details — the LLM does not invent new ones.

Findings are requested in the screen language. For English and Japanese, an output-language instruction is added at the end of the prompt. Small local models may not follow it fully and can mix in Korean.

Without an LLM connected, the rule-based diagnosis stands on its own.

Report

Where you turn analysis into a document.

1. Select the target tags for the report. (nothing is selected by default; select-all is available)
2. Choose a period — presets (1 hour · 6 hours · 24 hours · All) are anchored per tag, to that tag’s latest data; a range set in the calendar applies as-is to every selected tag. If some tags have no data in a custom range, a warning shows where their data actually is.
3. Optionally turn on Include LLM opinion and pick a provider.
4. Press Generate report.

One report file is created for each selected tag. The file name looks like tagname_timestamp.html.

Reports are made in the screen language when generation starts. Changing the language during a batch does not affect that batch. Existing reports do not change, so generate again if you need another language.

Generation runs in four steps — query data → compute indicators → generate opinion → save — with progress shown. If it takes too long, Cancel generation stops it. Finished reports are saved under the signed-in account’s folder (<account>/vibration-reports/, with the account name in lowercase).

Report structure

The Summary (요약) · Full analysis (전체 분석) tabs at the top of the report switch the view. Summary is built for a quick read; Full analysis lays out every chart and all evidence in numbered sections (01 Severity through 05 Action details). The View evidence (근거 보기) and View in-depth analysis (심층 분석 보기) links in the summary jump to the matching section of the full analysis.

Verdicts, evidence and actions in the report are written as terse Korean bullet phrases (개조식). The LLM opinion consists of a one-line key summary plus three parts — current state · why it looks that way · recommended actions — and is marked Provisional (잠정 판정) when the top evidence score is below 70 or a measurement gate is active.

The FFT in the spectrum section is drawn over the default range where the fault frequencies are visible; the Full band (전대역) button widens it to the highest frequency the sensor can measure. In the reference frequency list below the chart (shaft harmonics · bearing fault frequencies), the component the evidence points to is outlined.

Printing outputs the full analysis in the light theme: charts in a single column, the FFT at the range chosen on screen, and the measurement info on one row.

Print with the print button at the top of the report (Print · Save as PDF). A report opened from the app sits in a frame isolated from the console (so the report file cannot reach your login or API token), and the browser’s Ctrl+P prints only that outer frame.

Saved reports are self-contained HTML with no external references, and their charts are live rather than baked images — wheel zoom, drag, and tooltips work, and they redraw to match the theme. Handing over the single file is enough for someone else to explore it the same way.

Each report carries a judgment fingerprint, computed from the judgment engine’s rules. If the fingerprint and the Thresholds cell match but the numbers differ, the data changed; if the fingerprint differs, the package (judgment engine) changed.

Archive

The archive on the right lists saved reports in these columns.

Use the eye button on each row to preview and the trash button to delete (a confirmation appears; deletion cannot be undone). Page through when there are many.

Tips

• Start with one or two tags to confirm behavior, then widen the scope.
• A grade needs RPM and the bearing model under Equipment. For acceleration or displacement sensors, set the unit too — left empty it is treated as velocity (mm/s) and the grade height goes off.
• To see “different from usual” changes as evidence, run baseline learning while the equipment is healthy.
• If monitoring must continue with the app closed, switch to server monitoring.
• Export events periodically with Save for separate archiving.
• When you see an extreme indicator, check the measurement validity warning first — a large share of misdiagnoses come from the sensor, not the machine.

Navigation

• Previous: First setup
• Back to contents
• Next: Alarms & events