Trends & Analytics

The Trends Workbench is where you investigate your data. Plot any signals together, scrub through history, run analysis tools over them, and save what you find. It's a Seeq-style investigation tool built directly on the data Control Seat already collects, so there's no separate data-science stack to stand up.

The Trends Workbench

Open the Trends Workbench and add the tags you want to study. From there you can:

• Plot multiple signals together on synced charts — stack them (one chart per signal) or overlay them on a shared axis. There's also an XY / scatter mode for correlating two signals.
• Pick a time range, relative (now-24h) or absolute.
• Zoom, pan, and scrub with a shared cursor across every chart. Click an event capsule to zoom all charts to that window.
• Work live — charts in the live range refresh on a rolling interval, so an open investigation keeps updating.

Each signal has its own display controls: chart type (line, area, bar, scatter), interpolation (linear, smooth, step), color, Y-axis mode (auto, manual, log), and the time-bucket aggregate used when downsampling (last, avg, min, max, rms).

Conditions & capsules

A condition is a named predicate over your data (for example, "pump above 500 kW"). Where a condition is true, Control Seat paints a capsule — a shaded time band drawn right on the chart. Conditions can come from:

• A formula predicate evaluated over the signal.
• A periodic schedule (calendar-based, e.g. weekday business hours, weekends).
• Hand-picked time ranges (a manual condition).

Capsules let you frame an investigation around events ("show me every time we were in this state") instead of raw time.

Analysis tools

The Tools panel groups more than twenty analysis tools into seven categories. Each one runs over your historical data and can be saved into the worksheet.

Identify — find events.

• Anomaly Detection — flag unusual readings (z-score, IQR, MAD, percentile, auto; plus stuck-value and CUSUM drift).
• Change-point Detection — find step changes in a signal's mean, with the before/after values.
• Periodic / Manual Conditions — build conditions on a schedule or from chosen time ranges.

Quantify — measure.

• Signal from Condition — aggregate a signal over a condition's capsules (avg, min, max, sum, count, rms, duration).
• Statistics — min, max, mean, std dev, slope, R², count over a range.
• Histogram — value distribution and percentiles.
• Rate of Change — average, min, and max derivative of a signal.
• Scorecard — a persisted KPI tile with threshold-based color zones.

Predict — project forward.

• Forecast — extend a signal into the future (see Forecasting).
• Trendline — overlay a linear regression line with R² and slope.
• Multi-variate Prediction — fit one signal as a linear combination of others (OLS or Ridge regression).

Cleanse — process the signal.

• Smoothing — rolling-mean smoothing.
• FFT / Frequency — frequency spectrum with peak detection and a log-scale (dB) option, useful for vibration and electrical analysis.

Compare — match patterns.

• Profile Search — find matches of a chosen template waveform elsewhere in a signal.
• Reference Profile — build an envelope from reference capsules and flag where a signal drifts outside it.

Visualize — annotate the chart.

• Boundary — shade an envelope from an upper/lower tag pair.
• Reference Lines — horizontal setpoints or spec limits.
• Value Bands — shaded ranges for good and alarm zones.

Build — derive new signals.

• Formula — compute a new signal from existing tags, with live preview.

Forecasting

The Forecast tool projects a signal forward and draws a confidence band around the projection. Several models are available:

• Linear and Holt (exponential smoothing) for simple trends.
• Holt-Winters and Fourier for seasonal signals.
• Auto — a grid search across those models that keeps the best fit.
• AutoARIMA — a server-side grid search over ARIMA terms, scored on a holdout window.

Seasonality is detected automatically (and can be overridden), and you control the training window and forecast horizon, with quick presets for +1h, +1d, and +1wk.

Saving an investigation

Investigations persist as worksheets, saved per user:

• A worksheet keeps your signals, layout, time range, conditions, and saved analysis results.
• The journal is a rich-text writeup attached to the worksheet. Use the slash menu to drop in snapshots, tag references, condition references, and AI prompts.
• Pin annotations on the chart and they mirror into the journal.
• Open multiple worksheets in tabs, each with its own undo/redo history.

Ask AI

Ask about your data in plain language. The AI uses the same analytics tools — anomaly detection, change points, distributions, correlation, rate of change — to characterize what's happening and write up what changed. In the workbench, "Ask AI" builds a prompt from the visible signals, time range, and active conditions so the answer is grounded in exactly what you're looking at.

The AI can also operate the workbench for you: add signals, set the time range, create conditions, run analysis, and save the result as a worksheet. Ask it to "investigate why line 1 tripped last night" and it can assemble the whole investigation, not just describe it. (Configure your AI provider in settings first.)

Under the hood

The analytics run on Control Seat's historian. On cloud, that's a per-tenant ClickHouse database; on self-hosted, it's the bundled PostgreSQL. Statistics, distributions, anomalies, change points, rate of change, FFT, forecasting, regression, and pattern search are computed server-side in Go (pushed down to the historian where possible), so no external ML or data-science platform is required.

History for most sources is recorded into that historian. For AVEVA PI and Ignition, history is federated — read live from the source system rather than copied in. See Tags & Data Sources for how each connector behaves.

Next Steps

• Tags & Data Sources — connect data and enable history on the tags you want to analyze
• Dashboards — turn an investigation into a live view to share
• Flows — automate off anomalies and trends with the gateway rule engine
• Scripting — go further with binding, page, and gateway scripts