Heat Maps, Scatter, and Bullet Charts

Heat Map#

Registered as Heatmap. It shows one measure across two dimensions as a grid of coloured cells — typically hour of day against day, or day against month.

This is the chart for finding patterns you did not know to look for. A line chart of a year of hourly data is an unreadable smear; the same data as a heat map makes the pattern obvious at a glance:

• A cold band every Sunday — the plant does not run weekends
• A dark column at 06:00 and 14:00 — shift changeover losses
• A gradual seasonal drift across months
• A block of anomaly that turns out to be a period when an instrument was faulty

It is also the honest way to display a large set of similar measures — forty tags' utilisation, one row each — where forty lines would be a tangle.

Two things determine whether it works:

The colour scale. A scale spanning the full theoretical range flattens all real variation into one shade. Set the range to the operating window so genuine differences are visible.

Sequential, not rainbow. A single-hue sequential scale reads as ordered. A rainbow scale creates false boundaries where the hue changes and misleads readers about where the meaningful thresholds are.

Scatter Chart#

Plots two measures against each other rather than against time, so it answers "does this relate to that?"

Real uses:

• Energy consumption against production rate — is the plant more efficient at higher throughput?
• Reject rate against line speed — where does quality start to suffer?
• Yield against a process temperature — is there an optimum?
• Consumption against ambient temperature — how much of the load is weather?

The scatter chart is where process optimisation arguments get settled, because a cloud of points showing a relationship is far more persuasive than two trend lines that happen to move together.

Two cautions worth stating out loud when presenting one:

Correlation is not causation. Two measures both driven by production volume will correlate with each other and neither causes the other.

Both measures must be at the same interval, and the interval matters. An hourly scatter includes transient states — ramping up, shutting down — that muddy the relationship. A daily one is smoother but hides the operating detail. Try both.

Bullet Chart#

Registered as Bullet. A compact horizontal bar showing a value against a target marker, with qualitative bands behind it.

It is the most space-efficient way to show actual against target, and it is the right answer for a column of KPIs where a set of gauges would consume the whole screen. Ten bullets stack into the space of two gauges and are easier to scan, because the eye follows one axis.

Use bullets when the question is "are we hitting target across these ten things". Use a widget when there is one number that matters most. Use a gauge when you want deliberate emphasis on a single value and have the space to spend.

Event Based Timeline#

Events drawn as coloured bars along time. Covered in Analysing Data by Event, and worth naming here because it belongs in this group: it is the chart that makes a shift's story legible at a glance — runs, stoppages, and changeovers as a coloured strip.

Colours come from the event attribute's enumeration, so a deliberate palette makes it readable without a legend.

Comments Chart#

Displays commentary alongside data. This is how the explanation gets onto a dashboard rather than living in someone's email.

It matters more than it appears. A dashboard showing an unexplained dip generates the same question every morning until someone answers it. A comments chart lets the answer live next to the data — "planned maintenance, 14 August" — and the question stops being asked.

Where reporting is reviewed in a meeting, a comments chart is often what turns a set of numbers into a record of decisions.

Pass-Through Chart#

A chart that passes data through rather than rendering its own visualisation. Useful in composition, where another element consumes the data.

The layout components#

Four chart types exist to build the dashboard itself rather than to display measures:

Text is the one to use more. A dashboard with a clear heading and a one-line statement of what it is for gets used; an unlabelled grid of charts gets abandoned. It also carries the things a chart cannot: what "good" looks like, who owns the numbers, when they update.

Navigation is what makes a set of dashboards into a system. A summary dashboard whose panels link to area detail lets a reader follow a problem down instead of hunting through a tree.

Web Page embeds something outside Infohub — a SCADA screen, a supplier portal, an internal wiki page. Useful, and worth being conservative with: an embedded page that requires its own login, or that is slow, degrades the whole dashboard.

Map suits genuinely distributed operations — a water utility's pump stations, a mining group's sites, a fleet of remote installations. For a single plant, a map is decoration.

Choosing between them#

Diagnosing#

See also#

• How Infohub Charts Work
• Chart Type Catalogue
• Dashboard Layout Principles
• The Time Based Chart