Back to Cases

Comminution

Grinding & Classification Monitor: what the circuit did

A post-hoc evidence and comparison layer for a grinding circuit. Survey packs give circulating load and size-specific energy against immutable inputs. Multi-month historian exports give operating states, grind compliance, specific energy with its power basis, stability and deterministic events, all filterable by ore domain, liner age and control mode.

Request a Free Session

Demo with made-up data · In development

Who it is for

Built around one practical workflow first

  • Process engineers, plant metallurgists and control-room technical support who need to know what the circuit did, on what data.
  • Reliability engineers and concentrator superintendents comparing periods without a spreadsheet nobody can reproduce.

Workflow pain

Where the work usually slows down

  • A specific energy without its power basis cannot be compared to a published benchmark, and the difference between motor input and shell power is several percent.
  • An interpolated historian export of a heavily compressed tag looks complete and is not, and an I/O timeout read as zero moves every average on the page.
  • Two periods get compared without the ore domain, liner age or control mode that separates them.

Implementation design

What Eleticle builds and controls

What is built

  • A streamed, resumable import for multi-month CSV and XLSX exports, with digital-state parsing, interpolation and compression flags, timezone handling, per-signal resampling and an explicit gap policy.
  • Survey packs validated against a contract, with water-balance circulating load, size-specific energy, P80 and classification summaries stored as versions.
  • Deterministic operating states, mesh-first grind compliance by ore domain, specific energy with a declared power basis, analyser-versus-laboratory pairing, event rules with debounce, and window comparison led by distributions and duration curves.

What data is used

  • Historian exports following the documented export recipe, laboratory sizings, survey packs and context files, uploaded by your team.
  • Your signal catalogue and tag mappings, mesh aliases, state rules, targets by ore domain, time model and delay reason codes.

What AI may do

  • Summarise coverage, observations, context and recorded associations for a window.
  • Draft one labelled review note per due window that separates observation from uncertainty.
  • Answer questions from accepted results and event evidence.

What it does not do

  • Not advanced process control: it never writes to plant controls and never recommends a setpoint.
  • No OPC, PI or DCS subscription; it imports the exports your team uploads.
  • Relationships are labelled as association, never cause, and circulating load appears only as a survey calculation or a labelled proxy.

Human review

It records what the circuit did. People decide what to do next.

The assistant is scoped around clear boundaries: what it can read, what it can prepare, what must be reviewed, and what should stay manual.

Engineers acknowledge, assign, comment on, resolve or suppress events; every action is append-only.
A threshold change creates a new rule-set version and a new analysis run; it never rewrites a historical event.

Before and after

Example workflow change

Before

Circuit performance is argued from a trend screenshot and a spreadsheet that averages in the timeouts.

After

Every KPI carries its coverage, context and formula version, events carry their evidence, and two windows compare on distributions with the ore domain and liner age in view.

More workflows

Industrial operations

Private AI for industrial operations and technical workflows

QA/QC

Assay Guard: laboratory QA/QC review

Testwork

Testwork Atlas: testwork records with provenance

Metallurgy

Met Daily: the daily metallurgical review, signed

Start with one free session

Share the workflow you want to improve. Eleticle will help identify the safest useful first step.

Request a Free Session