Plant intelligence for concentrators, wash plants, and smelters.

Where the value is in a processing plant

Recovery and throughput in a concentrator are governed by a small number of interacting variables: mill load and power draw, cyclone pressure and density, flotation air, level, and reagent dosing, thickener bed and underflow density. The historian already records all of them. What is missing on most plants is the step between the trend and the decision.

In-Sight learns each signal's normal operating envelope by operating mode, so a mill running at reduced feed during a reline is not flagged as an anomaly, and a flotation cell drifting toward low froth depth at full feed is. Multivariate detection catches the combinations a single-tag alarm never sees, such as rising mill power with falling density and stable feed rate, which usually means a screen is blinding long before the recirculating load alarm trips.

• Grinding: mill power, bearing temperature, load, cyclone pressure and density, pebble return
• Flotation: air flow, level, froth depth, reagent dosing rate, concentrate grade where assayed
• Dewatering: thickener bed level, underflow density, flocculant dosing, filter cycle time
• Crushing and screening: crusher power, CSS, screen amps, bin levels, belt weightometers

Alarm management for control rooms that are already flooded

Most concentrator control rooms run at several times the ISA-18.2 target of one alarm every ten minutes per operator. In-Sight ingests alarms and events directly from the OPC UA server, measures the alarm system against the ISA-18.2 and EEMUA 191 benchmarks, and ranks the bad actors by contribution. Floods and chattering alarms are detected as they happen, and every alarm is correlated with the findings the detection layer raised on the same equipment at the same time. The two are never merged: an alarm remains an alarm, a finding remains a finding, and an engineer can see both in one timeline.

Investigations that read the plant's own documents

When a finding needs a cause, an engineer opens an investigation case. In-Sight builds a timeline of the affected tags, rolls related findings and alarms into the case, and ranks hypotheses by evidence. Its copilot has read the plant's datasheets, SOPs, and P&ID notes, so when it proposes that a thickener underflow density excursion is consistent with flocculant make-up strength, it cites the page. When it cannot source a claim, it says so rather than guessing.

Engineering limits are extracted from those same documents and proposed to an engineer with page-level provenance. Nothing is armed until a person confirms it. That matters on a mine, where an auto-armed limit from a superseded datasheet is worse than no limit at all.

From finding to work order to proof

A confirmed cause becomes a work order with the evidence chain attached, emailed to the planner or written back to SAP. The finding stays open until the fix is verified against live data, and recurrence is tracked after closure. Shift and monthly reports assemble themselves from live KPIs, charts, and the investigations closed in the period, generated as Word documents on a schedule.

Every confirmation, dispute, and disposition the team records becomes training signal. Baselines recalibrate on confirmed reality and detection precision is published in the app, so the plant can see the system getting better rather than take it on faith.

Built for the realities of a South African mine

In-Sight connects to OPC UA, OSIsoft PI, Modbus TCP, Siemens S7, and the other control and historian systems already on site, backfills history so day one starts with context, and monitors connection health and data gaps on every source. It can run on a single instance inside the plant network or in a dedicated cloud environment, with role-based access and a full audit trail of who decided what.

ExciteData, the company behind In-Sight, has delivered data and engineering work for teams at Royal Bafokeng Platinum, Bokoni Platinum Mines, Modikwa, and Richards Bay Minerals. The product is shaped by that work: it assumes intermittent connectivity, legacy control systems, and engineers who will not trust a number they cannot trace.

Questions we are asked

Which mining plant systems can In-Sight connect to?

OPC UA, OSIsoft PI, Modbus TCP, Siemens S7, EtherNet/IP, MQTT, DNP3, IEC 61850, BACnet, SAP OData, REST APIs, and CSV files. Historical backfill is included so the platform starts with context rather than an empty chart.

Does In-Sight replace the historian or the control system?

No. It reads from the historian and the control network and adds the detection, investigation, action, and learning layer on top. Nothing is written to the control system.

How does In-Sight avoid false alarms on a plant that changes operating mode often?

Baselines are learned per operating mode and recalibrate on confirmed operator feedback, so a reduced-feed campaign or a reline does not look like an anomaly. Multivariate detection also looks at combinations of tags, not single thresholds.

Can In-Sight run inside the plant network without internet access?

Yes. The platform deploys as a set of containers on a single instance, either inside the plant network or in a dedicated cloud environment. AI features can be configured to use an approved endpoint.

What does a pilot look like?

One line or circuit, connected to the existing OPC UA or historian source, with history backfilled. The team brings a persistent problem and we run the full loop on it: detect, investigate, act, and learn, with the results in a report at the end.

Related