Deterministic Edge-AI for Layer 4 Autonomous Mill Control.
Moving plant operations from reactive post-mortem lab assays to continuous, feedforward metallurgical governance. GlassBox MINER integrates raw sensor physics directly into edge networks to maximize Net Smelter Return (NSR).
GlassBox ORE: The Sense Layer
Conventional plants fly blind, relying on delayed 2-hour lab assays and visually estimated block models. GlassBox ORE utilizes a multi-spectral physical layer to intercept massive, high-frequency raw data directly from the primary feed.
To eliminate vertical and horizontal particulate segregation (the "Brazil Nut Effect"), we deploy a heavy-duty mechanical cross-belt sampler. This extracts a true volumetric cut of the 60-inch SAG feed burden, miniaturizing it onto a highly stable 30 cm presentation bypass belt, maximizing the metrological fidelity of the optical arrays.
- High-Frequency Cross-Belt LIBS: Firing at 100–1,000 Hz, Q-switched nano-lasers vaporize microscopic surface points to capture light gangue, silicates, and ore hardness proxies (Si, Mg).
- Continuous Cross-Belt XRF: Penetrates the top layer of the bypass burden to establish a highly stable, energy-dispersive baseline for core heavy base metals and penalizable elements (Fe, Cu, S, As).
- Bypass Presentation Integrity: Mounted on a vibration-isolated slider-bed conveyor to ensure absolute focal depth precision for plasma optical emissions.
- Raw Spectral Ingestion: Bypasses vendor-supplied "black-box" smoothed SCADA averages. The IPC ingests uncompressed UDP/TCP optical spectra and photon counts directly from the sensor APIs.
- PRISMA Structuring Engine: Executes high-speed block-matrix mathematics to evaluate structural covariance. It isolates and amputates anomalies (water droplets, empty belt space) to flatten 60,000+ erratic laser strikes per minute into a clean bulk average.
- Oculus Robust Optimizer: A continuous dual-loop inverse solver running at the edge to track variance and apply real-time affine offsets, neutralizing physical sensor drift instantly.
PRISMA: Edge-Native Signal Processing
Industrial PLC historians are designed for 1 Hz scalar telemetry and will instantly crash if fed uncompressed optical emission arrays. GlassBox architecture solves this by intercepting the physics layer at the edge.
Before any chemistry data touches the plant network, the localized, IP67-rated Schneider Electric Harmony Edge Cluster mathematically scrubs the raw tensors. The output is a deterministic, metrologically sound compositional average handed off seamlessly to the plant SCADA.
From Data to Layer 4 Actuation
Raw intelligence is useless without autonomous execution. The GlassBox suite transforms real-time structural insight into immediate feedforward physical control—protecting capital equipment and maximizing in-circuit recovery.
Agnostic DCS Integration: Plant historians are built for 1 Hz scalar data, not uncompressed optical arrays. GlassBox MINER runs the heavy physics at the extreme edge, handing off only the refined, deterministic intelligence. This guarantees zero-friction OPC-UA/Modbus-TCP integration with legacy plant nervous systems, including Emerson DeltaV, AVEVA PI / System Platform, and Rockwell Automation.
| Sub-System | Operational Function | Edge Computation / Method |
|---|---|---|
| GlassBox REACTOR (The Act Layer) |
Autonomous Mill & Flotation Control | Governs Layer 4 control logic. By identifying hard silica gangue spikes via upstream LIBS, REACTOR computes and writes feedforward setpoints directly to the plant DCS (e.g., dropping SAG feed rates by 5%) before the ore hits the mill, while strictly preserving human operator override states. |
| GlassBox HEAP (The Think Layer) |
Hydrological Stability & Recovery | Utilizes 3D poroelastic Finite-Difference Time-Domain (FDTD) lattices to map fluid saturation inside leach pads. Correlates piezometer telemetry to predict localized slumping risks and pregnant solution pathways. |
| GlassBox ASSET (System Health) |
Predictive Fatigue Monitoring | Monitors sub-threshold vibrational frequencies across heavy crushers and HPGR circuits to mathematically isolate the signature of component mechanical fatigue weeks before catastrophic failure occurs. |