It is 2:17 a.m. on a Wednesday. The flotation circuit operator watches the concentrate grade trend crawl toward the lower spec limit for the third consecutive hour. The SPC dashboard shows green control limits on every parameter. Feed grade, reagent flow, pH, air addition, pulp density, mill power draw. All within the static UCL and LCL that were calculated during last quarter's process capability study. The operator knows something is wrong, but the control chart says everything is fine. By shift handover, the lab confirms the concentrate is off-spec. Eight hours of production tagged for reprocessing. The static control limits that were supposed to protect quality became the reason the defect was missed. This is the cost of fixed SPC limits in a process where ore variability, reagent effectiveness, and equipment response change every shift.
- Limits set from a 30-sample capability study at die tryout or circuit commissioning
- UCL and LCL remain unchanged until the next manual recalculation, typically quarterly or annually
- Ore feed hardness shifts, reagent effectiveness drifts with pulp chemistry, mill liners wear, but control limits stay fixed
- Result: the process leaves the control band, but the SPC chart signals green until the defect is confirmed by lab assay
- Control limits recalculated continuously using a rolling window of recent process data (default 100-200 samples)
- UCL and LCL adapt to real-time changes in feed mineralogy, grind size distribution, reagent response, and environmental conditions
- The model distinguishes common-cause variation from assignable-cause drift, so operators investigate only real signals
- Result: limits reflect what the process is doing right now, not what it was doing three months ago
Adaptive SPC limits do not require operators to be statisticians. The logic is engineered for the control room, not the lab. Here is what happens inside the system, and what the operator sees, in the sequence it actually occurs on shift.
Each ore processing circuit has distinct dynamics. The parameters that drift in a grinding circuit are different from those in flotation or thickening, and the adaptive limit model must be circuit-specific to capture the correct interaction effects. The table below maps the adaptive SPC approach for each major circuit stage.
The transition from static to adaptive SPC limits does not require a plant-wide automation upgrade. Most ore processing operations already have the sensor infrastructure and data historian capacity required. The implementation sequence follows a structured path that production teams can execute without disrupting ongoing operations.
The ore processing operations that consistently meet concentrate grade specifications, maintain recovery targets, and pass IATF 16949, AS9100, and ISO 9001 audits without recurring non-conformances share one characteristic: their SPC limits reflect what the process is doing right now, not what it was doing at commissioning. Static control limits are a snapshot of past capability. Adaptive limits are a real-time map of current process behaviour. The difference is the difference between investigating a defect after it is confirmed and preventing it before the next lab assay.
For the operator on shift, adaptive SPC limits mean fewer false alarms, earlier warning of real drift events, and a control chart that makes sense for the conditions in front of them, not the conditions their predecessor faced three months ago. For the metallurgist, adaptive limits mean process capability data that reflects actual operating conditions rather than historical averages. For the quality manager, adaptive limits mean documented, timestamped evidence of process monitoring that satisfies audit requirements without manual chart review.
The infrastructure is already in place. The sensors are installed, the historian is recording, and the operators are watching the screens. The missing layer is the adaptive logic that turns raw measurement data into control limits that change when the process changes. That layer is available now, and the operations that deploy it are the ones that will consistently stay inside specification while their competitors investigate the same defects shift after shift. Book a Demo to see adaptive SPC limits running on your circuit data, or Get In Touch to start the data audit for your operation.
No. Adaptive SPC limits do not widen arbitrarily. The UCL and LCL are recalculated from the actual standard deviation of the rolling window. If the process is stable and centred, the limits may actually tighten because the model has more data to estimate real variation. If the process drifts, the limits follow the drift, but the system flags assignable-cause deviations separately. The Cpk calculation uses the same formula as traditional SPC; the difference is that the limits reflect real-time process conditions rather than a snapshot from months ago. The result is a chart that is more sensitive to real drift and less sensitive to routine process variation. Book a Demo to see adaptive limit behaviour demonstrated on real ore processing data.
Yes. Adaptive SPC maintains separate limit profiles that can be configured for each ore type, feed grade range, and shift combination. The system automatically detects which ore type is being processed and which shift is operating, then applies the appropriate limit set. The day shift operator sees limits calculated from the day shift's process data, and the night shift operator sees limits from the night shift's data. This eliminates the shift-to-shift Cpk reporting variance that plagues static SPC systems and ensures that each operator is working with control limits that reflect their actual operating conditions. Get In Touch to discuss the ore type profile setup for your operation.
Adaptive SPC requires a minimum of 100 data points per parameter to establish the initial rolling window. At typical ore processing data collection rates (one measurement per minute to one per five minutes), this represents 2 to 8 hours of production data. For circuits with longer time constants, such as thickening where changes propagate over hours, a 200-sample window with a wider sampling interval is recommended. No lengthy historical data set is required for activation. The system learns and adjusts as data accumulates. Get In Touch to schedule a data readiness assessment for your circuits.
