Direct answer

A repeated, comparable upstream measurement of settling response is required, together with a baseline separated by material regime and a temporal association with the subsequent thickener response. The early signal can use the complete h(t) curve and derived variables, not only one final velocity.

Detecting a change does not yet demonstrate that it will affect the thickener. That claim requires evidence of lead time, a relationship that holds across regimes, sampling quality and the ability to distinguish a real material change from a deviation in the test method.

Comparability comes before the alert

A difference between two tests is interpretable only when relevant conditions were controlled or recorded. Samples must be compared on a defined basis of solids, water, flocculant, mixing, temperature, geometry and interface criterion.

Temporal representativeness must also be evaluated. A spot sample may not capture feed variability during the period that later appears in the thickener. Sampling design must be related to flow, stream mixing and the delay to the equipment.

The measurement system must be monitored before the process is monitored. A check standard, reference sample or duplicate tests can help detect drift in the camera, lighting, timing, preparation or interface interpretation.

Features worth tracking

  • the complete interface-height versus time curve;
  • velocity or slope over a previously defined interval;
  • time to reach one or more reference heights;
  • final height within a specified time window;
  • curvature changes or transition towards compression;
  • supernatant clarity or turbidity when a traceable method exists;
  • dispersion between replicates and the percentage of uninterpretable tests;
  • sample, water, reagent, mixing and operator or instrument metadata.

Not every feature should become an alarm. First identify which features are repeatable and which provide information beyond the available feed variables.

Build a baseline by regime

One historical mean can mix different mineralogies, particle-size distributions, concentrations or water qualities. The baseline should separate regimes that have an operating explanation and enough data to estimate their natural variation.

Control charts, EWMA or CUSUM are possible tools for detecting changes relative to a reference. Their use requires review of independence, distribution, sample size and baseline stability. A statistical limit is not a thickener operating limit.

The initial phase characterises historical behaviour and removes assignable causes in the measurement method. The monitoring phase compares new results against that reference. If the protocol or reference material changes, baseline continuity must be reassessed.

Link the early signal to the thickener

The analysis must reconstruct a common sequence: sampling time, test time, material transport, estimated arrival at the thickener, and the evolution of bed, torque, overflow, underflow and operating actions.

Useful lead time is the interval between reliable detection and the first process response that permits action. It must be estimated from observed data, not only nominal hydraulic residence time. Mixing, recirculation, inventory and signal filtering can alter it.

To evaluate usefulness, compare episodes with and without an alert and record false positives, missed changes and conditions in which the relationship fails. A retrospective correlation is not enough to automate a decision.

Procedure for investigating an alert

  • confirm that the test followed the protocol and that its replicate is coherent;
  • review whether concentration, particle-size distribution, mineralogy, water or reagent changed;
  • check the condition of the measurement system and sample;
  • estimate when the represented material would reach the thickener;
  • observe the subsequent response without confusing it with operator actions;
  • classify the event as a material change, preparation change, measurement problem or unresolved cause;
  • update the baseline only through a documented rule.

The alert output should be a testable hypothesis and an observation window, not an automatic dosage or discharge instruction.

Common errors and limits

  • comparing tests that use different protocols;
  • mixing all materials into one population;
  • using velocity as the only feature;
  • changing limits after every event until alerts disappear;
  • ignoring false positives and missed changes;
  • claiming anticipation without demonstrating the delay to the thickener;
  • turning a laboratory alert into a control action without plant validation.

This page describes a detection architecture. It does not specify sampling frequency, statistical limits, setpoints or operating actions. Anticipation depends on representativeness, repeatability and the specific dynamics of the circuit.

Sources and evidence status