Key Takeaways

  • AI-based thickener control is a published, validated strategy — reinforcement-learning (proximal policy optimization) controllers have reported 10-15% flocculant savings in the literature.
  • There’s no single dominant named “AI thickener” product yet — this capability currently ships as an add-on within general advanced process control (APC) platforms from ABB, Metso, and Yokogawa rather than a standalone thickening-specific tool.
  • The AI target is flocculant dosing and underflow density control — getting the right amount of chemical in at the right time as feed conditions (solids %, mineralogy) fluctuate.
  • Same general closed-loop AI/APC pattern as grinding and flotation control — this is part of a broader plant-wide advanced control adoption rather than an isolated project.

TL;DR

AI thickener control means applying a model-predictive or reinforcement-learning controller to flocculant dosing and underflow density — available today as a module within APC platforms (ABB, Metso, Yokogawa) or via AI-native platforms like Imubit — rather than as a dedicated off-the-shelf thickening product.

How Do I Optimize Thickening With AI?

Thickening is a dewatering step where getting flocculant dosing right matters a lot: too little and you don’t get adequate solid-liquid separation, too much and you’re wasting an expensive reagent while potentially degrading underflow quality. The problem is a genuinely dynamic control challenge, because incoming feed characteristics — %solids, particle size distribution, mineralogy from a changing ore blend — shift continuously, and a fixed dosing rate tuned for average conditions leaves money on the table either way.

Published research applying reinforcement learning specifically to this problem — using proximal policy optimization (PPO) to learn a thickener dosing policy directly from plant data — has reported flocculant savings in the 10-15% range, which is a meaningful reagent-cost reduction at plant scale. That said, be clear-eyed that this is a validated control strategy more than a shrink-wrapped product: there isn’t yet a single named vendor selling “AI thickener control” as a discrete SKU the way TOMRA sells ore sorters. In practice, this capability shows up as a module within the same general advanced process control (APC) platforms already handling grinding and flotation — ABB Ability Expert Optimizer, Metso’s control suite, and Yokogawa’s process automation all offer thickener/dewatering optimization as part of a broader plant-wide APC deployment, and AI-native platforms like Imubit apply the same closed-loop neural-network approach here as they do to grinding.

The practical adoption path mirrors grinding and flotation AI: this is rarely a standalone thickening project. If you’re already engaging an APC vendor or an AI-native closed-loop platform for comminution or flotation, ask specifically what their thickener/dewatering module covers — you’ll typically get more value bundling this in than sourcing a separate point solution.

Try It With Geocluster

Sorting out which APC vendor’s thickener module is actually backed by real deployment data (versus marketing) — and whether a reinforcement-learning approach is mature enough for your plant — is exactly the kind of evidence-gathering Geocluster is built to support.