In most DLE plants, the electrode stack and the brine feed inlet are physically close to each other, but informationally distant. The electrode responds to what the brine was when it left the inlet sensor, not what it is right now at the electrode face. The gap between those two measurements is where most predictive control value lives.
Transit Time Is the Core Problem
When brine enters the DLE cell at the feed inlet, it takes a finite time to reach the electrode stack. That time, the hydraulic transit time, depends on the distance from the inlet sensor to the electrode bed midpoint, the volumetric flow rate, and the cross-sectional area of the flow path. In a compact DLE module operating at 3 to 5 cubic meters per hour, the transit time from inlet to electrode midpoint is typically 8 to 20 minutes. In a larger pilot-scale cell at lower space velocity, it can be 25 to 40 minutes.
During that transit time, the brine composition at the inlet can change. Seasonal brine variability in a salar feed changes slowly enough that 20 minutes of lag is insignificant. But the short-cycle variability sources that matter most for electrode control, including mixing dynamics at the evaporation pond outlet, rain events that dilute the surface brine layer, and variability introduced by upstream pre-treatment processes, operate on timescales of minutes to hours. A composition shift at the inlet that takes 20 minutes to propagate to the electrode will affect cycle performance if the control system is not adjusted before the brine front arrives.
Fixed-setpoint control cannot account for this at all. It applies constant electrode settings regardless of what is happening at the inlet. The first indication something is wrong comes from cycle efficiency or product purity metrics, which respond only after the variable brine has already been processed through the electrode. By the time a fixed-setpoint system operator notices a purity excursion and adjusts the setpoint, several cycles of off-spec production may have already occurred.
Leading Indicators in Upstream Process Data
The inlet sensor readings are leading indicators for electrode performance precisely because they arrive 8 to 40 minutes before the brine they describe. That predictive window is the resource a feedforward control strategy uses.
The signals we track upstream for this purpose are Li+ concentration, Mg/Li molar ratio, conductivity (as a proxy for total dissolved solids and ionic strength), temperature, and pH. Each of these affects a different aspect of electrode behavior: Li+ concentration drives the absolute loading rate per unit time; Mg/Li ratio drives selectivity behavior as discussed in the voltage gradient article; conductivity affects the electrical double layer properties and the apparent electrode potential; temperature affects intercalation kinetics and equilibrium constants; pH affects fouling risk and, for some electrode materials, the thermodynamics of Li+ intercalation directly.
Not all of these signals are equally independent. In most brines, conductivity correlates with TDS, which correlates with individual ion concentrations, so conductivity is often a useful proxy for detecting a brine composition shift even before the individual ion measurements confirm the nature of the shift. A conductivity step change at the inlet, for example, triggers a provisional forecast of what Li+ and Mg/Li values are likely incoming based on the historical correlation between conductivity and those parameters for that specific brine source. The ion-selective electrode measurements, which respond slightly slower, then confirm or correct the forecast.
The Predictive Control Model
The feedforward model in the EELI system takes current inlet sensor readings and applies the hydraulic transit time to project what the electrode will be processing at its current flow rate in T minutes, where T is the estimated transit time based on current flow rate. It then evaluates whether the projected inlet composition requires a change to the current electrode setpoint, executes that setpoint change T minutes before the new brine front arrives, and monitors actual electrode performance as the brine arrives to validate the forecast and update the model if the actual performance deviates from prediction.
The transit time estimation itself has a feedback correction. The predicted transit time is based on nominal cell geometry and measured flow rate. But actual transit time can vary due to channeling in the packed electrode bed (preferential flow paths that carry brine faster than average), temperature-induced density gradients in the brine, or partial bed plugging from fouling that changes the local flow path. The model compares the timing of predicted composition changes (based on when the inlet saw a shift) with the actual timing of performance changes in the electrode (which respond when the new brine arrives), and adjusts the transit time estimate accordingly over successive cycles.
This self-calibration is important for production accuracy. A transit time estimate that is 5 minutes too short means setpoint adjustments arrive 5 minutes after the brine front has already been at the electrode. A 5-minute overshoot in a 15-minute adsorption phase is a 33% lag relative to the phase duration. Over time, the self-calibrating transit estimate converges to within 1 to 2 minutes of actual transit time in most cell configurations, which is close enough to provide useful feedforward control.
Brine Feed Variability Patterns by Source Type
The variability patterns differ significantly across brine source types, which affects how the feedforward model is parameterized for each.
Continental salar brines from evaporation ponds tend to show two distinct variability timescales: slow seasonal variation (months, driven by evaporation rate and rainfall) and faster day-to-week variation driven by stratification within the pond, mechanical mixing from wind or pump operations, and the specific location and depth of the extraction intake. The slow seasonal component is handled by periodic recalibration of the base setpoint schedule. The faster daily variation is handled by the real-time feedforward model.
Geothermal brines show different variability: fluid composition changes are slower (geothermal reservoir chemistry is more stable than a surface pond), but temperature variation at the surface is larger and faster, because geothermal fluid cools as it moves through surface piping and seasonal ambient temperature affects the cooling rate. For geothermal sources, temperature is often the dominant feedforward variable, with composition following more slowly behind it.
Produced water from active oil fields shows the most rapid variability, driven by changing well production rates, manifold switching as wells are brought on or taken offline, and chemical treatment programs (scale inhibitors, corrosion inhibitors) that affect the brine chemistry within the produced water collection system. For produced water sources, the feedforward model benefits most from high-frequency monitoring (sub-minute) at the inlet, because the composition changes can propagate faster than a 2-minute sampling interval would catch.
What Upstream Process Data Cannot Predict
There are limits to what can be predicted from inlet data alone. Conditions that develop inside the electrode bed, including the gradual accumulation of fouling deposits, the slow capacity fade of the electrode material over its service life, and localized channeling or dead zones that develop as the electrode ages, are not visible in the inlet sensor stream. These require independent monitoring signals from within or at the outlet of the electrode bed itself.
The design of the monitoring architecture reflects this. Inlet sensors handle the feedforward prediction function. Effluent sensors at the electrode outlet handle the feedback correction function, detecting discrepancies between predicted and actual electrode performance. The combination of both makes the control model more robust than either alone. Inlet-only control (pure feedforward) misses electrode-internal degradation. Outlet-only control (pure feedback) sees the consequence only after it has happened. The combination allows the system to anticipate most variability in advance and catch what it could not anticipate in time to correct before the next cycle.