All case studies

Case study · Corn wet milling

One pound of starch per bushel is 160 tons a day.

An assessment prepared for a corn wet-milling facility. At this scale small upstream losses stop being small, and by the time they reach wastewater the product is gone. So AquaMesh put NIR starch detection on the process stream and wired it straight into the operational AI — the sensor sees the starch leaving, the platform tells the operator which knob to turn.

Corn processed
320,000 bushels a day
Starch equivalent moving
10.2M lb a day
Water
20.5M gallons a day
One lb per bushel
160 tons a day

Throughput and water figures are publicly reported operating scale. Starch quantity uses the EPA average wet-milling yield. No actual loss rate is assumed for this facility.

Where the problem shows up

Find the loss before it reaches wastewater.

In starch processing, small upstream losses become lost yield, excess water load, higher energy use and more treatment demand. Each step has its own way of letting product escape.

Six process steps and how each loses starch: separation (hydrocyclone and screen drift let fines escape), washing (multi-stage washing can carry product into overflow), CIP (rinse steps send recoverable product to drain), dewatering (moisture variability raises dryer load and hides loss), drying (excess water carried forward becomes avoidable energy cost) and wastewater (lost starch raises COD/TSS load, sludge and treatment burden).

Wastewater is where the problem appears. The value is in finding where it started. Lab results come back hours later, when the load has already moved through the plant.

The two halves, integrated

NIR reads the starch. The AI says what to do about it.

Neither half is much use alone. A spectrum nobody acts on is a chart; an AI with no starch signal is guessing. AquaMesh ships them as one chain.

  1. 1

    Connect what exists

    Historian, PLC/SCADA, lab results, flow, pressure, temperature, conductivity, turbidity, utilities, production and CIP records.

  2. 2

    Add NIR where the process is blind

    AquaSpectra™ with the NIR window reads the starch signature directly in the stream, alongside solids, turbidity and organic load — placed only where nothing measures product today.

  3. 3

    Learn what drives the loss

    The platform ties the NIR signal to feed variability, machine behaviour, wash conditions and dewatering, so a rise points at a cause instead of an alarm.

  4. 4

    Hand the operator an action

    Recommendations first, in the operator's language, with the value attached. Only after validation can AquaMesh execute approved changes inside the plant's own limits — and verify the result.

The measurement

Two windows. One fingerprint.

AquaSpectra captures UV/Vis (190–440 nm) and NIR (900–1700 nm) spectra every few seconds. The NIR window is what sees starch directly. A one-time model of the plant's own product becomes the reference, and the platform watches for deviation from it rather than trying to hold an absolute calibration.

Three panels: UV/Vis and NIR absorbance spectra at clean, rising, high and very high starch loss; a reference starch fingerprint across 200–1700 nm; and spectral deviation from that reference for small, moderate and large changes.
Because the model tracks change rather than absolute values, it is insensitive to drift and calibration, and it holds across recipes, seasons and raw materials.
Example indicators, expressed as deviation from the plant's own reference
IndicatorDeviationWhat it means
NIR starch signal (900–1700 nm)↑ 2.8 σDirect starch signature increase
Starch Loss Index (yield protection)↑ 2.6 σStarch escaping to wastewater
fDOM (UV254)↑ 2.1 σHigher soluble organics
COD/BOD proxy (TOC + UV254/SAC254)↑ 1.9 σRising organic load
Turbidity / solids↑ 1.4 σHigher suspended solids
High-COD wastewater per tonne of starch produced
10–20 m³
Typical secondary effluent
200–250 mg/L COD
COD removal in healthy anaerobic digesters
>89%
Lag on lab results while the load moves
Hours

Industry reference figures, not measurements from this facility. Sources: Anaerobic Digestion Starch Wastewater Guide; iFactory (BOD, COD & compliance); LAR corn-milling TOC monitoring; MDPI 13(10):1762 flocculation study.

In the platform

What the operator actually gets when the NIR signal moves.

Not a spectrum. A named cause, the knob to turn, the tags that prove it, and what the change is worth per hour — with accept, resolve and dismiss on the same card.

NIR reads the stream

The starch signature, every few seconds, in-situ — no sample, no lab queue.

The model names a cause

Deviation from the plant's own reference, tied to the unit that moved first.

The operator gets an action

One setpoint, one shift to observe, inside limits the plant already set.

The result is verified

The same signal confirms whether the yield actually came back.

The proposal

Start small. Prove value in 90 days.

One process area, quantified: what the losses are, what they cost, and whether the recommendations move the number.

  1. 1

    Days 1–30 · Establish the baseline

    Connect to existing data, map the process and define KPIs, quantify current losses and variability, identify the high-impact opportunities.

  2. 2

    Days 31–60 · Find and prioritize

    Detect and rank the top loss mechanisms, correlate drivers across process, quality and effluent, deliver real-time insight, validate with operators and lab data.

  3. 3

    Days 61–90 · Prove and plan

    Quantify the impact of the recommended actions, build the economic case, define the scope for control and sensing, and lay out the deployment roadmap.

No automated control in the pilot. Insights and recommendations first; control comes only after validation and approval.

What is your process losing?

The same assessment starts with your data: the historian you already keep, your lab results, and the process you already run.

Anonymized case study built from publicly reported operating scale and industry references. Deviation figures are illustrative of the method, not measurements from this facility.