FROM PLANT DATA TO PLANT DECISIONS: How Dexian is revolutionizing bioenergy with AI

India’s bio-energy industry has spent years investing in machines, sensors, control systems and increasingly sophisticated plants. What is pertinent here is that the next opportunity may not require another piece of equipment. It may lie in making better use of the data that plants are already generating.

This is where Dexian is positioning its AI technology. The company’s approach is built around a simple idea – industrial plants do not necessarily need more data; they rather need to transform existing data into better decisions on the shop floor. The AI layer of Dexian simply sits above the plant’s existing distributed control system (DCS) and advanced process control (APC) to identify gaps that conventional control systems doesn’t seem to address today.

The opportunity is particularly significant as India’s bio-energy industry moves beyond simply adding capacity. Ethanol production, CBG, cogeneration and other bio-energy operations are increasingly being judged by how efficiently they convert feedstock into usable energy and products.

Moving from Capacity to Efficiency

India has achieved nationwide E20 ethanol blending, while the broader bio-energy ecosystem is expanding into compressed biogas and sustainable aviation fuel.

Yet installed capacity does not automatically translate into production or profitability. The industry is increasingly focused on utilization, recovery and process efficiency.

For Dexian, this changes the role of industrial data.

A DCS can regulate what is continuously measured. But several important decisions still depend on laboratory readings taken only periodically, operator judgement or information scattered across different units. Losses can also accumulate between interconnected processes without being visible as a single problem.

Dexian’s AI layer is designed to work in these gaps rather than compete with existing control systems.

In simple terms, the DCS keeps the plant stable, while Dexian’s AI attempts to identify where value is being lost and recommends what can be done about it.

AI built around existing plant data

Dexian’s work spans sugar, ethanol, grain distilleries, cogeneration and CBG operations.

Across its engagements, the company has connected control-system and historian data with enterprise and laboratory information. One program covers 13 integrated sugar, cogeneration and distillery plants, averaging around 15,000 tonnes of cane crushed per day. Other engagements include distillery operations covering fermentation, distillation and dehydration, cogeneration and biorefinery systems, a 9,500 TCD milling assessment and a refinery crystallization operation with seven months of one-minute data from 10 vacuum pans.

The objective is not simply to collect this information. It is to turn it into measurements or recommendations that operators can use.

In sugar mills, for example, Dexian’s models can use roller speed, torque, imbibition flow and cane rate to estimate pol and moisture in bagasse between laboratory tests. The system can then recommend roller-gap and imbibition-water settings.

Similarly, where Brix is measured in the laboratory only every few hours, a soft sensor can estimate it every minute, giving operators visibility during the periods when no laboratory reading is available.

The same principle extends to crystallisation. A cut-point advisor can combine the Brix estimate with valve conditions and seeding information to provide a confidence score and a simple go-or-no-go recommendation.

Extending the model beyond sugar

The technology is not limited to sugar recovery. In cogeneration, Dexian models steam demand, boiler conditions and turbine export to identify opportunities for energy optimization. Since steam saved in one part of the plant can translate into additional power available for export, the value can extend across the system.

In distilleries, the AI layer can assess fermentation conditions such as yeast pitch, wort Brix, temperature, pH and carbon dioxide evolution to predict ethanol yield before a batch is completed.

It can also analyze distillation and dehydration parameters, including reflux, tray temperatures and molecular-sieve cycling, against energy consumption and first-pass release.

For CBG plants, the same architecture can be applied to digester stability. Feedstock mix, temperature, pH, volatile fatty acid-to-alkalinity ratio and gas composition can be used to predict methane yield before the digester moves away from its desired operating range.

The common thread is that the AI does not require the plant to start from scratch. It works primarily with data already being generated by existing systems.

First, make sure the data can be trusted

Dexian’s experience also points to a less glamorous but critical part of industrial AI: data quality.

In assessments across Indian plants, the company found recurring problems such as incomplete historian data, missing links between process, laboratory and dispatch records, and physically impossible values being recorded as valid data.

In one plant’s historian exports, Dexian identified 55 distinct data inconsistencies, including 1,249 Brix readings above 95 degrees. The daily data-fault rate at one point exceeded 20 percent.

This makes data readiness an important first step.

Dexian’s philosophy is to establish whether the available data is reliable before attempting to build sophisticated models. The logic is straightforward: an AI model trained on bad data can end up automating existing errors rather than solving them.

From model to measurable plant-floor insight

The company’s work has already produced several specific findings.

In one crystallisation application, stirrer current was found to distinguish between a standard cut and a shifted cut, with readings of 71.4 amps and 66.0 amps respectively. The finding was statistically significant and led to a rule requiring stirrer current above 69 amps for the cut gate.

A Brix soft sensor was also developed and validated at 65 percent accuracy against laboratory references, with a pathway identified to improve accuracy to 85 percent as data coverage expands. Five operating rules were subsequently derived and handed back to the plant for enforcement through its existing DCS.

These are relatively narrow interventions, but that is precisely the point. Dexian’s approach is to identify specific operational decisions where data can produce measurable improvement rather than promise blanket automation.

(The author is Senior Director – Head of Sales at Dexian)