Fermentation plants: running blind between two samples

Every media lot behaves differently

What happens. Yeast extract, peptones, corn steep liquor, molasses or plant hydrolysates come from natural sources, and every lot is a little different. The recipe stays the same, yet one batch grows fast and the next lags, foams more or makes less product. Often the lot is only suspected after the batch has failed.

What it costs. Failed or weak batches, lost tank time, investigations and deviation reports, and in the end higher prices for premium or chemically defined media bought to be safe.

Why the usual fix fails. Testing each lot in shake flasks takes time and does not show how it behaves in the production tank. A fixed feed recipe cannot react when the culture behaves differently.

How a fixed-price remote diagnosis checks it. From your batch records with lot numbers, off-gas data, feed and base addition and the final titres I check whether lot effects show up early in the run and which signals reveal them. The fixed-price remote diagnosis tells you whether feeding that adapts to the culture’s actual state can absorb lot variation, and what it would take.

Biomass is not measured online in the production tank

What happens. In the laboratory, cell density probes are common. In the production tank, sterility, validation and cost keep them out, so biomass is known from a sample every few hours: optical density or dry weight, measured after the fact. Between samples the operators work from the recipe and their experience.

What it costs. Feeding, induction and harvest decisions are taken on old information. Batches drift without anyone noticing until the next sample, and some fail.

Why the usual fix fails. More sampling means more work and more risk of contamination, and it still gives snapshots. A probe installed in a large tank needs calibration that drifts with the medium.

How a fixed-price remote diagnosis checks it. The off-gas analyser, feed pumps and base addition already show how the culture breathes and grows. From these logs and your sample values I check how well the biomass can be computed continuously. The fixed-price remote diagnosis tells you whether such a computed value is good enough to run the process on, and what it would take.

Background: Computing unmeasured states for control

Feed profiles run on the clock, not on the culture

What happens. The feed follows a profile fixed during development: so much sugar or methanol at this hour. When the culture grows slower than planned, substrate builds up and the cells switch to by-products such as acetate or ethanol. When it grows faster, it starves.

What it costs. Lower titre and yield, by-products that inhibit growth or burden purification, and variation from batch to batch that nobody can explain.

Why the usual fix fails. A more cautious profile avoids overfeeding but leaves productivity on the table in every good batch. Manual corrections depend on who is on shift and come after the next sample.

How a fixed-price remote diagnosis checks it. From your feed rates, off-gas data, dissolved oxygen, base addition and sample results I estimate how fast the culture was really growing compared with the profile. The fixed-price remote diagnosis shows whether a feed controlled on the estimated growth would keep the culture closer to its optimum, and what it would take.

Oxygen runs short at high cell density

What happens. As the culture gets dense, its oxygen demand climbs. The dissolved oxygen controller raises stirrer speed, then air flow, then back pressure or oxygen enrichment, until everything is at its limit and the dissolved oxygen still falls. Operators then cut the feed by hand.

What it costs. Oxygen limitation means by-products and lower yield. Running stirrer, compressor and oxygen supply flat out costs energy and gas, adds heat that the cooling must remove and drives foaming.

Why the usual fix fails. A cascade of separate controllers reacts only after the oxygen has dropped. Cutting the feed is late and conservative, and a bigger stirrer or compressor moves the limit without managing it.

How a fixed-price remote diagnosis checks it. From your stirrer power, air flow, pressure, dissolved oxygen, off-gas and feed data I check how close the runs come to the oxygen transfer limit and when. The fixed-price remote diagnosis tells you whether predictive control of stirrer, air, pressure and feed together could hold the culture at the limit without crossing it, and what it would take.

Background: Predictive control of process plants

Antifoam goes in generously

What happens. The foam probe triggers a shot of antifoam, and the shots are set large enough that foam never reaches the exhaust filter. Some plants dose on a timer in addition. Nobody knows how much antifoam the culture really needed.

What it costs. Antifoam lowers oxygen transfer, so stirrer and air have to work harder. It costs money, can disturb the culture and makes filtration and purification harder downstream.

Why the usual fix fails. Smaller shots risk a foam-out, a wet exhaust filter and a lost batch, so nobody dares. A second probe tells you when the foam is already high, not when it will be.

How a fixed-price remote diagnosis checks it. From your probe signals, antifoam additions, air flow, stirrer power, pressure and the phase of the run I estimate how the foam builds up and when it really needs treatment. The fixed-price remote diagnosis shows whether dosing on an estimated foam state could reduce antifoam use safely, and what it would take.

Profiles from the laboratory fail in the large tank

What happens. Feed, aeration and temperature profiles that worked in the laboratory and pilot tanks give less product in the production tank. Mixing takes longer, so cells pass through zones of high sugar and low oxygen; dissolved carbon dioxide is higher; the jacket cannot remove heat as fast. The profile is the same, the conditions are not.

What it costs. Months of trial runs in an expensive tank, launches that slip, and a process that runs below the yield promised by development.

Why the usual fix fails. Scaling by one rule, such as equal power per volume or equal oxygen transfer, keeps one condition and changes the others. Trial and error in the large tank is slow and every run is costly.

How a fixed-price remote diagnosis checks it. From your laboratory, pilot and production data, tank geometry and stirrer data I check which conditions differ between scales and which of them explain the yield gap. The fixed-price remote diagnosis tells you whether a scale-up model and control designed for the large tank could close the gap, and what it would take.

Substrate concentration is unknown between samples

What happens. Glucose, glycerol or methanol in the broth are measured from samples every few hours, often with a delay for the laboratory. In between, the concentration can run to zero or climb to a level that inhibits or poisons the cells. The feed pump has no idea.

What it costs. Starvation slows production; accumulation causes overflow metabolism or toxicity, especially with methanol. Either can cost the batch.

Why the usual fix fails. Online substrate analysers exist but are expensive, need a sterile sample stream and maintenance, and are rare at production scale. Feeding more cautiously avoids accumulation and gives away productivity.

How a fixed-price remote diagnosis checks it. The oxygen uptake, carbon dioxide release and base addition react quickly when the substrate runs short or builds up. From these logs, your feed records and your sample values I check how well the substrate concentration can be computed between samples. The fixed-price remote diagnosis tells you whether that value can steer the feed, and what it would take.

The titre is known only after the run

What happens. How much product is in the tank is measured in the laboratory, often a day or more later. Decisions during the run, whether to extend, induce, change the feed or stop, are made without knowing how production is going.

What it costs. Runs stopped too early waste product; runs extended too long waste tank time and may lose product to degradation. A weak batch is noticed only when it is too late to save it.

Why the usual fix fails. Faster assays help but still lag. Taking more samples adds work and contamination risk and gives snapshots, not a trend.

How a fixed-price remote diagnosis checks it. From your growth, feed and off-gas data and the titres of past runs I check how well product formation can be computed during the run. The fixed-price remote diagnosis shows whether a computed titre is reliable enough to base run decisions on, and what it would take.

Background: Computed measurements

Temperature and pH loops oscillate in large tanks

What happens. In a large tank the jacket and coils respond slowly and base takes time to mix in. The controllers were tuned during commissioning, often on water or a quiet phase of the run. In the active phase, temperature swings and the base pump overshoots, so the pH rises and falls around the setpoint.

What it costs. Cells under stress, lower productivity, more base and acid used, and local zones of high pH near the dosing point.

Why the usual fix fails. Detuning the loops calms them down but makes them too slow when the heat load climbs. Tuning by hand during a production run is a risk nobody wants to take.

How a fixed-price remote diagnosis checks it. From your temperature and pH trends, controller outputs, jacket flows and base additions I identify how the tank really responds in each phase. The fixed-price remote diagnosis tells you whether retuning, settings that change with the phase of the run, or model-based control would calm the loops, and what it would take.

The harvest is set by the clock

What happens. The run ends after a fixed time, because that is what the recipe says. Whether product formation peaked hours earlier and the product is already degrading, or whether the culture is still producing, is not known when the decision is taken.

What it costs. Product lost to degradation or cell lysis, or tank time spent on a culture that had stopped producing. Lysed cells also make harvest and purification harder.

Why the usual fix fails. Ending on a sample result comes too late, because the result arrives after the moment has passed. A fixed time is easy to plan but fits only the average batch.

How a fixed-price remote diagnosis checks it. From your run data and the titres and quality results of past batches I check whether the state of the culture at harvest can be estimated during the run. The fixed-price remote diagnosis shows whether harvesting on the estimated state would raise the yield per tank and per week, and what it would take.

Background: Optimal paths and setpoint plans

Chromatography columns are loaded cautiously

What happens. The capture column is loaded to a fixed amount per litre of resin, set well below the binding capacity measured during development. The titre of the feed varies and the resin ages over its cycles, so the safe load is not constant, but the recipe is.

What it costs. Resin is often among the most expensive consumables, and a column loaded below its capacity needs more cycles, more buffer and more time. Overloading would lose product in the flow-through.

Why the usual fix fails. Re-measuring capacity regularly costs time and material. A larger safety margin protects against breakthrough by giving away capacity every cycle.

How a fixed-price remote diagnosis checks it. From your feed titres, ultraviolet absorbance traces, cycle counts and yields I estimate how close each load came to breakthrough. The fixed-price remote diagnosis tells you whether loading on an estimated breakthrough could use more of the resin safely, and what it would take, including what it means for validation.

Membrane fouling ends filtration runs early

What happens. Crossflow and depth filtration steps run at a fixed flux until the pressure climbs to its limit. How fast the membrane fouls depends on cell debris, antifoam and the state of the harvest, which change from batch to batch. Some runs end long before the volume is through.

What it costs. Lower throughput, extra membrane area bought to be safe, more cleaning, and filters replaced before their time.

Why the usual fix fails. A lower fixed flux slows every run to protect the worst one. Oversizing the membrane area costs money on every batch.

How a fixed-price remote diagnosis checks it. From your pressures, flows, permeate volumes and batch data I estimate how the fouling develops and which batch properties drive it. The fixed-price remote diagnosis shows whether flux controlled on the estimated fouling state could get more through each run, and what it would take.

Steam sterilisation in place runs long fixed cycles

What happens. Before every batch, tank and lines are sterilised with steam on a fixed programme: heat up, hold, cool down. The hold time starts when the coldest monitored point reaches temperature, and the whole programme is set long enough for the slowest case.

What it costs. Hours of tank downtime per batch, steam and cooling water, and wear on seals and valves from long hot cycles.

Why the usual fix fails. Shortening the cycle by hand is not an option in a validated process. Adding temperature sensors shows where the cold spots are but does not shorten the heat-up.

How a fixed-price remote diagnosis checks it. From your temperature records of past cycles, steam pressures and valve sequences I check where the time goes, mainly in heat-up and cool-down, and how much is margin. The fixed-price remote diagnosis tells you whether an optimised cycle within the required limits is feasible and what it would take, including the revalidation.

Contamination is detected too late

What happens. A contamination shows up on the agar plate or under the microscope a day or more after it started. By then the batch is usually lost, and the tank and lines must be cleaned and sterilised before the next run.

What it costs. A lost batch, days of tank time, raw materials, an investigation, and sometimes a delivery that cannot be met.

Why the usual fix fails. More frequent sampling adds work and is itself a contamination risk. Alarms on single signals, such as dissolved oxygen or pH, trip too often or too late, because the culture’s own variation hides the signal.

How a fixed-price remote diagnosis checks it. A contaminant changes how the culture consumes oxygen, releases carbon dioxide and needs base, often before anything is visible. From the run data of normal and contaminated batches I check whether such deviations show up early against the estimated state of the culture. The fixed-price remote diagnosis tells you how early a contamination could be flagged, and what it would take.

Continuous fermentation runs far from its washout limit

What happens. In continuous culture the dilution rate decides productivity. Push it close to the maximum growth rate of the cells and they are washed out of the tank faster than they grow. Because nobody knows exactly where that limit lies today, the plant runs at a safe distance.

What it costs. Productivity lost every hour the process runs, a plant that needs more volume for the same output, and a full restart if washout happens anyway.

Why the usual fix fails. The growth rate depends on medium, temperature and the age of the culture, so a fixed limit set during development is either too tight or unsafe. Manual increases are made cautiously and reversed at the first sign of trouble.

How a fixed-price remote diagnosis checks it. From your flow rates, off-gas, biomass samples and feed data I estimate the current growth rate and how far the plant runs from washout. The fixed-price remote diagnosis shows whether control near the limit, on the estimated growth rate, is feasible, and what it would take.

The process moves to a contract manufacturer and is developed again

What happens. A process developed in-house moves to a contract manufacturer. Their tanks have different geometry, stirrers, sensors and control systems. The recipe parameters do not carry over, and engineering runs are needed to find out why the first batches miss.

What it costs. Expensive engineering runs, months of delay, and a transfer that is paid for twice: once in development and once again at the contract manufacturer.

Why the usual fix fails. Copying the recipe parameter by parameter ignores the differences in equipment. Correcting by trial and error at the contract manufacturer costs a tank slot per attempt.

How a fixed-price remote diagnosis checks it. From your development data, the recipe and the equipment data of the receiving plant I check which conditions will differ and which parameters must change to keep the culture’s conditions the same. The fixed-price remote diagnosis tells you whether a model-based transfer could cut the engineering runs, and what it would take.

Start-ups scale up without a control engineer

What happens. Precision-fermentation start-ups have strong biologists and process developers, but often nobody who designs automation, sensors and control for a production tank. The first scale step goes to a pilot plant or a contract manufacturer, and the control questions arrive with it.

What it costs. Scale-up delays, runs lost to control problems, and decisions on equipment and data that are hard to reverse later.

Why the usual fix fails. Hiring a full-time control engineer takes months and is hard to justify before the process is proven. Relying on the equipment supplier or the contract manufacturer leaves the start-up without its own view of the process.

How a fixed-price remote diagnosis checks it. From your current run data, recipes and scale-up plan I check which measurements, control loops and data the next scale step needs, and where the gaps are. The fixed-price remote diagnosis tells you what control work the scale-up requires, in what order, and what it would take.

Cooling, aeration and sterilisation eat the margin

What happens. Chillers remove the heat of fermentation, compressors supply the air, and steam sterilises media, tanks and lines. Each of these systems runs on its own fixed setpoints, often sized and set for the most demanding product, whatever runs today.

What it costs. For bulk fermentation products, where margins are thin, energy can be a large part of the production cost. Peaks in steam and cooling demand set the size of utilities and the power bill.

Why the usual fix fails. Efficiency projects on single units, a new chiller or compressor, help that unit but do not touch how the plant is run. Lowering setpoints by hand risks a batch and gets reversed.

How a fixed-price remote diagnosis checks it. From your utility metering, chiller and compressor logs, batch schedule and fermentation data I check where energy goes per batch and how much follows the process and how much is fixed margin. The fixed-price remote diagnosis shows whether energy-optimal operation of utilities and batches would pay off, and what it would take.

Background: Economic predictive control

Sensor data is not kept, so runs cannot be compared

What happens. The controller shows trends during the run, but the data is overwritten or kept only as a printed batch record. Off-gas, feed and dissolved oxygen of past runs are scattered over spreadsheets, if they exist at all.

What it costs. Nobody can compare a good batch with a bad one, and every problem is investigated from scratch. Models, computed measurements and better recipes all need this data and cannot be built without it.

Why the usual fix fails. Exporting data by hand after each run is forgotten when the plant is busy. Buying a large data platform before knowing which data matters often ends as an expensive archive nobody uses.

How a fixed-price remote diagnosis checks it. From what your control system and laboratory already record I check which signals are needed to learn between runs, how they should be stored and how they can be lined up by batch. The fixed-price remote diagnosis tells you what a lean data layer would take and what it would let you analyse.

Operators learn on real batches

What happens. New operators learn by watching and then running real batches. Rare events, such as a foam-out, a failing feed pump or an oxygen limitation, are met for the first time in production, often on the night shift.

What it costs. Batches lost to mistakes, long training periods, and knowledge that leaves with experienced staff.

Why the usual fix fails. Written procedures describe the normal case, not how the culture reacts when something goes wrong. Generic training simulators do not behave like your process.

How a fixed-price remote diagnosis checks it. From your run data and recipes I check whether a process model can be built that reproduces your tanks well enough for training, including the rare events. The fixed-price remote diagnosis tells you whether a simulator from that model is feasible, and what it would take.

Hybrid models and predictive control of bioreactors stay research projects

What happens. Many teams want hybrid models of their bioreactors, part balance equations, part learned from data, and predictive control built on them. The work starts with a thesis or a pilot project, shows promising plots, and then stalls, because nobody can keep the model running in production.

What it costs. Money and time spent on models that never reach the plant, and processes that keep running on fixed recipes.

Why the usual fix fails. Hiring someone with both bioprocess and control experience is hard; such people are rare. Buying a software package does not tell you whether your data supports the model.

How a fixed-price remote diagnosis checks it. From your run data, sensor list and recipes I check whether a hybrid model can be identified with the data you have, what it could predict, and what predictive control on it would need. The fixed-price remote diagnosis tells you whether the step to the plant is realistic, and what it would take.

Background: Control that adapts to drift

Mid-size fermenters are scarce, so every scale-up trial must count

What happens. Many fermentation projects stall between laboratory proof and commercial investment, because food-grade pilot capacity of middle size is missing (Oost NL). A commercial plant can cost hundreds of millions of dollars, so firms depend on lenders and long-term buyers (FoodNavigator). Slots in pilot and demonstration plants are few.

What it costs. Every scale step costs yield and time, and a trial that fails costs a slot that may not come back for months.

Why the usual fix fails. Going to the larger tank with the laboratory recipe and adjusting there is exactly what scarce slots cannot afford. Booking extra runs as a reserve is expensive and often impossible.

How a fixed-price remote diagnosis checks it. From your laboratory and pilot data and the data of the target plant I check which conditions will change at the next scale and how the recipe and control should be prepared for it. The fixed-price remote diagnosis tells you whether a scale-up model can make the first runs in the larger tank count, and what it would take.