Steel mills: scrap quality, furnace power and casting by rule

Scrap quality varies, and copper cannot be removed

What happens. Electric steel is made from scrap, and scrap is a mix. Its composition is known only roughly. Copper, tin and other elements stay in the melt, because the furnace cannot remove them. They show up later as cracks in the rolling mill or as a steel that has to be sold as a lower grade.

What it costs. Downgraded heats, cracks and rework in the rolling mill, and the price difference between good and poor scrap, which is paid in advance by the charge decision.

Why the usual fix fails. Buying only clean scrap is expensive and does not always exist. Diluting with pure iron costs money and energy. A safety margin on the charge mix makes every heat more expensive to avoid the rare bad one.

How a fixed-price remote diagnosis checks it. From charge records, scrap grades and analyses of the finished heats I check how well the composition of a charge can be computed beforehand. The fixed-price remote diagnosis answers whether a charge mix computed against the target composition lowers the cost of materials without raising the risk.

Melt temperature and composition are known only from dip samples

What happens. In the electric arc furnace, the temperature is measured by a probe that is dipped now and then, and the composition by a sample sent to the lab. Between two samples, the furnace runs blind. The tap is decided on the last value and on experience.

What it costs. Power-on time that is longer than needed, delays while waiting for the lab result and heats that are tapped too hot or too cold.

Why the usual fix fails. More dip samples interrupt the process and use probes that cost money. A fixed time per heat ignores the scrap. Waiting for the lab result stops the furnace.

How a fixed-price remote diagnosis checks it. From the furnace logs (power, electrode position, off-gas, slag practice, samples) I check how well melt temperature and composition can be computed between the samples. The fixed-price remote diagnosis answers whether a computed melt state can guide the tap and the power profile.

Background: Computing unmeasured states for control

The power profile of the arc furnace follows the recipe, not the cost

What happens. The furnace power is set by a recipe: steps of voltage and current by time. The recipe fits an average charge. The price of power changes through the day, the scrap burns differently each time, and the electrodes and the roof wear differently with each profile.

What it costs. Power at a higher price than needed, electrode consumption and refractory wear, and a longer time to tap when the recipe does not fit the charge.

Why the usual fix fails. A recipe per scrap type is a start and cannot cover all mixes. Moving melting to the night is limited by the logistics of the mill. Hand changes by the operator depend on the shift.

How a fixed-price remote diagnosis checks it. From furnace logs, price series and the data on electrode and wear I check how much a profile that considers the price and the state of the melt could change. The fixed-price remote diagnosis answers whether an economic power profile is feasible within the limits of your furnace and your grid contract.

Background: Economic predictive control

Heat losses of the ladle during waiting are unknown

What happens. After tapping, the ladle waits for the ladle furnace or the caster. Steel cools in that time, faster in a cold ladle, slower in a hot one. The wait varies with the schedule, and the temperature at the next stage is a surprise.

What it costs. Reheating in the ladle furnace with electricity, a cast temperature that is too high for safety and heats that must be held back because they arrive too cold.

Why the usual fix fails. A temperature measurement is possible only now and then. A fixed addition of temperature at tapping is a guess for an average wait. Insulating lids help, but only on the ladles that have them.

How a fixed-price remote diagnosis checks it. From logs of tapping temperature, waiting times, ladle history and later measurements I check how well the temperature of the steel in the ladle can be computed in between. The fixed-price remote diagnosis answers whether a computed ladle temperature can reduce reheating and the reserve at tapping.

Background: Computing unmeasured states for control

The mould level in continuous casting oscillates

What happens. In the mould of the caster, the level of the steel must stay constant. The level oscillates, driven by the flow of steel, the stopper or slide gate and bulging of the strand. The operator sees the level swing and the controller working against it.

What it costs. Surface defects and inclusions in the slab, which cost grinding, downgrading or rejection, and a lower casting speed to keep the level calm.

Why the usual fix fails. Tuning the standard controller harder makes it nervous and the valve wears. Slowing the caster is certain and costs output. Dampers and electromagnetic brakes work at the hardware level and do not remove the cause.

How a fixed-price remote diagnosis checks it. From level, valve and speed recordings I check which disturbances drive the oscillation and how much a predictive level control, which knows the valve and the delay, could calm it. The fixed-price remote diagnosis answers whether the control can be improved without a change of hardware.

Background: Predictive control of process plants

Shell thickness and the end of the liquid core in the strand are unknown

What happens. Inside the cast strand, the steel solidifies from outside in. How thick the shell is at the mould exit and where the liquid core ends cannot be measured. Secondary cooling sprays water by a table per grade and speed. When the speed changes, the table is out of date.

What it costs. Breakouts, where liquid steel runs out of a thin shell, are a severe risk and a long stop. Over-cooling and under-cooling cause cracks and segregation, which come out as quality problems later.

Why the usual fix fails. A cautious casting speed reduces the risk and costs output. A table of cooling per grade ignores the transients. A sensor in the strand is not possible in the harsh zone.

How a fixed-price remote diagnosis checks it. From the casting logs (speed, temperatures, water flows, quality outcomes) I check how well a solidification model can compute shell and crater end and how the cooling could follow. The fixed-price remote diagnosis answers whether a computed strand state can control the secondary cooling and warn of risk.

Background: Computing unmeasured states for control

Hot-rolled thickness varies with temperature and hardness of the strip

What happens. In hot rolling, a strip that is a little colder or harder than the average leaves the stand thicker. The gauge control corrects after the exit measurement, and the error is already in the strip. At the head and the tail, and at a wedge in temperature, the variation is largest.

What it costs. Thickness outside the tolerance, trimming and downgrading of the coil, and a safety margin on the target thickness, which uses more steel than necessary.

Why the usual fix fails. Feedback control reacts after the error. A thicker target is safe and costs material. A table of settings per grade does not know the actual temperature of this strip.

How a fixed-price remote diagnosis checks it. From rolling logs (forces, gaps, speeds, temperatures, exit thickness) I check how much of the variation can be computed before the strip is in the stand. The fixed-price remote diagnosis answers whether feedforward from an estimated strip state can tighten the thickness.

Background: Computing unmeasured states for control

Flatness actuators in hot rolling interact

What happens. Strip flatness is formed by several actuators: roll bending, shifting, cooling of the rolls, and gap tilt. Each acts on the shape and also on the others. The operator or a loop per actuator turns one and gets the effect on the other.

What it costs. Strip with waves or camber, rejects, and trouble in the later threading and coiling.

Why the usual fix fails. Loops per actuator fight each other, so the settings are made slow and soft. Experienced operators solve it by feel, which does not hand over. More actuators add more interaction.

How a fixed-price remote diagnosis checks it. From recordings of flatness, actuator moves and strip data I check how strong the interaction is and what a control that treats all actuators together could do. The fixed-price remote diagnosis answers whether multivariable control brings a flat strip with fewer rejects at your stand.

The cooling path decides the properties and is not controlled closely

What happens. After the last pass, the strip is cooled on the run-out table. How fast it cools decides the microstructure and with it strength and toughness. The water is set by a table. The strip temperature at the end is measured only at the coiler, and the path in between is not known.

What it costs. Mechanical properties that scatter from coil to coil, rework or downgrading after testing, and a margin of alloy or process that makes the steel more expensive.

Why the usual fix fails. More valves are not enough without knowing the strip temperature along the table. A table of water for each grade ignores speed and entry temperature. Waiting for the test result of the coil shows it too late.

How a fixed-price remote diagnosis checks it. From logs of entry temperature, speed, water valves and the coiling temperature I check how well the cooling path can be computed and set. The fixed-price remote diagnosis answers whether a controlled cooling path narrows the scatter of properties.

Background: Computing unmeasured states for control

Strip tension swings at speed changes in cold rolling

What happens. In cold rolling, the strip is pulled between the stands and the reels. At acceleration, at a weld or at a change of speed, the tension swings. The tension is measured at few places, and the loops that hold it were set once.

What it costs. Strip breaks with a long stop, thickness variation from tension swings and scrap at the start and end of each coil.

Why the usual fix fails. Slower acceleration reduces the swings and the output. Tuning the loops harder risks oscillation. A tension meter at every stand is costly and does not give the full picture.

How a fixed-price remote diagnosis checks it. From drive and tension recordings I check how well the tension between the stands can be estimated and how a control that uses the estimate could handle the speed changes. The fixed-price remote diagnosis answers whether the swings and breaks can be cut without new hardware.

Background: Computing unmeasured states for control

The composition of the scrap mix varies from heat to heat

What happens. The melt shop charges scrap of many sources. The yard sorts it by grade, and still each basket has a different content of residual elements. Copper cannot be taken out of the melt, so an unlucky basket spoils the heat.

What it costs. Heats downgraded or diluted, and a cautious charge that uses less of the cheap scrap than the steel could take.

Why the usual fix fails. Sorting the scrap more finely costs labour and yard space. Testing every load is slow. The safe recipe is always the expensive one.

How a fixed-price remote diagnosis checks it. From scrap delivery notes, charge records and analyses of the heats I estimate the content of each scrap type and check what a charge optimisation could use of the cheaper grades. The fixed-price remote diagnosis answers how reliably the composition of a basket can be estimated and what the optimisation would bring.

Power peaks during melting drive the grid charge

What happens. Furnaces draw a lot of power in short time. Peak power at the grid connection sets the capacity charge for the month or the year. Two furnaces that happen to be in the high-power phase together make the peak. Today the operators try to avoid this by feel and a telephone call.

What it costs. Capacity charges for peaks that were avoidable, and less room for cheap hours because melting cannot be freely moved.

Why the usual fix fails. A fixed order of heats ignores the delays of the day. A hard limit on power makes heats longer. A battery is expensive against the size of the peak.

How a fixed-price remote diagnosis checks it. From power logs, the heat schedule and the price and charge terms I check how much the peaks could be reduced by planning the melting against the price and the peak. The fixed-price remote diagnosis answers whether a melt plan with this target fits your logistics.

Background: Economic predictive control

Hydrogen direct reduction is a new process without operating experience

What happens. Plants for steel from hydrogen-reduced iron are built for the first time at industrial size. Nobody has years of experience of how the shaft, the gas loops and the heaters behave together. The operating team learns the process while the plant starts.

What it costs. A long and costly start-up, plant time that is not productive and a quality of the iron that scatters during the learning phase.

Why the usual fix fails. Copying the practice of gas-based plants works in part and misses the new effects of hydrogen. A pilot plant is smaller than the real one. The supplier’s guarantee covers a design point, not the whole operation.

How a fixed-price remote diagnosis checks it. From the process design, balance data and the first operating data I check how a model of the plant can describe the reduction, gas and heat, and what a control concept would need. The fixed-price remote diagnosis answers where a model can reduce the start-up risk before the plant runs at full load.

Background: Predictive control of process plants

Ladle and slab logistics are planned by hand

What happens. Ladles, cranes and slabs move through the mill in a tight rhythm. The dispatcher plans the order of cranes and ladles by experience. A delay in the furnace or at the caster moves everything, and the plan is redone by phone.

What it costs. Heat losses when steel waits, ladles at the wrong place, crane time and delays that spread to the caster or the rolling mill.

Why the usual fix fails. A fixed schedule breaks at the first disturbance. More cranes or ladles are capital for a problem of planning. A scheduling tool built for another mill does not know the layout and the rules of this one.

How a fixed-price remote diagnosis checks it. From the logs of ladle movements, crane times and heat schedule I check where the waiting arises and how much a scheduling optimisation could shorten it. The fixed-price remote diagnosis answers whether a plan that is recalculated during the shift is feasible.

Background: Optimal paths and setpoint plans

The site gas balance is run by hand and gas is flared

What happens. On an integrated site, blast-furnace and coke-oven gas are used in boilers, heaters and the power plant. Supply and demand do not match every minute. The gas dispatcher balances with holders and flares by experience, and when a consumer drops out, gas goes to the flare.

What it costs. Energy that is burnt without use, purchase of natural gas to cover shortfalls and emissions that could be avoided.

Why the usual fix fails. Bigger holders are capital for a planning problem. A fixed priority of consumers does not follow the day. Dispatching by experience does not hand over and does not see the next hours.

How a fixed-price remote diagnosis checks it. From gas flows, holder levels and consumer plans I check how well the next hours can be forecast and what an optimised distribution would save. The fixed-price remote diagnosis answers whether a gas network optimisation is worthwhile on your site.

Background: Economic predictive control

Mechanical properties are known only after testing

What happens. Strength and toughness of a coil or a bar are measured on a sample, hours or days after production. By then the next coils have been made with the same settings. If a property is off, nobody knows from when.

What it costs. Rework or downgrading of material that has been produced, testing capacity and a safety margin in the alloy or process.

Why the usual fix fails. Testing more samples gives more data and the same delay. A rule on the chemistry alone misses the effect of the process. A learned model on a few thousand tests can fail on a new grade.

How a fixed-price remote diagnosis checks it. From process data and test results of past coils I check how well the properties can be estimated during production and which process variables matter. The fixed-price remote diagnosis answers whether estimated properties can warn within minutes and be trusted.

Background: Computing unmeasured states for control

Drive control for the power supply of arc furnaces is a rare skill

What happens. The power supply of an arc furnace has transformers, converters and fast control of the electrode current. It couples to the grid, which sets limits for flicker and harmonics. Engineers who know both the drives and the furnace process are few, and projects wait for them.

What it costs. Delays in furnace projects, commissioning that depends on a single person and trouble with the grid operator when the limits are violated.

Why the usual fix fails. A converter supplier delivers its standard control, which does not know the furnace. A furnace supplier knows the process and not the converter. Hiring is slow.

How a fixed-price remote diagnosis checks it. From drawings, the converter data and the recordings of electrode current and grid values I check what limits the control and which part of the control design can be handed over as a task with a model and documentation. The fixed-price remote diagnosis shows what can be delivered remotely.

Background: Predictive control of drives

Online measurement and control in steelmaking need models next to data

What happens. Steelmakers collect a lot of data and want computed measurements and control based on physical models, with learning methods added where physics does not reach. Teams have data scientists or process engineers, rarely both with control knowledge. Projects stop at a good-looking model that never runs online.

What it costs. Projects that take years to reach the plant, models that fail at the next grade and skilled people who are tied up in prototypes.

Why the usual fix fails. Data science without process knowledge finds correlations that break. A pure physical model without data does not match the plant. Handing the topic to a software house leaves the plant questions unanswered.

How a fixed-price remote diagnosis checks it. From a typical process dataset and your question I check which part is physics, which part data, and how a computed measurement could be built to run online and stay valid. The fixed-price remote diagnosis answers what a first working version would need.

Background: Computing unmeasured states for control

Efficiency audits list measures that nobody has time to implement

What happens. In Austria, large companies must either introduce an energy management system or carry out an energy audit at regular intervals under the efficiency law (Austrian Economic Chamber). The audit delivers a list of measures. The plant people are busy with production, and the list stays on the shelf.

What it costs. Audit costs without savings, a duty that is formally met and a plant that stays as it was. The rules for the next years add further effort.

Why the usual fix fails. An auditor assesses on estimates, not on the operating data of the process. A consultant’s report does not change settings. Hiring an energy manager is not possible for each site.

How a fixed-price remote diagnosis checks it. From process data, energy meters and the audit list I check which measures can be computed and which can be run by control and optimisation at the plant. The fixed-price remote diagnosis answers what each measure is worth in your data and which can be implemented with the least risk.

New arc furnaces in running steelworks mean a variable charge and large power

What happens. Steelmakers replace blast furnaces by electric arc furnaces inside running works. The new furnaces take scrap, hot metal and direct-reduced iron in varying shares. The electrical demand is large, the charge mix changes by the day and the existing site has to adapt.

What it costs. High risk during the start-up, power cost and grid connection that decide the margin from the first heat, and a steel quality that depends on the charge.

Why the usual fix fails. Rules from other plants fit another mix. A design based on a design charge is good for one case. Trial and error in the first year is costly when the furnace is the heart of the mill.

How a fixed-price remote diagnosis checks it. From planned charge mixes, power data and furnace design I check how the energy use and quality depend on the mix and what a computed optimisation of charge and power would decide. The fixed-price remote diagnosis answers where a model can help before the furnace starts.

Background: Economic predictive control

Product rules will ask for a carbon footprint per product

What happens. The European ecodesign rules are being set product group by group, and iron and steel come early (European Commission). The requirements may cover the carbon footprint, recycled content and a digital product passport. A steel producer must state values per product, not an average for the mill.

What it costs. The effort of collecting data, the risk of a value that is too high against competitors and a lack of proof for claims about recycled content.

Why the usual fix fails. A yearly average of the mill is easy to compute and not product-specific. A life-cycle study per product is expensive. Estimates from default tables give the worst case, which hurts commercially.

How a fixed-price remote diagnosis checks it. From energy and material flows of the process I check how a footprint per heat or coil can be computed from existing data and how stable it is. The fixed-price remote diagnosis answers whether the footprint for each product can be derived without additional meters.

Background: Computing unmeasured states for control

Grid charges are expected to rise and reward flat load profiles

What happens. The state subsidy that has lowered the grid charges for transmission is planned to be reduced for the coming years (DIHK). For a steelworks, which draws power in large blocks, grid charges are a large part of the cost, and with them the question of how flat the load profile can be kept.

What it costs. A power bill that rises, and a cost that depends on a few peak quarter-hours as well as on the total energy.

Why the usual fix fails. Negotiating a better contract changes the price, not the profile. A load limit by hand stops production at the wrong times. A battery is large against a furnace.

How a fixed-price remote diagnosis checks it. From load profiles of the main consumers, the production plan and the charge terms I check which loads can be moved or limited, and by how much the bill could fall. The fixed-price remote diagnosis answers whether a peak and load-profile optimisation fits the way you produce.

Background: Economic predictive control