Maximize Efficiency and Reliably Adapt Metals Production Schedules to Energy Availability (Part 2)
06 Jan 2023 - Energy, Production
Government regulations and sustainability measures impose cost pressure on steelmakers, but also drive the trend towards more efficient energy consumption. Yet, when it comes to efficiently managing energy consumption and scheduling production, steel producers face enormous difficulties due to their complexity, requiring high level of expertise and big data. However, steelmakers can reliably determine their energy consumption values and successfully adapt their production schedules to energy availability. All it requires are proactive methods and robust solutions.
Methods of Deriving Specific Energy Consumption Values
Many industries have repetitive businesses. For such industries, most productions are done multiple times annually. The production follows a defined sequence of operations and processes on specific lines using the same process parameters for a given product. Through data mining, we can calculate the mean or, after removing possible outliers, the average and standard deviation of energy consumption by product, production step and process parameter for all relevant processes in the route.
For some steps, like batch annealing, the specific energy consumption is only process dependent. Material properties like total batch weight and contained coil thickness might influence the total cycle duration. The general energy consumption profile for an anneal cycle will still be the same and just end earlier for a given annealing recipe.
However, if the business is less repetitive or contains products with low repetition rates, getting reliable data is harder. One way is to review individual production reports and confirm the specific energy consumption manually. The other is to apply analytical methods using product and process properties of similar production processes and derive functions to predict the energy consumption for a production step.
The foundation for deriving reliable prognosis data is a solid data basis composed of on-time production report messages and synchronized energy consumption message, preferably from connected level 2 systems. Handwritten report sheets or manual entries into production reports are prone to errors when it comes to time punctuality and mapping with Energy Management System (EMS) reports.
Identifying the relevant product and process properties is a challenge. Whoever said, “Make an educated guess” first, is probably right on the spot. Most applications of Gaussian process regression assume that the parameters of the process are linear independent. Applying these methods results in a prediction model. However, it is essential to get rid of statistical outliers. To do so in an automated way, the size of the dataset for producing the same product has to be big enough to identify outliers.
Methods to identify outliers
- Mean and standard deviation
- Median and median absolute deviation
- Interquartile range
- Dixon’s Q test
- Grubb’s test
- Rosner’s test, etc.
Machine learning uses a set of methods to identify outliers and remove them from the dataset.
It might become necessary to run multiple loops to generate a model based on the given dataset, removing outliers from the dataset every time. Even then, it is important to verify the prognosis quality of the derived function with real production data. Mapping this to Annealing recipes and Electric Arc Furnace (EAF) processes typically follow a given recipe and have a predictable energy consumption profile. Figure 1 illustrates the energy consumption for heating in a recovery annealing cycle for multi-pass cold rolling.
In cold rolling mills, the energy consumption depends on input material properties and the required output material properties. Beside the reduction rate / material displacement and the coil width, the steel grade and grain shape have a direct impact on the required energy.
Typically, a closed control loop takes care of the rolling process. Here analytical methods can predict the required energy to roll a coil to a specific product. We can assume that the input material properties defined by previous production processes are known within some small tolerances. Therefore, applying the analytically derived formulas will result in an acceptable energy consumption forecast for this production step in the route.
Energy consumption forecasting in galvanizing processes depends on the process on one hand – galvanize or galvanneal – and on the other hand, the input material properties needed to achieve a defined input temperature and process speed.
Depending on the time that it takes to produce a material on a specific line and the type of energy consumption forecasting model, the specific energy demand can be expressed as an average value by product, either stored or calculated by a formula, or time dependent via a detailed consumption curve for a given process start.
Complex Production Scheduling in the Steel Plant
The precondition to forecast energy consumption is that date and time can be predicted when a production step will be executed on a selected line. Unfortunately, production scheduling in the steel industry is very hard and indeed recognized as one of the most difficult and complex industrial scheduling problems.
To achieve a good plant productivity, it is mandatory to plan production and to streamline production flows, so that production orders are completed on time in full, while the equipment utilization is maximized and average lead times are minimized. The production planning models do not consider all constraints of the individual production lines, but give constraints to detailed scheduling.
When the operator releases the sequences, they are put in a continuous time model; production rates are applied to determine the job durations; a resource calendar defines available operation time intervals for each line per day, week and year. By linking the newly released schedules to previously released ones, the start time and duration of each job is calculated.
Typically, line schedules are created and released to production for a time span ranging from 8 hours up to 5 days. Detailed schedules typically exist for all energy consumption relevant production lines.
Remark: Reactive scheduling might change the line schedules later due to production events like quality issues with individual materials or production problems at the line. Here it will become important later to respect the energy constraints set by the original line schedule.
Adapting Production Schedules to Energy Availability
The aim of Industrial Demand-Side Management is to modify the overall consumption time profile. The motivation for this is to achieve savings in electricity charges in accordance with the contractual power and grid connection parameters. Adapting production schedules to the forecasted available energy supplied by the grid, and provided by internal sources, is doing exactly this by modifying the consumption profile to the availability constraint.
There are different levels to achieve this:
- Visualization of forecasted availability and consumption, showing mismatches and violations, enables the user to act and resolve problems. Figure 2 shows the energy demand forecast of a planning and scheduling scenario with maximum energy availability constraints for a set of time buckets.
- The colorization by consuming line makes it obvious that consumption on “Tandem” and “ColdRoll” have high levels in times of low energy availability. Instead of stopping one line completely, the problem can be resolved by changing the sequence of the planned line schedules. Figure 3 shows the result after resequencing of the schedules.
Switching the sequence of schedules at the two rolling mills is one form of “Load Shifting”. Although energy is not saved, energy costs can be significantly reduced. If “load shifting” cannot resolve the problem of energy consumption exceeding contractual limits in a time bucket, “Peak Clipping” is applied to avoid a significant rise in energy costs. Shutting down complete production lines or a drastic reduction of rolling speed are examples for this hard intervention in the production processes with the potential to cause quality issues.
Identifying potential situations where “peak clipping” measures could become necessary, enables production planning and control to take preventive measures. Changing the sequence of line schedules works well if there is a relation between applied schedule templates and energy consumption within the schedule or in specific schedule zones. Even then, the schedule durations and material availability set limits to load shifting.
What Other Approach Could Be Applied?
A second approach is to use the energy constraints in the creation of the line schedules. The energy constraint is time dependent and has to be applied as a time series or bucketized curve. In many cases, line schedule creation is modeled as a sequencing problem. The optimization methods exchange, insert or remove individuals or groups of jobs in a sequence. After each operation, the total costs of the sequence are calculated to rate the quality of the newly created sequence. This has low computational costs when only a limited number of jobs example left and right neighbor contribute to the cost for inserting or removing a job.
If this is not the case, the execution time to explore the solution space and get good results rises quickly with the number of jobs in the schedule. Applying energy constraints is transforming the sequencing problem into a scheduling problem to respect the time dimension. Deviation from the “wanted” energy level determines a new cost component in the overall cost function. With each operation – insert, remove, exchange – all jobs in the sequence after the place of operation can be shifted from one time bucket to another.
Therefore, a recalculation for all jobs after the inserted, replaced or removed one(s) is required. The position in continuous time and the contribution to time buckets can change.
Therefore, the energy cost component per time bucket has to be recalculated for all buckets, starting at the bucket in which the operation occurred. Whether the additional cost component is over-constraining the model, resulting in poor scheduling results, has to be analyzed for each application and production line.
The same has to be said for the impact on the duration it takes to solve the optimization problem for typical schedule sizes.
This second approach provides another challenge: If multiple production lines have product/process dependent energy consumption, how can the target/max-constraint energy curve to the individual line specific scheduling models be distributed?
If there is only one main contributor, the target/max-constraint curve is the energy availability minus the forecasted demand for the whole plant with exception of the production line in scope. For the multiline problem, a similar approach is applicable. The target/max-constraint curve is the energy availability minus the forecasted demand for the whole plant with exception of production lines with product/process dependent energy consumption. Here we have to add the energy forecast of line schedules from these lines again, if they have been released to production execution and cannot be modified by production scheduling any more. The fixed horizon may differ for the production lines depending on the duration of the released schedules.
The resulting target/max-constraint curve can then be distributed to the individual scheduling modules. It is important to distribute the target/max-constraint curve only between the lines with remaining production capacity in a time interval. The distribution factor depends on the
- Remaining capacity per interval,
- The mean energy consumption and
- The product/process dependent energy variation.
This approach supports multi-user scenarios, but does not compensate deviations from the target values. Ideally, they should be distributed to the target/max constraint curves of the other lines. This could only be achieved by sequentially solving the individual optimization models and feeding back the results as new fixed forecasts. Doing so in multiple cycles with re-optimization of schedules created in the last cycle is an iterative approach to get to the best overall fit between forecasted supply and demand. However, this would require a completely automated system and probably result in unacceptable calculation time.
A Lasting Solution for Grid Stability
Outsourcing industrial demand-side management to the grid operators is not an option for the metals industry.
Key for a successful Industrial Operation
- The complexity of planning and scheduling the supply chain
- Maintaining production and process parameters to ensure final product quality
- Respecting line and stock constraints in scheduling and
- Ensuring customer satisfaction by maintaining on time in full (OTIF), etc.
To overlook these factors will affect competitiveness. Proactively adapting the production from planning to execution, following signals from grid operators, as well as from energy providers, will contribute to grid stability, reduce energy costs and contribute in the transition to a pure renewable energy supply. Forecasting models exist on the market and are being implemented.
Examples are the PSImetals Online Heat Scheduler for the liquid phase and PSImetals Order Scheduler for the solid production, which enable users to monitor and adjust production scheduling, respecting energy constraints and building energy cost efficient production plans.
The importance and monetary incentive to predict consumption and follow the prediction will grow with an increased share of volatile renewable energy in the grid. Therefore, it is important to act now and investigate flexibility potentials to adjust energy consumption to availability.
Series: Energy Consumption Forecast
- Part 1: Maximizing Efficiency: How to Align Your Metals Production With Energy Consumption Forecast
- Part 2: Maximize Efficiency and Reliably Adapt Production Schedules to Energy Availability
Discover PSImetals Release 5.24!
Find out more
Dr. Stefan Albers
Business Consultant, PSI Metals
salbers@psi.de
After 4 years as Head of Production Planning and Scheduling in a cold rolling mill, Stefan joined PSI Metals in 2007 as a Software Engineer and Project Manager. Due to his enormous experience and expertise, he became a Consultant and Solution Architect. Stefan has exceled in this role, becoming a global and prominent voice in planning and scheduling as well as overall customer business process where he has contributed to various projects worldwide.
Follow Us on
