A Stable Plant Is Not Necessarily An Optimal Plant
Laboratory studies and pilot plants provide the first operating conditions for a new industrial process. They are essential, but they remain approximations.
The real process only becomes visible at full scale. Raw-material quality changes. Equipment wears. Heat transfer deteriorates. Catalysts age. Ambient conditions fluctuate. Measurement errors and disturbances become significant. At the same time, production must continue, customer specifications must be met and unnecessary risks must be avoided.
Once a plant is running reliably, operating conditions are therefore often kept within narrow ranges. This stabilizes production, but it also limits learning. Historical data may contain thousands of observations while providing surprisingly little information about what would happen under better operating conditions.
The central idea of Evolutionary Operation, or EVOP, is simple:
An industrial process should generate not only product, but also information about how to improve that product and the process producing it.
EVOP introduces small, planned variations into normal operation. The changes are deliberately kept within acceptable production limits. By repeating them systematically, engineers can distinguish real process effects from industrial noise. Operating conditions are moved only when the accumulated evidence supports the decision.
The following cases show how this philosophy has created measurable value in minerals processing, polymer extrusion, biotechnology and real-time equipment control.
Case 1: Reducing Ozone Consumption In Full-Scale Kaolin Processing
At a BASF kaolin plant, the objective was to reduce the operating cost of a continuous ozonation and bleaching process with a production capacity of approximately 40 tonnes per hour.
The final product brightness depended on the interaction between ozone dosage and compressed-air dosage. However, testing this relationship at full scale was difficult. The large production rate meant that any unsuitable operating condition could generate a considerable amount of lower-quality material.
To control this risk, the experimental work was divided into four blocks conducted during separate periods. This reduced the plant’s exposure to off-specification production and allowed the analysis to account for changes in feed quality, upstream performance, weather, production rate and other uncontrolled conditions.
The work identified a practical operating region in which sufficient compressed air improved contact between ozone and the kaolin slurry. Too little air resulted in poor contact, while excessive air diluted the ozone and reduced its effectiveness.
A verification trial showed that the ozone dosage could be reduced from 1.0 to 0.8 pounds per tonne while maintaining the required final brightness. This represented a 20% reduction in ozone use and reported annual savings of approximately $960,000 at the prevailing production rate.
Strictly speaking, this case was conducted as a blocked response-surface Design of Experiments study rather than a classical EVOP program. Nevertheless, it illustrates precisely the industrial problem for which EVOP was developed: the plant must learn, but the learning process cannot be allowed to disrupt production.
The Lesson
Full-scale experimentation becomes acceptable when changes are controlled, exposure is limited and the economic objective is clearly connected to product-quality constraints.
Case 2: Revealing Hidden Capacity In Dow Film Extrusion
A Dow Chemical plant operated several film extruders, each containing multiple heating zones in addition to the heat generated by screw friction.
The number of possible combinations of heating-zone temperatures and screw speeds was enormous. In the absence of a structured optimization method, individual operators developed their own preferred settings. A machine could appear to run smoothly, yet its temperature profile could change significantly when a different operator took over.
EVOP was introduced on one extruder to identify which heating zones had the greatest influence and to move the process systematically toward better operating conditions.
Within a relatively short time, the selected machine achieved an output approximately 20% higher than comparable machines. As the study continued, output increased further. The process was explored until undesirable product characteristics began to appear, defining the practical boundary of the operating region. At that point, output had increased by approximately 37%.
A related application examined the use of recycled material. A conventional one-factor-at-a-time increase in the recycle ratio had previously caused a substantial deterioration in product quality. EVOP instead investigated several factors together.
By coordinating the process settings, the amount of recycled material could ultimately be doubled while product quality improved. The historical source estimated the resulting savings at approximately $4,000 per month.
The Lesson
The best operating condition is rarely determined by one factor in isolation. EVOP makes interactions visible and replaces operator-specific recipes with a reproducible learning process.
It also shows that optimization is not simply a search for the highest possible throughput. The goal is to find the best combination of production rate, quality, material use and process stability.
Case 3: Improving Enzyme-Fermentation Yield At Novozymes
Fermentation processes are particularly challenging to optimize.
Biological variability, raw-material differences and long batch durations make it difficult to distinguish the effect of a deliberate process change from normal variation. The conditions identified in laboratory development may provide a useful starting point, but the full-scale organism, equipment and operating environment create their own response.
At Novozymes, EVOP was applied to the full-scale fermentation of an industrial enzyme. The program investigated three operating factors:
- pH
- temperature
- nutrient dosage
Process yield was used as the principal response.
The experimental arrangement was divided into two blocks. This meant that stable conditions were required for only four or five consecutive batches at a time, rather than for the entire duration of the program. The blocking strategy reduced the influence of longer-term process variation and made the investigation more compatible with normal production.
The reported outcome was a 45% improvement in process yield. The case demonstrated that EVOP could be used as a practical production method in industrial biotechnology—not merely as a statistical analysis performed after production.
The Lesson
EVOP is not limited to continuous chemical plants.
In a batch process, the learning cycle is measured in batches rather than minutes or hours. The underlying logic remains the same: introduce small variants, repeat them, accumulate evidence and move the operating conditions only when the improvement is sufficiently reliable.
The case also demonstrates the value of blocking. Industrial processes rarely remain completely unchanged throughout a long optimization program. A suitable EVOP structure allows learning to continue despite these longer-term shifts.
Case 4: Real-Time Evop Control Of A Cone Crusher
The fourth case moves EVOP from a manually managed improvement program toward real-time process optimization.
Cone-crusher performance changes continuously as material characteristics fluctuate and the crusher liners wear. A fixed operating speed may perform well at one point in the liner lifetime but poorly at another. The optimum is therefore not a single setting that can be identified once and then used permanently.
Researchers developed a real-time control algorithm in which material-flow feedback was combined with EVOP. The system varied two parameters:
- the initial eccentric speed
- the rate at which that speed was changed
The algorithm learned across several EVOP phases while the closed-side setting, liner condition and incoming material continued to evolve.
The reported two-variable control strategy achieved 6.9% higher performance than the best fixed-speed reference. Differences of approximately 20–30% were observed between the best and worst runs within EVOP phases, while the performance range over the lifetime of the liners exceeded 100 tonnes per hour.
These results show that the economically relevant optimum was moving materially during normal operation.
The Lesson
EVOP does not have to remain a paper-based system in which operators manually record results and engineers periodically update an information board.
The underlying method can be integrated into a control algorithm that continues to test, evaluate and adapt as the process changes. This case provides a direct bridge between traditional EVOP and modern autonomous process-improvement systems.
What These Cases Have In Common
The four applications involve very different processes. Kaolin bleaching, film extrusion, microbial fermentation and crushing have different equipment, timescales and constraints.
Yet the same five principles explain their success.
1. THE RELEVANT OPTIMUM EXISTED AT FULL SCALE
Laboratory work, pilot studies and standard recipes provided useful starting conditions. They did not reveal the complete industrial optimum.
The important effects became visible only in the real production environment.
2. THE OPTIMUM WAS NOT PERMANENT
Feed quality, biological activity, environmental conditions, wear and process history changed the response over time.
Optimization could therefore not be treated as a one-time project.
3. FACTOR INTERACTIONS MATTERED
In the kaolin process, compressed air changed the effectiveness of ozone. In extrusion, recycled-material use depended on other operating conditions. In fermentation, pH, temperature and nutrient dosage had to be considered together.
One-factor-at-a-time adjustments would have provided only a partial picture.
4. REPETITION CREATED CONFIDENCE
Individual industrial observations are noisy. A single apparently successful run may be caused by an uncontrolled disturbance rather than by the deliberate process change.
EVOP repeats small variants until a consistent effect can be distinguished from normal variation.
5. PRODUCTION CONSTRAINTS WERE PART OF THE METHOD
The cases did not treat production as a laboratory. Changes were limited by product quality, equipment restrictions and operational acceptance.
Close cooperation with the people responsible for the process was essential. Collaboration is useful in conventional experimentation; in EVOP, collaboration is critical.
What Prevented Traditional Evop From Scaling
The method itself has never been the main limitation.
Traditional EVOP programs required operators to record conditions and responses, engineers to perform calculations, teams to update information boards and supervisors to decide when sufficient evidence had been collected to begin a new phase.
This approach could work very well, but it demanded discipline over long periods. Programs could slow down when production priorities changed, when additional laboratory analysis became burdensome or when the responsible personnel moved to other tasks.
The industrial environment has changed substantially since the first EVOP applications.
Modern plants have process historians, connected instrumentation, automated quality checks, edge computing and digital workflows. Data can be collected and evaluated continuously. Safety limits can be encoded explicitly. Process changes can be documented, traced and reversed.
These capabilities do not replace the EVOP philosophy. They make it practical to apply that philosophy continuously.
The Next Chapter: Pia
The historical cases demonstrate that small, structured changes can uncover major improvements in capacity, yield, material efficiency and operating cost.
They also demonstrate that the optimum of a real plant is often a moving target.
PIA—the Process Improvement Agent—builds on this industrial experience. Its purpose is to transform EVOP from a periodic, manually coordinated improvement program into an always-on and guardrailed optimization workflow.
PIA does not need to make large or disruptive changes. It works through controlled variation around the current operating point. It evaluates whether observed changes are repeatable, whether they respect quality and safety constraints and whether they create measurable economic value.
The objective is straightforward:
Keep the plant on specification. Keep the process learning.
A first application does not require an entire site or an extensive digital-transformation program. A suitable pilot can begin with:
- one production process
- one economically meaningful response
- two or three adjustable operating factors
- clearly defined safety and quality limits
The historical record shows that EVOP can work.
The opportunity now is to make it continuous, traceable and scalable.
References
1. Lazić, Ž. R. (2022). Full Scale Plant Optimization in Chemical Engineering: A Practical Guide. Wiley-VCH.
2. Box, G. E. P., and Draper, N. R. (1969). Evolutionary Operation: A Statistical Method for Process Improvement. Wiley.
3. Kvist, T., and Thyregod, P. (2005). “Using Evolutionary Operation to Improve Yield in Biotechnological Processes.” Quality and Reliability Engineering International, 21(5), 457–463.
4. Hulthén, E., and Evertsson, C. M. (2011). “Real-time Algorithm for Cone Crusher Control with Two Variables.” Minerals Engineering, 24(9), 987–994.
Reporting note: The performance and cost figures are reproduced as reported in the respective historical sources. Monetary values are nominal and have not been adjusted for inflation. The BASF kaolin example was a blocked response-surface DOE study rather than a classical EVOP application; it is included because it demonstrates the full-scale optimization constraints that EVOP is designed to address.