Experience with Digital Process Optimization of Zinc Roasting …
381
pure data-based approach would recommend a feed increase to bring the temperature
backup. This conclusion is based on the positive correlation of feed flow and bed
temperature usually observed in roasting plant data but obviously wrong, potentially
even hazardous, because the complex reality of roasting phenomena is not sufficiently
captured by the data. Methods for decision-making based on process know-how take
an entirely different approach by analyzing causes and effects based on the laws
of physics and thermodynamics. This way, much more effects and complexity are
captured than with the data-based approach. In the aforementioned example, process
know-how would dictate that something is wrong with the oxygen/feed ratio and that
further feed increase will not lead to a restored bed temperature.
The same fundamental issue with data-based approaches also lead to difficult,
sometimes even impossible traceability of results. Origins of results and made decisions are often not clear and understandable because the approach abstracts interpretable data and thus greatly reduces interpretability. In contrast, approaches based
on process know-how draw conclusions with clear relationships between causes and
effects, which maintain interpretability of the results. This way, digital systems can
also give situation-dependent operational advice and explanations to the operator
to avoid decay of operator know-how for example. Data-based methods struggle to
achieve this.
Data-based methods rely on previously acquired data about the process to learn the
inherent behavior of reality. In practice, although large amounts of data are more and
more available, these data are often shallow with respect to covered operating ranges,
different feed materials processed, operating philosophy, plant modifications, and so
on. Each of these aspects has a strong influence on plant behavior. If the data does
not contain meaningful ranges and variations for all of them, data-based methods
struggle to achieve a necessary degree of predictive capability. Approaches based on
know-how can overcome this difficulty because the knowledge can be extrapolated
and applied to new situations.
Finally, extensibility of digital systems is very important. Operational priorities,
plant setups, plant automation, feed materials but also process insight can change over
time and digital solutions need to consider these variations in order to be successful.
Introducing this additional information or knowledge into an existing know-howbased system is usually easily done. If a pure data-based approach is followed,
inclusion of the new information into the existing set of data-based information can
be difficult to achieve. For that reason, it is also difficult to include new customer
requirements into existing solutions.
All in all, approaches including process know-how are favorable over pure databased approaches because they feature many benefits. Despite the drawbacks of
purely data-based methods, they can help close gaps between know-how and reality
and are hence a good complementation of know-how-based approaches.
381
pure data-based approach would recommend a feed increase to bring the temperature
backup. This conclusion is based on the positive correlation of feed flow and bed
temperature usually observed in roasting plant data but obviously wrong, potentially
even hazardous, because the complex reality of roasting phenomena is not sufficiently
captured by the data. Methods for decision-making based on process know-how take
an entirely different approach by analyzing causes and effects based on the laws
of physics and thermodynamics. This way, much more effects and complexity are
captured than with the data-based approach. In the aforementioned example, process
know-how would dictate that something is wrong with the oxygen/feed ratio and that
further feed increase will not lead to a restored bed temperature.
The same fundamental issue with data-based approaches also lead to difficult,
sometimes even impossible traceability of results. Origins of results and made decisions are often not clear and understandable because the approach abstracts interpretable data and thus greatly reduces interpretability. In contrast, approaches based
on process know-how draw conclusions with clear relationships between causes and
effects, which maintain interpretability of the results. This way, digital systems can
also give situation-dependent operational advice and explanations to the operator
to avoid decay of operator know-how for example. Data-based methods struggle to
achieve this.
Data-based methods rely on previously acquired data about the process to learn the
inherent behavior of reality. In practice, although large amounts of data are more and
more available, these data are often shallow with respect to covered operating ranges,
different feed materials processed, operating philosophy, plant modifications, and so
on. Each of these aspects has a strong influence on plant behavior. If the data does
not contain meaningful ranges and variations for all of them, data-based methods
struggle to achieve a necessary degree of predictive capability. Approaches based on
know-how can overcome this difficulty because the knowledge can be extrapolated
and applied to new situations.
Finally, extensibility of digital systems is very important. Operational priorities,
plant setups, plant automation, feed materials but also process insight can change over
time and digital solutions need to consider these variations in order to be successful.
Introducing this additional information or knowledge into an existing know-howbased system is usually easily done. If a pure data-based approach is followed,
inclusion of the new information into the existing set of data-based information can
be difficult to achieve. For that reason, it is also difficult to include new customer
requirements into existing solutions.
All in all, approaches including process know-how are favorable over pure databased approaches because they feature many benefits. Despite the drawbacks of
purely data-based methods, they can help close gaps between know-how and reality
and are hence a good complementation of know-how-based approaches.