198
7 Summary and Outlook
In this work failures have not been considered, as the timing of failures in the
model will always be different from their actual future occurrence in production.
Thus, a small deviation in production quantities between the real production and
the simulation model has been noticed. As the consideration of failure probabilities would have a negative effect on the ability to detect consumption peaks, they
might stay undetected due to a machine failure occurrence in the model but not
in the real production, machine failures are not part of the simulation model.
One prerequisite for the proposed method is the existence of measured energy
consumption data at adequate levels of resolution. While energy monitoring systems are gradually being recognized and installed as efficiency tools in many
companies, such an accurate tracking of the energy consumption on machine
level in a very detailed resolution is rather the exception. To grant flexibility
in case load data is not available, the simulation approach has been tested with
mean values instead of energy load profiles. For each machine state a mean value
has been added as a parameter, which is taken as the consumption value for the
entire duration of the state. While this did not have any influence on the detection
and avoidance of unproductive states and only a small deviation regarding the
total energy consumption of the line through the use of mean values was noticed, the existing peak loads in the consumption profiles disappeared completely.
The model accuracy should therefore be chosen depending on the objective of the
investigation. A combination of the use of energy data tables and mean values
is also conceivable if not all machines have yet been connected to the energy
data acquisition in a production but should already be taken into account. There
are only limited advantages of the use of real energy profiles, when peak reduction is not an issue in production. As the use of the exact energetic load profiles
requires intensive data work, it might be advantageous to use the mean value version which reduces modeling efforts and the experiment run in the simulation and
optimization experiments drastically.
Both, the fictional and the real example have shown that a lexicographic optimization provides better results than the optimization of both objectives in one
single experiment. Since the reduction of the total energy consumption is the primary goal, it makes sense to define two consecutive optimization experiments.
This procedure also has the advantage that not both objectives need to be pursued in every project, but only the total energy consumption or only the peak
loads can be optimized. An optimization of both objectives in a single optimization experiment would technically be possible by using a cost model. In practice,
however, this approach failed because the energy companies were not willing to
reveal consumption-based grid fees. Thus, it was impossible to assess the peak
consumption reductions monetarily.
7 Summary and Outlook
In this work failures have not been considered, as the timing of failures in the
model will always be different from their actual future occurrence in production.
Thus, a small deviation in production quantities between the real production and
the simulation model has been noticed. As the consideration of failure probabilities would have a negative effect on the ability to detect consumption peaks, they
might stay undetected due to a machine failure occurrence in the model but not
in the real production, machine failures are not part of the simulation model.
One prerequisite for the proposed method is the existence of measured energy
consumption data at adequate levels of resolution. While energy monitoring systems are gradually being recognized and installed as efficiency tools in many
companies, such an accurate tracking of the energy consumption on machine
level in a very detailed resolution is rather the exception. To grant flexibility
in case load data is not available, the simulation approach has been tested with
mean values instead of energy load profiles. For each machine state a mean value
has been added as a parameter, which is taken as the consumption value for the
entire duration of the state. While this did not have any influence on the detection
and avoidance of unproductive states and only a small deviation regarding the
total energy consumption of the line through the use of mean values was noticed, the existing peak loads in the consumption profiles disappeared completely.
The model accuracy should therefore be chosen depending on the objective of the
investigation. A combination of the use of energy data tables and mean values
is also conceivable if not all machines have yet been connected to the energy
data acquisition in a production but should already be taken into account. There
are only limited advantages of the use of real energy profiles, when peak reduction is not an issue in production. As the use of the exact energetic load profiles
requires intensive data work, it might be advantageous to use the mean value version which reduces modeling efforts and the experiment run in the simulation and
optimization experiments drastically.
Both, the fictional and the real example have shown that a lexicographic optimization provides better results than the optimization of both objectives in one
single experiment. Since the reduction of the total energy consumption is the primary goal, it makes sense to define two consecutive optimization experiments.
This procedure also has the advantage that not both objectives need to be pursued in every project, but only the total energy consumption or only the peak
loads can be optimized. An optimization of both objectives in a single optimization experiment would technically be possible by using a cost model. In practice,
however, this approach failed because the energy companies were not willing to
reveal consumption-based grid fees. Thus, it was impossible to assess the peak
consumption reductions monetarily.