· · · Institutional Repository

The objective of this research is to limit the risks of fully automated operation of drinking water treatment plants and to improve their operation by using an integrated system of process models and emulated process automation software. This thesis contains the design of such an integrated system. The use of the system is investigated in the three identified applications, i) optimization of process control, ii) training of operation supervisors and iii) virtual commissioning of process automation software.

The flow through a unit of a drinking water treatment plant is one of the most important parameters in terms of a unit's effectiveness. In the present paper, a new EPAnet library is presented with the typical hydraulic elements for drinking water treatment processes well abstraction, rapid sand filtration and cascade and tower aeration. Using this treatment step library, a hydraulic model was set up, calibrated and validated for the drinking water treatment plant Harderbroek. With the actual valve position and pump speeds, the flows were calculated through the several treatment steps. A case shows the use of the model to calculate the new setpoints for the current frequency converters of the effluent pumps during a filter backwash.

This research deals with the contribution of process simulation models to the factory acceptance test (FAT) of process automation (PA) software of drinking water treatment plants. Two test teams tested the same piece of modified PA-software. One team used an advanced virtual commissioning (AVC) system existing of PA-emulation and integrated process simulation models, the other team used the same PA-emulation but basic parameter relations instead of the process simulation models, the VC-system. Each test team found one (different) error of the thirteen errors put into the software prior to the experiment; the majority of the errors was found prior to the functional test. The team using the AVC-system found three errors, the team using the VC-system found four, but the AVC-team judged 1% of the test items ‘not possible’, the VC-team 17%. It was concluded that the hypothesis that with AVC more errors could be found than with VC could not be accepted. So, for the FAT of PAsoftware of drinking water treatment plants, the addition of basic parameter relations to PA-emulation satisfied. Not the exact process behavior helped to find errors, but the passing of process thresholds.