Software sensors are a real alternative to true sensors
Introduction
This investigation was performed at the Ejby Mølle WWTP in Odense, Denmark. This is a three-stage plant where preprecipitation precedes a Biodenipho® extended aeration unit (40,000 m3). The last treatment step is sand filtration. The load on the plant is 220,000 PE.
In the Biodenipho® process nitrification and denitrification occur in the same tank but in alternating phases. Aeration stops and a denitrification phase begins when the NH4 concentration drops below 0.6 mgN L−1. Aeration is restarted again when the NH4 concentration reaches 1.6 mgN L−1 or when the redox potential drops below −120 mV (all redox potentials in this paper are relative to a Ag|AgCl reference electrode) (Cecil, 2003). The bend in the redox curve, which indicates that the nitrate concentration is zero (Koch and Oldham, 1985) occurs at potentials greater than −120 mV at the Ejby Mølle plant. As shown in Fig. 1, there are two pairs of aeration tanks. The flow to and from the tanks is controlled by movable weirs at each end of the tank. When the aeration stops in one tank of a pair and the aeration starts in the other tank the weirs open and close redirecting the flow to the tank that is not aerated.
There is an opening between the two tanks in a pair such that there can be influent to one tank and effluent from the other. However, at this WWTP the flow is normally to and from the same tank. When the flow to the plant is large due to rain, there is settling in the aerations tanks. This reduces the sludge load on the secondary settling tanks and increases the hydraulic capacity. Settling occurs in one of the two tanks in each pair. In the other tank, there is influent flow and aeration. Effluent is from the tank in which there is settling.
Fig. 1 shows the positions of the meters in the aeration tanks. The meters in each tank are:
- •
1 ammonium meter (the EVITA® INSITU sensor manufactured by HachLange)
- •
2 oxygen meters (two HachLange EVITA® series sensors, one at each end of the tank)
- •
1 redox potential (ORP) meter (a Broadley–James electrode connected to a Yokogawa transmitter).
The ammonium meter in the influent to the aeration tank is also an EVITA® INSITU sensor.
The flows shown in Fig. 1 are measured by magnetic flow meters. 70% of the flow to the unit (not including the return sludge) is directly from the primary settling tank. 30% comes from two sets of trickling filters operating in series. The first set of trickling filters pre-treats primary settling tank effluent. The second set of trickling filters performs nitrification. Therefore there is a flow of nitrate to the activated sludge unit. This nitrate load is estimated by multiplying the average concentration of nitrate in the flow by the measured flow from the trickling filters.
There are 6 surface aerators in each aeration tank. The oxygen concentration is controlled by turning the surface aerators on and off. During the aerated phases the oxygen concentration is normally between 0.5 and 1.0 mg L−1.
The ammonium meter is an auto analyzer with good accuracy and precision. The meter also has a short response time compared to other auto analyzers. The meter manufacturer reported that the response time is the sum of a 5 min dead time and a characteristic rise time of 1.3 min. This agrees well with performance observed at the plant. On the other hand, the meters have an up-time of less than 95%. The primary causes of down-time are automatic calibration and rinsing, manual maintenance and instrument failure.
Many large WWTPs operating the Biodenipho® process use nitrate meters instead of ORP. Rather than install nitrate meters at the Ejby Mølle plant it was decided to compute the nitrate concentration using a software sensor based on the existing instruments. Other expected benefits of a software sensor were estimations of the nitrification and denitrification rates.
The software sensor in this investigation employs an observer based estimator (OBE) of the ammonium plus ammonia concentration (NH4), the maximum nitrification rate, the nitrate plus nitrite concentration (NO) and of the maximum denitrification rate in aeration tanks. There are a number of different types of OBEs described in the literature (Dochain and Vanrolleghem, 2001). The Kalman filter is an OBE in common use in WWTPs (Bernard et al., 2006). OBEs combine a prediction of a measured state variable and a correction of that prediction based on the difference between the measurement and the prediction. The OBE in this investigation also adjusts the rates based on this difference. Farza et al. (1997) describe this type of OBE in detail.
Section snippets
The observer based estimator
The OBE is as follows:
Results
Table 3 shows the rmsd between the concentrations measured in the grab samples and the predicted concentrations for NH4 and NO. Fig. 2 shows the concentrations of NH4 and NO on February 25 2008; a day with average values for the rmsd. For most of the samples the predicted NH4 concentration was more accurate than the concentration measured by the NH4 meter because of the response time of the NH4 meter. Note that the NH4 meter was out of operation for about 40 min at 11:30. This was due to
Discussion
The variation in rnit shown in Fig. 3 is larger than expected. Temperature changes account for less than 45% of this variation. rnit. There was also a correlation between the mixed liquor suspended solids (MLSS) and rnit. However, temperature and MLSS concentration decreased in parallel during the period shown in Fig. 3. Therefore it is difficult to separate the effects of the two variables. Together the temperature and MLSS accounted for 55% the variation of rnit. Changing concentrations of
Conclusion
At the Ejby Mølle WWTP in Odense, Denmark a software sensor predicts the NH4 and NO concentrations in real-time. The software sensor uses the signals from NH4 meters and from ORP electrodes placed in each of the plant's 4 aeration tanks. The predicted NH4 concentration is used to control the length of the nitrification phase in a Biodenipho® activated sludge unit rather than the measured NH4 concentration because the software sensor has a shorter response time and a better up-time. The software
References (6)
- et al.
Nonlinear observers for parameter estimation in bioprocesses
Chem. Eng. Sci.
(1997) - et al.
State Estimation for Wastewater Treatment Processes. Wastewater Quality Monitoring and Treatment
(2006) Controlling nitrogen removal using redox and ammonium sensors
Wat. Sci. Tech.
(2003)
Cited by (27)
Nitrogen recovery from the municipal wastewater treatment plants
2023, Resource Recovery in Municipal Waste WatersData-driven intelligent monitoring system for key variables in wastewater treatment process
2018, Chinese Journal of Chemical EngineeringCitation Excerpt :Meanwhile, most of these above soft sensors cannot be used in real WWTP [26,27]. In fact, the monitoring systems, based on data-driven soft sensors, have been widely accepted and adopted for the industrial processes [28,29]. And some researchers are trying to design a cost-effective platform for developing a monitoring system for WWTP in recent years [30,31].
Advanced strategies to improve nitrification process in sequencing batch reactors - A review
2018, Journal of Environmental ManagementCitation Excerpt :Aerobic phase length control and DO control are examples of strategies based on online measurements of ammonium, which have been proposed for energy savings (Hajji et al., 2016; Olsson, 2012). Nevertheless, the major disadvantages of these analyzers are their high investment and maintenance costs (Casellas et al., 2006; Ruano et al., 2009), significant time delay (Cecil and Kozlowska, 2010), and complex operation (Yang et al., 2010). To overcome these drawbacks several researchers have online monitored physical-chemical parameters such as DO, pH, ORP (Akin and Ugurlu, 2005; Won and Ra, 2011) and oxygen uptake rate (OUR) (Puig et al., 2005) (also called secondary variables) which can be used to estimate the duration of the nitrification and denitrification phases.
A data-derived soft-sensor method for monitoring effluent total phosphorus
2017, Chinese Journal of Chemical EngineeringStatistical monitoring and dynamic simulation of a wastewater treatment plant: A combined approach to achieve model predictive control
2017, Journal of Environmental Management