Pilot size matters: the case of a full scale pilot for P-recovery from digestate in Belgium

Aquafin owns and operates 281 wastewater treatment plants (WWTPs) in Flanders, Belgium, treating wastewater from around 5 million population equivalent (ppe). In 2012 the company decided to investigate P-recovery from digested sludge as the mineral form struvite. Struvite is a crystalized mineral with the chemical formula MgNH₄PO₄.6H₂O.

For water utilities, the problem of increasingly scarce phosphorus (P) globally is of some importance, because large amounts pass daily through WWTPs (Cordell et al. 2009). On the other hand, actual operating cost-benefits are still the major triggers for investment in relatively new technologies such as the recovery of P as struvite from digested sludge. The claimed benefits of this technique, including financial impact, relate mainly to the enhanced dewaterability of the treated sludge, reduced risk of clogged pipes and reduced return flow of P to the main WWTP (Marchi et al. 2015).

This paper describes how uncertainties about sensitive operating parameters could be decreased by moving from a theoretical economic analysis to a pilot-plant of suitable size, and how wrong conclusions could be drawn from small-scale tests.

Leuven WWTP treats the municipal wastewater of 120,000 ppe. It is also a central sludge treatment facility for the operator, so that additional liquid and dewatered sludge are transported there for further treatment. The sludge treatment line comprises thickening tables, an anaerobic digester, centrifuges and a sludge dryer.

Early in 2013, a struvite recovery process (NuReSys, Waregem, Belgium) was installed after the anaerobic digester. The WWTP removes P biologically, so there is considerable release of PO₄^3– in the digester, arising from cell hydrolysis. Because of this and other operating benefits noted above, the struvite recovery process was considered promising enough for investigation. The installation and recovery technique are described in Marchi et al. (2015).

Work began in 2012 with a literature-based feasibility study (e.g. Cornel & Schaum 2009) and suppliers data. After selecting the most promising technology supplier, a full-scale pilot was built and started working in April 2013. Four monitoring campaigns were carried out in the spring, summer, autumn and winter of 2013, when the operating settings – e.g., the MgCl₂ dosing ratio and the pH – were optimized. In 2014 and early 2015, the focus was on the optimization of nucleation, crystal growth and separation. The results enabled a detailed economic model of the plant to be built, which was used to carry out a sensitivity and uncertainty analysis.

The basis of the economic model's uncertainty analysis is described in Sin et al. (2011), Geerts et al. (2015) and Donckels et al. (2014). Initially, the sensitivity and uncertainty analysis was carried out by varying economic model inputs within plausible ranges obtained from the literature and/or the supplier, and selecting uniform distributions (e.g. STOWA 2012; TAUW 2012). After pilot testing, ranges were narrowed, and/or shifted, and/or received a Gaussian or Gamma distribution, when enough statistical evidence had been generated in the pilot testing. For each of the uncertainty analyses, 1,100 Monte Carlo simulations were carried out to shed light on the financial feasibility of struvite recovery in Leuven.

Even if a water utility invests in a pilot test, its total duration (e.g. related to seasonal variations in sludge) and size of the pilot are determinants of the financial study's total uncertainty. For variables that are measurable at between lab- and full-scale, the standard deviations of measured outcomes are directly related to the input variables' uncertainty ranges. Variables that cannot be assessed at any specific scale and/or during a specific test period will still lead to literature-based error ranges around themselves, often with uniform distributions, in the Monte Carlo runs.

At three stages during the project, the Monte Carlo simulations produced different financial feasibility results for struvite recovery at Leuven. The variation ranges of the input variables for the Monte Carlo runs are presented in Table 1. Moving from the desk study to the full-scale pilot test led to narrower error ranges (‘selection ranges' for the Monte Carlo runs) and changes in distributions. The increased testing period (stage 3 analysis) led to further narrowing of the error ranges around uncertain variables and to the definition of the inter-parameter dependency between dewaterability, and MgCl₂- and NaOH-dosing. In relation to man hours required per week, for example, it was noted that the expected work load was initially overestimated: the desk study range of 0.5 to 5 hours/week was reduced to the real value of 0.5 to 2 hours/week. The same was true for the range of possible improvement in sludge dewaterability (Table 1), for which full-scale testing indicated a range between +1 and +3% DM with normal distribution, and not −1 to +3% DM with uniform distribution. On the other hand, some uncertainty ranges remained valid throughout the studies, such as the polymer price variation range or the uncertainty range for man hour unit costs at Aquafin.

Further details of the Monte Carlo runs, including construction of the variable distributions are given in Geerts et al. (2015), which also describes how inter-parameter dependencies were taken into account for the dewaterability –Mg and –NaOH dosing relationships.

Table 2 illustrates qualitatively how certain input variables, including the sensitive ones (Geerts et al. 2015), can only be tested at relevant scale and during a sufficiently long trial.

Table 2

Qualitative assessment of test possibility and/or level of accuracy of assessing different variables in the desk study (1), short-term pilot test (6 months) (2) and longer-term pilot test (2 years) (3); red = impossible, orange = insufficient, green = satisfactory

View Large

The red, green and brown areas in Figure 3 represent the cases with production cost outcome exceeding 550 €/tonne struvite (‘predicted failure rates'), for the desk study (1), and full-scale study after six months (2) and two years (3), respectively. Three things, in particular, can be seen in Figure 3:

• ­ The right hand tail of the distributions, with very disadvantageous financial scenarios, decreased from the desk study (1), through the six-month (2) and two-year (3) pilot trials.

• ­ The distribution peak, representing the median possible scenario, moved slightly left, towards the more advantageous financial scenarios.

• ­ The chance of a financial scenario yielding a discounted payback period of 10 years for this installation, if the struvite sale price is below 0 €/tonne (left hand tail <0 €/tonne) remained the same throughout the three financial analyses (respectively 4%, 7% and 5% for tests 1 to 3 respectively).

Thus, for example, an arbitrary cutoff value for recovered struvite of 550 €/tonne could be considered. This value is not real but enables discussion of the shifts in uncertainty. (The current (early 2016) market price for struvite is between 50 and 100 €/tonne.) As noted, the probability of achieving a discounted installation payback of 10 years, with a struvite price less than or equal to 550 €/tonne increased between tests 1 and 3 (see Figure 3). The greatest, estimated, probability increase occurred between the desk study (1) and the six-month full scale assessment (2). In decision-making terms, a management board using only the desk study analysis (1) and a struvite price of € 550/tonne, would put the project on hold (estimated 36% success rate). A management board working with the final financial analysis (3) would deploy this type of project because the estimated success rate is 64%.

The proportional recovery of struvite and resulting possible improvement in sludge dewaterability, are the economic model's most sensitive parameters for this technique with respect to Leuven WWTP, in terms of standardized regression coefficients (Geerts et al. 2015). These two variables are hard to assess with enough precision in a desk study, so the initial distribution of possible financial outcomes is too broad. They are also hard to test in lab- or small-scale studies, so a full-scale pilot was installed at Leuven. The main causes of the differences between the analytical results from tests 1 and 3 (Figure 3) include better understanding of the physical recovery of struvite crystals, and knowledge of the improvement in sludge dewaterability, in terms of both absolute effect, and its relationship with Mg- and Na- dosing.

Apart from the issues of sensitive model parameters, other benefits related to the use of a full-scale pilot became apparent. These included the ability to test its inclusion within and its effects on the WWTP, the effect on pipe clogging, the effect on biological phosphorus removal within the WWTP, and the CO₂ stripping efficiency and thus NaOH consumption. These checks could not have been done at smaller scale.

Finally the total test period matters. A longer test will not necessarily reduce the extremes of uncertainty (distribution tail positions, Figure 3) because greater extremes can become apparent with increased time. However, for a promising technology, the average operation centers automatically around the median of the distributions with time, and the weight of both tails tends to decrease.

Putting an exact monetary value on the improved financial risk assessment is speculative. This makes the assessment of the benefit of improved risk estimation against the total cost of such a full scale installation difficult. From the Monte Carlo runs of financial assessments 1 to 3, the 1,100 profitability results can be integrated by determining the vector sum of every point result and scaling it to 1. This is a probabilistic estimate of the unit production cost per tonne of struvite/year, to yield a discounted payback of 10 years. In fact, one could call it the ‘average’ unit production cost of struvite, but based on probability distributions with varying boundary conditions in financial assessments 1, 2 and 3. This value was estimated at € 896/tonne after the desk study (financial assessment 1) and finally at € 485/tonne after the 2 years of full-scale piloting (financial assessment 3). At present, Aquafin could deploy this technique only at large P-bio-removal WWTPs. If it were deployed at all such plants, annual struvite production would be approximately 1,000 tonnes, so the probabilistic estimator can be multiplied by 1,000. The difference of this between financial assessments 1 and 3 exceeds € 411,000/year. This relatively large annual value, compared to the cost of full-scale testing, illustrates theoretically the benefit of using full-scale tests for decision making in such situations.

Saying that ‘longer, bigger and case-specific testing’ is better is forcing an open door. This statement is quantified here, however, with a concrete example of use of a full-scale pilot for P-recovery from digested sludge at Leuven WWTP. As a utility for whom wastewater treatment is the core business, ‘extra’ innovative projects will always require sufficient, trustworthy data for their defense. Investing with risk is less of a problem than investing without knowing the risk. Over the last three years, the financial assessment of this recovery technique has moved away from uncertain, literature-based data from other sludges, tested at different scales in different locations.

Upsizing a pilot test is not a miraculous way forward. However, the initial uncertainty and sensitivity analysis of any new technique should at least help to define the minimum size of pilot plant needed to evaluate correctly the sensitive parameters and thus yield a more reliable economic analysis, in addition to the demonstration value of larger scale pilots. Given the current lack of corroborating results on this modern P-recovery technique, investing in a full-scale pilot was well justified, in this particular case, as well as necessary to sharpen its financial analysis.