Improved management of water resources in process industry by accounting for fluctuations of water content in feed streams and products

Highlights

• Fluctuating waters are of primary importance for several water-consuming industries.

• Common strategies not consider fluctuating water, limiting the overall recovery.

• Fluctuating waters could prohibitively increase complexity and computational burden.

• Statistical information can allow counting fluctuation in reconciliation procedures.

• A proper strategy using ordinal optimization can overcome the resulting complexity.

Abstract

In this paper, a general strategy is presented for optimizing water resources in industrial activities taking into account fluctuations of water flow-streams and water contents in feed and products as well as in the utility system. Indeed, considering fluctuations can be of primary importance for several water intensive industries such as the food or the paper&pulp industry. Furthermore, whenever the water balance includes the water content of products (as is typically the case for the food industry) fluctuations in the composition of products can considerably affect the overall balance due to water being a major component of many foods. Therefore, the currently used optimization strategies that disregard water fluctuations, can lead to a severe bias in the water balance and thus overlook the potential of water recovery and reuse.

Including this aspect in widely used algorithms for process monitoring (data reconciliation) and optimization might lead to a possibly prohibitive increase in the computational burden. In particular, it is shown in this paper that the resulting optimization algorithm should solve a dynamic non-convex stochastic Mixed Integer Nonlinear Programming problem.

To tackle this (typically NP-hard) problem, a strategy has been developed that combines ordinal optimization for dealing with statistical information and nonlinear, non-convex deterministic algorithms.

The outlined strategy has been applied to a complex process in the food industry (production of starch and starch-based products). The results seem to confirm the general validity of the algorithm developed.

Introduction

Water is the main component of almost all living species, and a key element in human life. Indeed, practically all human activities depend directly or indirectly on water. On a global average, the main water-demanding sectors are agriculture, industries and urbanization, absorbing around 70, 20 and 10 % of the amount withdrawn from water resources, respectively [1].

In particular, the demand for water from industries worldwide is expected to increase by 400 % in 2050 [2] in view of the rapid growth in global industrialization. While several industrial sectors can be considered water intensive, including electrochemical processes [3], the garment and textile, the pulp&paper and the automotive industry [4], the industrial food sector – defined as the set of industrial activities that process agricultural and farming raw materials – is one of the most demanding industrial activities in terms of water: since water contributes, on average, about 80 % of the total weight of food materials, it determines the majority of its properties (structure, appearance, taste, and stability) and the ways in which a product can be processed. In addition to being an ingredient in a large number of products, water generally enters the food industry, similarly to most other industrial activities, in two main ways: i) process applications (e.g. as a reactant) and ii) non-process applications (e.g. resources and washing, cooling and heating) [5].

The pressure for reducing the amount of water consumed is further increased by the ever stricter regulations on wastewater discharges, which force continuously rising standards of ‘Best Available Technologies’ (BATs) and/or a simultaneous increase of effluent taxes. Some examples of this twofold strategy can be found in Jetoo [6] and Möller-Gulland et al. [7]. Indeed, highly polluted wastewater is produced in significant quantities by the processing industry, which can cause intense damages to the environment, especially if accidentally discharged [8]. The food industry is no exception, as relevant amounts of organic compounds, nitrogen, phosphorous and heavy metals are released with wastewater. In particular, depending on the different food processed, Noukeu et al. [9] report for the mentioned substances the following concentration ranges (mg/L): COD = 457–357’725; NH₄⁺ = 1.23–2925; PO₄³⁻ = 3.567–3240; Pb⁺ = 0.083–1.025; Cd⁺ = 0.052–0.158. The improper release of wastewaters with similar amounts of contaminants can severely pollute the water resources of the soil, surface and underground, requiring specific reclamation strategies to contain/remove the contamination [[10], [11], [12]]. Additionally, as for the meat industry [13], wastewater may also contain chromium and tannins that are directly involved in the deaths of aquatic animals [14,15].

The water footprint and water neutrality concepts were introduced to estimate the pressure of a specific product or an industrial activity on water resources. However, while the former does not provide any information on how to reduce water consumption or its impact on the environment, the latter still has some drawbacks, which hinder its establishment as a strong and meaningful concept [16]. Indeed, the definition of water neutrality as the requirement that "for every new development, the total water use in the region after development must be equal to or less than total water use in the region before development" is generally considered to be potentially ambiguous as key issues such as "development" and "region" are not clearly specified [17].

In principle, the techniques developed for general applications with a view to optimizing water consumption can be applied to the food industry provided the amount of water in final products is duly taken into account.

Water pinch analysis (WPA) and mathematical optimization are two alternative or complementary strategies that are frequently used to minimize water consumption and wastewater generation. Indeed, while the latter approach is more rigorous and general because it does not require the approximations contained in the pinch analysis (e.g. constant activity and transfer coefficients) it can occasionally lead to convergence problems. Thus, the two methods can be employed alternatively after considering the complexity of the model and the consequences of the simplifications contained in the pinch analysis or complementarily using the results provided by the pinch analysis as starting values for the subsequent optimization problem.

The steadily increasing price of water (over 40 % from 2010 to 2015 in the largest USA cities [18]) has forced both process designers and plant operators to consider treatment efficiency, as well as potential reuse and recycling options as key issues in the general optimization procedure.

Unlike the energy case, savings in freshwater and wastewater in countries where availability of water still does not constitute a particular problem are unlikely, on their own, to justify the cost of carrying out a detailed water pinch study and implementing the recommended projects. This results in a slower integration of these concepts into the standard industrial design procedures, particularly in the case of retrofit projects focused on existing facilities. Furthermore, water networks subjected to existing and not recently built industrial complexes, typically only have the minimum instrumentation required by the operation, and measurements of most of the water flows are not available. This lack of information hampers the implementation of retrofit projects, starting from the data extraction phase when it is essential to assess with good reliability the freshwater grades and flow rates supplied to the process units and to treatment facilities in the current state [19].

Undeniably, this trend is not uniform in all cases: there are some industrial sectors (for example wood pulp and paper) that involve a relevant consumption of freshwater, and regional areas (for example the Middle East), where water scarcity hampers industrial development or where the actual cost of freshwater is, in absolute terms, higher or comparable to the cost of energy, that have seen the flourishing of various project and applications aimed at optimizing this resource.

However, in the food industry, additional difficulties arise from the multiple use of water, in particular because of the frequent incorporation of water streams into products, which adds further limits to the quality of the water circulating in the plant and their use/regeneration before being discharged. This also leads to the impossibility to directly measure the water content of process streams (raw materials, intermediate and finished products), which is then considered jointly to the flowrate of the process streams. Nevertheless, fluctuations in the mixed streams due a variable water content should be properly addressed in the overall data reconciliation process, because they can have an incidence even higher than measurement errors in process streams, especially in the presence of evaporation losses, which cannot be measured and involve a compensation effect that might hide the unbalances due to water content oscillations.

In this contribution, the focus is in particular on the development of a suitable model for the optimal management of water resources, which allows considering water content fluctuations in both feed streams and products, and is to be considered the main result of the work.

Indeed, water balances can be considerably affected by the presence of comparatively large fluctuations of streams and product compositions. Neglecting them can lead to a suboptimal operational performance of the control system and consequently to an increased water consumption as verified in the industrial case considered.

The reconciliation of dynamic data has been analyzed using different approaches, such as those based on Kalman filtering [20] or modified versions of it (Bai et al., 2006) [52], on nonlinear programming algorithms for the minimization of deviations on least squares over a pre-determined time window [21], and on integral methods as suggested by Bagajewicz and Jiang [22].

Their use implies the knowledge of the analytical model of the process (typically given by a DAE system), the selection of a limited time window and the impossibility of monitoring the presence of sources and sinks (such as unit hold-up fluctuations and leaks). Unfortunately, in the case analyzed, the use of the overall process model could hardly be taken into consideration due to the prohibitive computational effort necessary to simulate the high-dimensional, non-convex, severely nonlinear problem resulting from the model of complex (and frequently approximately known) set of reactions that take place in the process. Similarly, the time window that could be considered is given by the frequency of quality control tests that determine the water content of products, which turns out to be much higher than the one required by methods that compute time derivatives using discretization techniques. However, the water content fluctuations in the product (which correspond to sinks and leaks in the approaches described) are indeed available on a statistical basis. These three basic differences require the use of a different approach that will be illustrated in the following section.

The next section provides information on the difficulties underlying this type of optimization problem, while Section 3 illustrates the overall mathematical model, highlighting the differences with respect to processes where no additional uncertainty is present due to fluctuating amounts of water in the products. Subsequently, the application of the model developed to a real case study is outlined: general information on the considered plant is provided (a real bio-refinery for raw maize processing) along with some results obtained using the procedure described in this document – the amount of information is limited for reasons of confidentiality. Since corn refineries are among the major water-demanding food industries, requiring approximately 4300 million m³ of freshwater and producing about 3800 million m³ of wastewater worldwide [5], the application of suitable solutions for the management of water consumption and wastewater generation in these industrial units can be of great importance.

The development of practical methodologies to optimize the use of water resources even in complex but relevant activities, as in the case of the food industry, will become increasingly critical in the coming decades, and is also in line with strategies such as the Circular economy, the recourse to BATs, and the ‘near-zero discharge’ of hazardous wastes [23]. A transition to a circular economy could create significant synergies for the widespread adoption of water reuse as an alternative water supply, overcoming many of the barriers that water reuse is still facing, ranging from public perception to price and regulatory challenges.