Search

Principal component analysis (PCA) is by far the most widespread tool for unsupervised learning with high-dimensional data sets. It is popularly studied for exploratory data analysis and online pro...

Due to the growing amount of data from in-situ sensors in environmental monitoring, it becomes necessary to automatically detect anomalous data points. Nowadays, this is mainly performed using supervised machine learning models, which need a fully labelled data set for their training process. However, the process of labelling data is typically cumbersome and, as a result, a hindrance to the adoption of machine learning methods for automated anomaly detection. In this work, we propose to address this challenge by means of active learning. This method consists of querying the domain expert for the labels of only a selected subset of the full data set. We show that this reduces the time and costs associated to labelling while delivering the same or similar anomaly detection performances. Finally, we also show that machine learning models providing a nonlinear classification boundary are to be recommended for anomaly detection in complex environmental data sets. © 2020 The Authors, Environmental Modelling & Software, 134, ISSN:1364-8152, ISSN:1873-6726

In engineering practice, model-based design requires not only a good process-based model, but also a good description of stochastic disturbances and measurement errors to learn credible parameter values from observations. However, typical methods use Gaussian error models, which often cannot describe the complex temporal patterns of residuals. Consequently, this results in overconfidence in the identified parameters and, in turn, optimistic reactor designs. In this work, we assess the strengths and weaknesses of a method to statistically describe these patterns with autocorrelated error models. This method produces increased widths of the credible prediction intervals following the inclusion of the bias term, in turn leading to more conservative design choices. However, we also show that the augmented error model is not a universal tool, as its application cannot guarantee the desired reliability of the resulting wastewater reactor design. © 2020 Elsevier Ltd Marc B. Neumann acknowledges financial support provided by the Spanish Government through the BC3 María de Maeztu excellence accreditation 2018–2022 (MDM-2017-0714) and the Ramón y Cajal grant (RYC-2013-13628); and by the Basque Government through the BERC 2018-2021 program.

The aim of this paper is to describe the state-of-the art computer-based techniques for data analysis to improve operation of wastewater treatment plants. A comprehensive review of peer-reviewed papers shows that European researchers have led academic computer-based method development during the last two decades. The most cited techniques are artificial neural networks, principal component analysis, fuzzy logic, clustering, independent component analysis and partial least squares regression. Even though there has been progress on techniques related to the development of environmental decision support systems, knowledge discovery and management, the research sector is still far from delivering systems that smoothly integrate several types of knowledge and different methods of reasoning. Several limitations that currently prevent the application of computer-based techniques in practice are highlighted.

Ammonia control in municipal wastewater treatment plants typically requires maintenance-intensive instrumentation. A low maintenance alternative is sought for small- to medium-scale applications. To this end, a pH-based soft-sensor is proposed to detect ammonia peak load events. This soft-sensor is based on a newly developed technique for qualitative trend analysis and is combined with a rule-based controller. The use of qualitative trend analysis makes this soft-sensor tolerant towards sensor drifts and thereby reduces the maintenance effort. The method allows controlling any process in which relative changes in the measured output are informative about the system output.

The study of techniques for qualitative trend analysis (QTA) has been a popular approach to address challenges in fault diagnosis of engineered processes. Such challenges include the lack of reliable extrapolation of available models and lack of representative data describing previously unseen circumstances. Many of these challenges appear in biological systems even when normal operation can be assumed. It is for this reason that QTA techniques have also been proposed for the purpose of fault detection, automation, and dynamic modeling. In this work, we adopt a shape-constrained spline function method for the purpose of unknown input estimation. Thanks to data collected at laboratory-scale in a biological reactor for urine nitrification, this novel approach has been demonstrated successfully for the first time.

To this day, obtaining reliable characterization of sludge settling properties remains a challenging and time-consuming task. Without such assessments however, optimal design and operation of secondary settling tanks is challenging and conservative approaches will remain necessary. With this study, we show that automated sludge blanket height registration and zone settling velocity estimation is possible thanks to analysis of images taken during batch settling experiments. The experimental setup is particularly interesting for practical applications as it consists of off-the-shelf components only, no moving parts are required, and the software is released publicly. Furthermore, the proposed multivariate shape constrained spline model for image analysis appears to be a promising method for reliable sludge blanket height profile registration.

The development and validation of methods for fault detection and identification in wastewater treatment research today relies on two important assumptions: {\em (i)} that sensor faults appear at distinct times in different sensors and {\em (ii)} that any given sensor will function near-perfectly for a significant amount of time following installation. In this work, we show that such assumptions are unrealistic, at least for sensors built around an ion-selective measurement principle. Indeed, long-term exposure of sensors to treated wastewater shows that sensors exhibit important fault symptoms that appear simultaneously and with similar intensity. Consequently, our work suggests that focus of research on methods for fault detection and identification should be reoriented towards methods that do not rely on the assumptions mentioned above. This study also provides the very first empirically validated sensor fault model for wastewater treatment simulation and we recommend its use for effective benchmarking of both fault detection and identification methods and advanced control strategies. Finally, we evaluate the value of redundancy for the purpose of remote sensor validation in decentralized wastewater treatment systems.

The control of nitrite-oxidizing bacteria (NOB) challenges the implementation of partial nitritation and anammox (PN/A) processes under mainstream conditions. The aim of the present study was to understand how operating conditions impact microbial competition and the control of NOB in hybrid PN/A systems, where biofilm and flocs coexist. A hybrid PN/A moving-bed biofilm reactor (MBBR; also referred to as integrated fixed film activated sludge or IFAS) was operated at 15 °C on aerobically pre-treated municipal wastewater (23 mgNH4-N·L−1). Ammonium-oxidizing bacteria (AOB) and NOB were enriched primarily in the flocs, and anammox bacteria (AMX) in the biofilm. After decreasing the dissolved oxygen concentration (DO) from 1.2 to 0.17 mgO2·L−1 - with all other operating conditions unchanged - washout of NOB from the flocs was observed. The activity of the minor NOB fraction remaining in the biofilm was suppressed at low DO. As a result, low effluent NO3− concentrations (0.5 mgN·L−1) were consistently achieved at aerobic nitrogen removal rates (80 mgN·L−1·d−1) comparable to those of conventional treatment plants. A simple dynamic mathematical model, assuming perfect biomass segregation with AOB and NOB in the flocs and AMX in the biofilm, was able to qualitatively reproduce the selective washout of NOB from the flocs in response to the decrease in DO-setpoint. Similarly, numerical simulations indicated that flocs removal is an effective operational strategy to achieve the selective washout of NOB. The direct competition for NO2− between NOB and AMX - the latter retained in the biofilm and acting as a “NO2-sink” - was identified by the model as key mechanism leading to a difference in the actual growth rates of AOB and NOB (i.e., μNOB < μAOB in flocs) and allowing for the selective NOB washout. Experimental results and model predictions demonstrate the increased operational flexibility, in terms of variables that can be easily controlled by operators, offered by hybrid systems as compared to solely biofilm systems for the control of NOB in mainstream PN/A applications.HighlightsHybrid PN/A systems provide increased operational flexibility for NOB controlAOB and NOB enrich primarily in the flocs, and AMX in the biofilm (“NO2-sink”)AMX use NO2− allowing to differentiate AOB and NOB growth ratesA decrease in DO or an increase in floc removal leads to selective NOB washout from flocsThe activity of the minor NOB fraction in the biofilm is suppressed at limiting DO

Environmental process modeling is challenged by the lack of high quality data, stochastic variations, and nonlinear behavior. Conventionally, parameter optimization is based on stochastic sampling techniques to deal with the nonlinear behavior of the proposed models. Despite widespread use, such tools cannot guarantee globally optimal parameter estimates. It can be especially difficult in practice to differentiate between lack of algorithm convergence, convergence to a non-global local optimum, and model structure deficits. For this reason, we use a deterministic global optimization algorithm for kinetic model identification and demonstrate it with a model describing a typical batch experiment. A combination of interval arithmetic, reformulations, and relaxations allows globally optimal identification of all (six) model parameters. In addition, the results suggest that further improvements may be obtained by modification of the optimization problem or by proof of the hypothesized pseudo-convex nature of the problem suggested by our results. Display Omitted Objective function bounds for global biokinetic parameter optimization are proven.Global deterministic parameter estimation of a six-parameter model is possible.Globally optimal parameters are obtained to fit the model to experimental data.

Obtaining high quality data collected on wastewater treatment plants is gaining increasing attention in the wastewater engineering literature. Typical studies focus on recognition of faulty data with a given set of installed sensors on a wastewater treatment plant. Little attention is however given to how one can install sensors in such a way that fault detection and identification can be improved. In this work, we develop a method to obtain Pareto optimal sensor layouts in terms of cost, observability, and redundancy. Most importantly, the resulting method allows reducing the large set of possibilities to a minimal set of sensor layouts efficiently for any wastewater treatment plant on the basis of structural criteria only, with limited sensor information, and without prior data collection. In addition, the developed optimization scheme is fast. Practically important is that the number of sensors needed for both observability of all flows and redundancy of all flow sensors is only one more compared to the number of sensors needed for observability of all flows in the studied wastewater treatment plant configurations.

Anomaly detection is critical to process modeling, monitoring, and control since successful execution of these engineering tasks depends on access to validated data. Classical methods for data validation are quantitative in nature and require either accurate process knowledge, large representative data sets, or both. In contrast, a small section of the fault diagnosis literature has focused on qualitative data and model representations. The major benefit of such methods is that imprecise but reliable results can be obtained under previously unseen process conditions. This work continues with a line of work focused on qualitative trend analysis which is the qualitative approach to data series analysis. An existing method based on shape-constrained spline function fitting is expanded to deal explicitly with discontinuities and is applied here for the first time for anomaly detection. An experimental test case and a comparison with the principal component analysis method bear out the benefits of the qualitative approach to process monitoring.

The advent of affordable computing, low-cost sensor hardware, and high-speed and reliable communications have spurred ubiquitous installation of sensors in complex engineered systems. However, ensuring reliable data quality remains a challenge. Exploitation of redundancy among sensor signals can help improving the precision of measured variables, detecting the presence of gross errors, and identifying faulty sensors. The cost of sensor ownership, maintenance efforts in particular, can still be cost-prohibitive however. Maximizing the ability to assess and control data quality while minimizing the cost of ownership thus requires a careful sensor placement. To solve this challenge, we develop a generally applicable method to solve the multi-objective sensor placement problem in systems governed by linear and bilinear balance equations. Importantly, the method computes all Pareto-optimal sensor layouts with conventional computational resources and requires no information about the expected sensor quality.

The computation and modeling of extents has been proposed to handle the complexity of large-scale model identification tasks. Unfortunately, the existing extent-based framework only applies when certain conditions apply. Most typically, it is required that a unique value for each extent can be computed. This severely limits the applicability of this approach. In this work, we propose a novel procedure for parameter estimation inspired by the existing extent-based framework. A key difference with prior work is that the proposed procedure combines structural observability labeling, matrix factorization, and graph-based system partitioning to split the original model parameter estimation problem into parameter estimation problems with the least number of parameters. The value of the proposed method is demonstrated with an extensive simulation study and a study based on a historical data set collected to characterize the isomerization of &alpha, pinene. Most importantly, the obtained results indicate that an important barrier to the application of extent-based frameworks for process modeling and monitoring tasks has been lifted.

Sensor maintenance is time-consuming and is a bottleneck for monitoring on-site wastewater treatment systems. Hence, we compare maintained and unmaintained sensors to monitor the biological performance of a small-scale sequencing batch reactor (SBR). The sensor types are ion-selective pH, optical dissolved oxygen (DO), and oxidation-reduction potential (ORP) with platinum electrode. We created soft sensors using engineered features: ammonium valley for pH, oxidation ramp for DO, and nitrite ramp for the ORP. Four soft sensors based on unmaintained pH sensors correctly identified the completion of the ammonium oxidation (89-91 out of 107 cycles), about as many times as soft sensors based on a maintained pH sensor (91 out of 107 cycles). In contrast, the DO soft sensor using data from a maintained sensor showed slightly better (89 out of 96 cycles) detection performance than that using data from two unmaintained sensors (77, respectively 82 out of 96 correct). Furthermore, the DO soft sensor using maintained data is much less sensitive to the optimisation of cut-off frequency and slope tolerance than the soft sensor using unmaintained data. The nitrite ramp provided no useful information on the state of nitrite oxidation, so no comparison of maintained and unmaintained ORP sensors was possible in this case. We identified two hurdles when designing soft sensors for unmaintained sensors: i) Sensors' type- and design-specific deterioration affects performance. ii) Feature engineering for soft sensors is sensor type specific, and the outcome is strongly influenced by operational parameters such as the aeration rate. In summary, the results with the provided soft sensors show that frequent sensor maintenance is not necessarily needed to monitor the performance of SBRs. Without sensor maintenance monitoring small-scale SBRs becomes practicable, which could improve the reliability of unstaffed on-site treatment systems substantially.

The development of a wide array of process technologies to enable the shift from conventional biological wastewater treatment processes to resource recovery systems is matched by an increasing demand for predictive capabilities. Mathematical models are excellent tools to meet this demand. However, obtaining reliable and fit-for-purpose models remains a cumbersome task due to the inherent complexity of biological wastewater treatment processes. In this work, we present a first study in the context of environmental biotechnology that adopts and explores the use of extents as a way to simplify and streamline the dynamic process modeling task. In addition, the extent-based modeling strategy is enhanced by optimal accounting for nonlinear algebraic equilibria and nonlinear measurement equations. Finally, a thorough discussion of our results explains the benefits of extent-based modeling and its potential to turn environmental process modeling into a highly automated task.

Sensor drift is commonly observed across engineering disciplines, particularly in harsh media such as wastewater. In this study, a novel stabilizing controller for nitrification of high strength ammonia solutions is designed based on online signal derivatives. The controller uses the derivative of a drifting nitrite signal to determine if nitrite-oxidizing bacteria (NOB) are substrate limited or substrate inhibited. To ensure a meaningful interpretation of the derivative signal, the process is excited in a cyclic manner by repeatedly exposing the NOB to substrate-limited and substrate-inhibited conditions. The resulting control system successfully prevented nitrite accumulations for a period of 72 days in a laboratory-scale reactor. Slow disturbances in the form of feed composition changes and temperature changes were successfully handled by the controller while short-term temperature disturbances are shown to pose a challenge to the current version of this controller. Most importantly, we demonstrate that drift-tolerant control for the purpose of process stabilization can be achieved without sensor redundancy by combining deliberate input excitation, qualitative trend analysis, and coarse process knowledge.

International audience; This paper serves as a problem statement of the issues surrounding uncertainty in wastewater treatment modelling. The paper proposes a structure for identifying the sources of uncertainty introduced during each step of an engineering project concerned with model-based design or optimisation of a wastewater treatment system. It briefly references the methods currently used to evaluate prediction accuracy and uncertainty and discusses the relevance of uncertainty evaluations in model applications. The paper aims to raise awareness and initiate a comprehensive discussion among professionals on model prediction accuracy and uncertainty issues. It also aims to identify future research needs. Ultimately the goal of such a discussion would be to generate transparent and objective methods of explicitly evaluating the reliability of model results, before they are implemented in an engineering decision-making context.

The identification of kinetic models can be simplified via the computation of extents of reaction on the basis of invariants such as stoichiometric balances. With extents, one can identify the structure and the parameters of reaction rates individually, which significantly reduces the number of parameters that need to be estimated simultaneously. So far, extent-based modeling has only been applied to cases where all the extents can be computed from measured concentrations. This generally excludes its application to many biological processes since the number of reactions tends to be larger than the number of measured quantities. This paper shows that, in some cases, such restrictions can be lifted. In addition, in contrast to most extent-based modeling studies that have dealt with simulated data, this study demonstrates the applicability of extent-based model identification using laboratory experimental data.

Empirical model identification for biological systems is a challenging task due to the combined effects of complex interactions, nonlinear effects, and lack of specific measurements. In this context, several researchers have provided tools for experimental design, model structure selection, and optimal parameter estimation, often packaged together in iterative model identification schemes. Still, one often has to rely on a limited number of candidate rate laws such as Contois, Haldane, Monod, Moser, and Tessier. In this work, we propose to use shape-constrained spline functions as a way to reduce the number of candidate rate laws to be considered in a model identification study, while retaining or even expanding the explanatory power in comparison to conventional sets of candidate rate laws. The shape-constrained rate laws exhibit the flexibility of typical black-box models, while offering a transparent interpretation akin to conventionally applied rate laws such as Monod and Haldane. In addition, the shape-constrained spline models lead to limited extrapolation errors despite the large number of parameters. (C) 2017 Elsevier Ltd. All rights reserved.