A machine learning framework to improve effluent quality control in wastewater treatment plants.

Due to the intrinsic complexity of wastewater treatment plant (WWTP) processes, it is always challenging to respond promptly and appropriately to the dynamic process conditions in order to ensure the quality of the effluent, especially when operational cost is a major concern. Machine Learning (ML) methods have therefore been used to model WWTP processes in order to avoid various shortcomings of conventional mechanistic models. However, to the best of the authors' knowledge, no ML applications have focused on investigating how operational factors can affect effluent quality. Additionally, the time lags between process steps have always been neglected, making it difficult to explain the relationships between operational factors and effluent quality. Therefore, this paper presents a novel ML-based framework designed to improve effluent quality control in WWTPs by clarifying the relationships between operational variables and effluent parameters. The framework consists of Random Forest (RF) models, Deep Neural Network (DNN) models, Variable Importance Measure (VIM) analyses, and Partial Dependence Plot (PDP) analyses, and uses a novel approach to account for the impact of time lags between processes. Details of the framework are provided along with a demonstration of its practical applicability based on a case study of the Umeå WWTP in Sweden involving a large number of samples (105763) representing the full scale of the plant's operations. Two effluent parameters, Total Suspended Solids in effluent (TSSe) and Phosphate in effluent (PO4e), and thirty-two operational variables are studied. RF models are developed, validated using DNN models as references, and shown to be suitable for VIM and PDP analyses. VIM identifies the variables that most strongly influence TSSe and PO4e, while PDP elucidates their specific effects on TSSe and PO4e. The major findings are: (1) Influent temperature is the most influential variable for both TSSe and PO4e, but it affects them in different ways; (2) PO4e depends strongly on the TSS in aeration basins - higher TSS concentrations in aeration basins generally promote PO4 removal, but excess TSS can have negative effects; (3) In general, the impact of TSS in aeration basins on TSSe and PO4e increases with the distances of the basin from the merging outlet, so more attention should be paid to the TSS concentration in the third or fourth aeration basins than the first and second ones; (4) Returning excessive amounts of sludge through the second return sludge pipe should be avoided because of its adverse impact on TSSe removal. These results could support the development of more advanced control strategies to increase control precision and reduce running costs in the Umeå WWTP and other similarly configured WWTPs. The framework could also be applied to other parameters in WWTPs and industrial processes in general if sufficient high-resolution data are available.

OPLS in batch monitoring - Opens up new opportunities.

In batch statistical process control (BSPC), data from a number of "good" batches are used to model the evolution (trajectory) of the process and they also define model control limits, against which new batches may be compared. The benchmark methods used in BSPC include partial least squares (PLS) and principal component analysis (PCA). In this paper, we have used orthogonal projections to latent structures (OPLS) in BSPC and compared the results with PLS and PCA. The experimental study used was a batch hydrogenation reaction of nitrobenzene to aniline characterized by both UV spectroscopy and process data. The key idea is that OPLS is able to separate the variation in data that is correlated to the process evolution (also known as 'batch maturity index') from the variation that is uncorrelated to process evolution. This separation of different types of variations can generate different batch trajectories and hence lead to different established model control limits to detect process deviations. The results demonstrate that OPLS was able to detect all process deviations and provided a good process understanding of the root causes for these deviations. PCA and PLS on the other hand were shown to provide different interpretations for several of these process deviations, or in some cases they were unable to detect actual process deviations. Hence, the use of OPLS in BSPC can lead to better fault detection and root cause analysis as compared to existing benchmark methods and may therefore be used to complement the existing toolbox.

Near-infrared chemical imaging (NIR-CI) on roll compacted ribbons and tablets--multivariate mapping of physical and chemical properties.

Near-infrared chemical imaging (NIR-CI) is an attractive technique within the pharmaceutical industry, where tools are continuously in demand to assess the quality of the intermediate and final products. The present paper demonstrates how NIR-CI in combination with multivariate methods was utilized to spatially map physical properties and content of roll compacted ribbons and tablets. Additionally, extracted textural parameters from tablet images were correlated to the design parameters of the roll compaction process as well as to the physical properties of the granules. The results established the use of NIR-CI as a complementary nondestructive tool to determine the ribbon density and map the density distribution across the width and along the length of the ribbons. For the tablets, the compaction pressure developed during compression increased with the lateral distance from the center. Therefore, NIR-CI can be an effective tool to provide information about the spatial distribution of the compaction pressures on the surface of the tablet. Moreover, low roll compaction roll force correlated to a heterogeneous type of texture in the API chemical image. Overall, texture analysis of the tablets enabled efficient investigation of the spatial variation and could be used to advance process understanding. Finally, orthogonal projections to latent structures (O2PLS) model facilitated the understanding of the interrelationships between textural features, design parameters and physical properties data by separately joint and unique variations.

Roll compaction process modeling: transfer between equipment and impact of process parameters.

In this study, the roll compaction of an intermediate drug load formulation was performed using horizontally and vertically force fed roll compactors. The horizontally fed roll compactor was equipped with an instrumented roll technology allowing the direct measurement of normal stress at the roll surface, while the vertically fed roll compactor was equipped with a force gauge between the roll axes. Furthermore, characterization of ribbons, granules and tablets was also performed. Ribbon porosity was primarily found to be a function of normal stress, exhibiting a quadratic relationship thereof. A similar quadratic relationship was also observed between roll force and ribbon porosity of the vertically fed roll compactor. The predicted peak pressure (Pmax) using the Johanson model was found to be higher than the measured normal stress, however, the predicted Pmax correlated well with the ribbon relative density/porosity and the majority of downstream properties of granules and tablets, demonstrating its use as a scale-independent parameter. A latent variable model was developed for both the horizontal and vertical fed roll compactors to express ribbon porosity as a function of geometric and process parameters. The model validation, performed with new data, resulted in overall good predictions. This study successfully demonstrated the scale up/transfer between two different roll compactors and revealed that the combined use of design of experiments, latent variable models and in silico predictions result in better understanding of the critical process parameters in roll compaction.

A quality by design approach to investigate the effect of mannitol and dicalcium phosphate qualities on roll compaction.

Roll compaction is a continuous process for solid dosage form manufacturing increasingly popular within pharmaceutical industry. Although roll compaction has become an established technique for dry granulation, the influence of material properties is still not fully understood. In this study, a quality by design (QbD) approach was utilized, not only to understand the influence of different qualities of mannitol and dicalcium phosphate (DCP), but also to predict critical quality attributes of the drug product based solely on the material properties of that filler. By describing each filler quality in terms of several representative physical properties, orthogonal projections to latent structures (OPLS) was used to understand and predict how those properties affected drug product intermediates as well as critical quality attributes of the final drug product. These models were then validated by predicting product attributes for filler qualities not used in the model construction. The results of this study confirmed that the tensile strength reduction, known to affect plastic materials when roll compacted, is not prominent when using brittle materials. Some qualities of these fillers actually demonstrated improved compactability following roll compaction. While direct compression qualities are frequently used for roll compacted drug products because of their excellent flowability and good compaction properties, this study revealed that granules from these qualities were more poor flowing than the corresponding powder blends, which was not seen for granules from traditional qualities. The QbD approach used in this study could be extended beyond fillers. Thus any new compound/ingredient would first be characterized and then suitable formulation characteristics could be determined in silico, without running any additional experiments.