Enhancing vehicular emissions monitoring: A GA-GRU-based soft sensors approach for HDDVs

Vehicle emissions have a serious impact on urban air quality and public health, so environmental authorities around the world have introduced increasingly stringent emission regulations to reduce vehicle exhaust emissions. Nowadays, PEMS (Portable Emission Measurement System) is the most widely used method to measure on-road NOx (Nitrogen Oxides) and PN (Particle Number) emissions from HDDVs (Heavy-Duty Diesel Vehicles). However, the use of PEMS requires a lot of workforce and resources, making it both costly and time-consuming. This study proposes a neural network based on a combination of GA (Genetic Algorithm) and GRU (Gated Recurrent Unit), which uses CC (Pearson Correlation Coefficient) to determine and simplify OBD (On-board Diagnosis) data. The GA-GRU model is trained under three real driving conditions of HDDVs, divided by vehicle driving parameters, and then embedded as a soft sensor in the OBD system to monitor real-time emissions of NOx and PN within the OBD system. This research addresses the existing research gap in the development of soft sensors specifically designed for NOx and PN emission monitoring. In this study, it is demonstrated that the described soft sensor has excellent R2 values and outperforms other conventional models. This research highlights the ability of the proposed soft sensor to eliminate outliers accurately and promptly while consistently tracking predictions throughout the vehicle's lifetime. This method is a groundbreaking update to the vehicle's OBD system, permanently adding monitoring data to the vehicle's OBD, thus fundamentally improving the vehicle's self-monitoring capabilities.

Development and application of control concepts for twin-screw wet granulation in the ConsiGmaTM-25: Part 2 granule size

Traditional operation modes, such as running the production processes at constant process settings or within a narrow design space, do not fully exploit the advantages of continuous pharmaceutical manufacturing. Integrating Quality by Control (QbC) algorithms as a standard component of production processes can mitigate the effect of diverse process disturbances and enhance process efficiency, particularly in terms of production costs and environmental footprint. This paper explores the potential of QbC algorithms for optimizing twin-screw wet granulation in the ConsiGmaTM-25 manufacturing line, specifically addressing granule size. It represents the second part of a study (Celikovic et al., 2024) focused on granule composition. The concepts proposed in this work rely on process analytical technology (PAT) equipment for real-time monitoring of the granulation CQAs and a dynamic process model linking the granulation process parameters and the monitored CQAs. The granule size model identified via the local-linear-model-tree (LoLiMoT) algorithm is used to develop both a model predictive controller (MPC) and a granule size soft sensor. The MPC employs this model as a core component for selecting optimal granulation parameters to ensure the production of granules with target size. A digital operator assistant is developed to address disturbances that cannot be mitigated via MPC but can be eliminated by the plant operators. This study systematically outlines a workflow, starting from conceptualization, moving through simulation development, and finally ending with real-world application on a production line. In this final step, all proposed concepts are transferred to the ConsiGmaTM-25 manufacturing line, where their performance is validated through selected disturbance scenarios.

Soft sensor for predicting indoor PM2.5 concentration in subway with adaptive boosting deep learning model

Public health depends on indoor air quality (IAQ), hence soft measurement techniques must be implemented in the subway environment for more precise and reliable monitoring of indoor particulate matter concentration levels. Adaptive boosting (AdaBoost), an ensemble learning technique, is simple to code and less prone to overfitting. Compared to a single model, it is better able to take into consideration the intricate elements included in air quality data. It is suggested to use an adaptive boosting of long short-term memory (AdaBoost-LSTM) model and kernel principal component analysis (KPCA) for ensemble learning. The kernel function and PCA are first coupled to create KPCA, which is a nonlinear dimensionality reduction method for IAQ. This removes the negative impacts of noise interference. The learning performance of LSTM is then enhanced using AdaBoost as an ensemble learning technique. The KPCA-AdaBoost-LSTM model can deliver higher modeling performance, according to the results. The R2 reached 0.9007 and 0.8995 when predicting PM2.5 in the hall and platform. SHapley Additive exPlanations (SHAP) analysis was used to interpret the input contributions of the model, enhancing the interpretability and transparency of the proposed soft sensor.

Deep learning based soft-sensor for continuous chlorophyll estimation on decentralized data

Monitoring the concentration of pigments like chlorophyll (Chl) in water-bodies is a key task to contribute to their conservation. However, with the existing sensor technology, measurement in real-time and with enough frequency to ensure proper risk management is not completely feasible. In this work, with the concept of data-driven soft-sensing, three hydrophysical features are used together with three meteorological ones to estimate the concentration of Chl in two tributaries of the River Thames. Data driven models, specifically neural networks, are used with three learning approaches: individual, centralized and federated. Data reduction scenarios are proposed in order to analyze the performance of each approach when less data is available. The best results in the training are usually obtained with the individual approach. However, the federated learning provides better generalization ability. It was also observed that in most of the cases the results of the federated learning approach improve those of the centralized one.

A new process analytical technology soft sensor based on electrical tomography for real-time monitoring of multiphase systems

BACKGROUND

Electrical tomography is widely recognized for its high time resolution and low cost. However, the implementation of electrical tomographic solutions has been hindered by the high computational overhead associated, which causes delays in the analysis, and numerical instability, that results in unclear reconstructed images. Therefore, it has been mostly applied offline, for qualitative tasks and with some delay. Applications requiring fast response times and quantification have been hindered or ruled out.

RESULTS

In this article, we propose a new process analytical technology soft sensor that maps directly electrical tomography signals to the relevant parameter to be monitored. The data acquisition and estimation steps occur almost instantaneously, and the final accuracy is very good (R2 = 0,994).

SIGNIFICANCE AND NOVELTY

The proposed methodology opens up good prospects for real-time quantitative applications. It was successfully tested on a pilot piping installation where the target property is the interface height between two immiscible fluids.

Soft sensor based rapid detection of trace chlorine dioxide (ClO2) concentration in water

Chlorine dioxide (ClO2) is a widely used sterilizer and a disinfectant across a multitude of industries. When using ClO2, it is imperative to measure the ClO2 concentration to abide by the safety regulations. This study presents a novel, soft sensor method based on Fourier transform infrared spectroscopy (FTIR) spectroscopy for measurement of ClO2 concentration in different water samples varying from milli Q to wastewater. Six distinct artificial neural network models were constructed and evaluated based on three overarching statistical standards to select the optimal model. The OPLS-RF model outperformed all other models with R2, RMSE, and NRMSE values of 0.945, 0.24, and 0.063, respectively. The developed model demonstrated limit of detection and limit of quantification values of 0.1 and 0.25 ppm, respectively, for water. Furthermore, the model also exhibited good reproducibility and precision as measured by the BCMSEP (0.064). The soft sensor-based method presented in the study offers significant advantages in terms of simplicity and speedy detection. In summary, the study presents development of a soft sensor that is capable of predicting the trace content of chlorine dioxide ranging between 0.1 to 5 ppm in a water sample by connecting FTIR with an OPLS-RF model.

An evolutionary deep learning soft sensor model based on random forest feature selection technique for penicillin fermentation process

Accurate and reliable measurement of key biological parameters during penicillin fermentation is of great significance for improving penicillin production. In this research context, a new hybrid soft sensor model method based on RF-IHHO-LSTM (random forest-improved​ Harris hawks optimization-long short-term memory) is proposed for penicillin fermentation processes. Firstly, random forest (RF) is used for feature selection of the auxiliary variables for penicillin. Next, improvements are made for the Harris hawks optimization (HHO) algorithm, including using elite opposition-based learning strategy (EOBL) in initialization to enhance the population diversity, and using golden sine algorithm (Gold-SA) in the search strategy to make the algorithm accelerate convergence. Then the long short-term memory (LSTM) network is constructed to build a soft sensor model of penicillin fermentation processes. Finally, the hybrid soft sensor model is used to the Pensim platform in simulation experimental research. The simulation test results show that the established soft sensor model, with high accuracy of measurement and good effect, can meet the actual requirements of engineering.

A soft sensor for simulating algal cell density based on dynamic response to environmental changes in a eutrophic shallow lake

There is a great need for timely monitoring and rapid water quality assessment to control the algal blooms that often occur in eutrophic lakes. While algal cell density (ACD) is a critical indicator of algal growth, field monitoring is laborious and time-consuming, and rapid assessment of algal blooms based on ACD is often not possible. To address the limitations of conventional ACD detection, we proposed a soft sensor approach that uses surrogate indicators to simulate ACD in machine learning models. We conducted a case study using monitoring data from Chaohu Lake collected between 2016 and 2019. We found that ensemble learning models, especially extreme gradient boosting (XGBoost), outperformed traditional machine learning algorithms by comparing various machine learning algorithms. Also, considering the influence of input variable selection on model performance, we combined the results of different filter methods in the multi-stage variable selection process. Finally, we screened out seven key variables out of the 43 initial candidate variables, including dissolved oxygen (DO), chlorophyll-a (Chl-a), Secchi disk depth (SD), pH, permanganate index (COD_Mn), week of the year (WOY), and wind velocity (WV). Their inclusion substantially improved data accessibility and supported the development of a rapid simulation model. The final model was capable of reliable spatiotemporal generalization, with an overall R² value of 0.761. On the theoretical side, our study makes a new attempt to simulate ACD values in a eutrophic lake. For practical purposes, the soft sensor can facilitate the rapid assessment of bloom conditions, which helps the local administration with emergency prevention and control.

Viable cell estimation of mammalian cells using off-gas-based oxygen uptake rate and aging-specific functional

This study developed an estimation routine for counting the viable cells in an in vitro fed-batch Chinese hamster ovary cultivation that relies on off-gas information and inlet gas mixture knowledge. We computed the oxygen uptake rate bound to the bioreactor exhaust gas outlet when the inlet gas mixture was stationary. Our mammalian biosynthesis analysis determined the stoichiometric parameters as a function of the average population age. We cross-validated an identical algorithm for mammalian and microbial cultivations and found that the' 99% confidence band of the model generally overlapped with the error bars defined from observations. The resulting RMSE and MAE averages were 0.188 and 0.14e⁹cells L^-1, respectively, when estimating the viable mammalian cell count. The validation for the estimation of total bacterial biomass yielded an MAE and RMSE of 1.78 g L^-1 and 2.53 g L^-1, respectively. Moreover, our proposed approach provides an online estimation of the average population age for both aerobically cultivated microorganisms.

Co-training based virtual sample generation for solving the small sample size problem in process industry

With the development of industrialization, the production scale and complexity of process industries are getting larger and larger. But, limited by the small amounts of samples and the uneven sample distribution in the process industry, it is difficult to establish accurate and efficient data-driven soft sensor models to predict some variables. To further develop the application of soft sensor models, generating new virtual samples based on the original sample distribution to extend the sample set is an ideal approach to solve this problem. In this paper, a novel virtual sample generation method based on the co-training of two K-Nearest Neighbor (KNN) models is proposed. First, according to the sparse parameter, sparse regions in each dimension of the feature space are identified. Second, the input features of virtual samples are generated in these sparse regions by performing interpolation operations. Third, the outputs of virtual samples are predicted by double KNN regressors based on co-training. The qualified virtual samples are screened and the model is updated using these virtual samples to improve the prediction accuracy of the double KNN models. To verify the effectiveness and superiority of the proposed virtual sample generation method based on the co-training (CTVSG), case studies are conducted using two standard functions and a Purified Terephthalic Acid (PTA) industrial dataset, where the effectiveness of CTVSG is confirmed.

A soft sensor model based on long&short-term memory dual pathways convolutional gated recurrent unit network for predicting cement specific surface area

The specific surface area of cement is an important index for the quality of cement products. But the time-varying delay, non-linearity and data redundancy in the process industry data make it difficult to establish an accurate online monitoring model. To solve the problems, a soft sensor model based on long&short-term memory dual pathways convolutional gated recurrent unit network (L/S-ConvGRU) is proposed for predicting the cement specific surface area. In this paper, first, as the linear coupling constraint inside the gated recurrent unit network (GRU) hinders the flow of information, parameters L and S are introduced into convolutional gated recurrent unit network (ConvGRU). L and S are decimals in the range (0, 1) which changed its internal linear constraint relationship and enhanced the feature extraction capability of the model. Then, two spatio-temporal feature extraction pathways are designed: long-term memory enhancement pathway and short-term dependence pathway, which capture long-term and short-term time-varying delay information from the sample data. Finally, the two feature extraction pathways mentioned above are applied to the L/S-ConvGRU model and the extracted spatio-temporal features are fused to achieve accurate prediction of the specific surface area of cement. The model was trained using raw data from the cement plant and the experimental results show that L/S-ConvGRU has higher precision and better generalization capability.

Semi-supervised echo state network with temporal-spatial graph regularization for dynamic soft sensor modeling of industrial processes

Echo state network (ESN) has been successfully applied to industrial soft sensor field because of its strong nonlinear and dynamic modeling capability. Nevertheless, the traditional ESN is intrinsically a supervised learning technique, which only depends on labeled samples, but omits a large number of unlabeled samples. In order to eliminate this limitation, this work proposes a semi-supervised ESN method assisted by a temporal-spatial graph regularization (TSG-SSESN) for constructing soft sensor model with all the available samples. Firstly, the traditional supervised ESN is enhanced to construct the semi-supervised ESN (SSESN) model by integrating both unlabeled and labeled samples in the reservoir computing procedure. The SSESN computes the reservoir states under high sampling rate for better process dynamic information mining. Furthermore, the SSESN's output optimization objective is modified by applying the local adjacency graph of all training samples as a regularization term. Especially, in view of the dynamic data characteristic, a temporal-spatial graph is constructed by considering both the temporal relationship and the spatial distances. The applications to a debutanizer column process and a wastewater treatment plant demonstrate that the TSG-SSESN model can build much smoother model and has better generalization capability than the basic ESN models in terms of soft sensor prediction results.

Exploration of deep learning models for real-time monitoring of state and performance of anaerobic digestion with online sensors

The immediate response to the state disturbances of anaerobic digestion is essential to prevent anaerobic digestion failure. However, frequent monitoring of the state and performance of anaerobic digestion is challenging. Thus, deep learning models were investigated to predict the state and performance variables from online sensor data. The online sensor data, including pH, electric conductivity, and oxidation-reduction potential, were used as the input features to build deep learning models. The state and performance data measured offline were used as the labels. The model performance was compared for several deep learning models of convolutional neural network (CNN), long short-term memory (LSTM), dense layer, and their combinations. The combined model of CNN and bidirectional LSTM was robust and well-generalized in predicting the state and performance variables (R² = 0.978, root mean square error = 0.031). The combined model is an excellent soft sensor for monitoring the state and performance of anaerobic digestion from electrochemical sensors.

Adaptive soft sensing of river flow prediction for wastewater treatment operation and risk management

Many wastewater utilities have discharge permits directly tied with the receiving river flow, so it is critical to have accurate prediction of the hydraulic throughput to ensure safe operation and environment protection. Current empirical knowledge-based operation faces many challenges, so in this study we developed and assessed daily-adaptive, probabilistic soft sensor prediction models to forecast the next month's average receiving river flowrate and guide the utility operations. By comparing 11 machine-learning methods, extra trees regression exhibits desired deterministic prediction accuracy at day 0 (overall accuracy index: 3.9 × 10^-3 1/cms²) (cms: cubic meter per second), which also increases steadily over the course of the month (e.g., MAPE and RMSE decrease from 41.46% and 23.31 cms to 3.31% and 2.81 cms, respectively). The overall classification accuracy of three river flow classes reaches 0.79 at the beginning and increases to about 0.97 over the course of the predicted month. To manage the uncertainty caused by potential false negative classification as overestimations, a probabilistic assessment on the predictions based on 95% lower PI is developed and successfully reduces the false negative classification from 17% to nearly zero with a slight sacrifice of overall classification accuracy.

Development of a wide-range soft sensor for predicting wastewater BOD5 using an eXtreme gradient boosting (XGBoost) machine

In wastewater monitoring, detecting extremely high pollutant concentrations is necessary to properly calibrate the treatment process. However, existing hardware sensors have a limited linear range which may fail to measure extremely high levels of pollutants; and likewise, the conventional "soft" model sensors are not suitable for the highly-skewed data distributions either. This study developed a new soft sensor by using eXtreme Gradient Boosting (XGBoost) machine learning to 'measure' the wastewater organics (in terms of 5-day biochemical oxygen demand (BOD5)). The soft sensor was tested on influent and effluent BOD5 of two different wastewater treatment plants to validate the results. The model results showed that XGBoost can detect these extreme values better than conventional soft sensors. This new soft sensor can function using a sparse input matrix via XGBoost's sparsity awareness algorithm - which can address the limitation of the conventional soft sensor with the fallibility of supporting hardware sensors even.

Enhanced virtual sample generation based on manifold features: Applications to developing soft sensor using small data

In the process industry, it is essential to establish a data-driven soft sensor to predict the key variable that is difficult to online measure directly. The accuracy performance of data-driven soft sensors relies heavily on data. Unfortunately, it is hard to acquire sufficient and informative data from the samples with limited number, which is called as the small sample problem. For handling the small sample problem, it is a good solution to generating virtual samples according to the distribution of original data. This paper proposes an enhanced method of virtual sample generation utilizing manifold features to develop soft sensors using small data. First, T-Distribution Stochastic Neighbor Embedding (t-SNE) is utilized to extract the features of input data. The main idea of generating virtual samples is to use the interpolation algorithm to obtain virtual t-SNE input features and then the random forest algorithm is utilized to get the virtual outputs using virtual t-SNE input features. Finally, virtual samples using the proposed t-SNE based virtual sample generation (t-SNE-VSG) can be achieved. For the sake of confirming the effectiveness and feasibility of the presented t-SNE-VSG, a standard data set is first used. What is more, a small data set from an actual industrial process of Purified Terephthalic Acid is used to establish a soft sensor model. The results from simulations show that the accuracy performance of the soft sensor established with small data can be effectively improved by adding the virtual samples generated by t-SNE-VSG. In addition, t-SNE-VSG achieves superior accuracy to state-of-the-art virtual sample generation methods.

Soft Sensor Modeling Method Based on Improved KH-RBF Neural Network Bacteria Concentration in Marine Alkaline Protease Fermentation Process

Marine alkaline protease (MAP) fermentation is a complex multivariable, multi-coupled, and nonlinear process. Some unmeasured parameters will affect the quality of protease. Aiming at the problem that some parameters are difficult to be detected online, a soft sensing modeling method based on improved Krill Herd algorithm RBF neural network (LKH-RBFNN) is proposed in this paper. Based on the multi-parameter RBFNN model, the adaptive RBF neural network algorithm and control law are used to approximate the unknown parameters. The adaptive Levy flight strategy is used to improve the traditional Krill Herd algorithm, improve the global search ability of the algorithm, and avoid falling into local optimization. At the same time, the location update formula of Krill Herd algorithm is improved by using the calculation methods of similarity and agglomeration degree, and the parameters of adaptive RBFNN are optimized to improve its over correction and large amount of calculation. Finally, the soft sensing prediction model of bacterial concentration and relative active enzyme in map process based on LKH-RBFNN is established. The root mean square error and maximum absolute error of this model are 0.938 and 0.569, respectively, which are less than KH-RBFNN and PSO-RBFNN prediction models. It proves that the prediction error of LKH-RBFNN model is smaller and can meet the needs of online prediction of key parameters of map fermentation.

Electromechanical tensile test equipment for stretchable conductive materials

The demand for soft and conductive materials has intensified due to the increased interest in soft robotics. Consequently, researchers strive to realize easy, fast, and cost-effective fabrication methods. To evaluate the mechanical properties of materials requires tensile testing. However, the availability of an electromechanical tensile test to assess the quality of the electromechanical properties of stretchable conductive materials has yet to be widely commercialized. This situation has hindered the development of soft and stretchable conductive materials. Here, we develop a customized electromechanical tensile test for soft and stretchable materials. We integrate three standalone devices using Python software and provide a graphic user interface (GUI) for easy operation of the equipment. We expect that our customized electromechanical tensile test will contribute to advances in soft robotics, especially soft and stretchable sensors. Furthermore, our electromechanical setup can aid in the development of laboratory equipment and the understanding of the electromechanical properties of stretchable conductive materials.

Online cement clinker quality monitoring: A soft sensor model based on multivariate time series analysis and CNN

The content of free calcium oxide (f-CaO) in cement clinker is an important index for cement quality. Aiming at the characteristics of strong coupling, time-varying delay and highly non-linearity in cement clinker production, a soft sensor model based on multivariate time series analysis and convolutional neural network (MVTS-CNN) is proposed for the online f-CaO content monitoring. Based on the process industry characteristics, the MVTS-CNN modeling involves the detailed analysis of coupling relationship and time-varying delay in cement production and the application of neural network in multivariate time-series feature extraction. The main researches and contributions are fourfold: First, the strong coupling in the production system is further analyzed, and the proposed model is focused on the data coupling between specific processes, not the control coupling. Second, a multivariate time series analysis method is designed to select the time series that may have direct impacts on the f-CaO content in different production conditions, which is founded on the information on time delay range and longest active duration. Third, a multivariate time series feature extraction method is designed and adopted in the MVTS-CNN model to extract the multivariate time series features, such as active duration difference features, coupling features, nonlinear features and key time series features. Fourth, a new timing matching method, which is combined the rough timing matching of multivariate time series and the detailed timing matching of key features, is proposed to deal with the time-varying delay in various production conditions. Compared with traditional CNN, support vector machines (SVM) and long-short term memory networks (LSTM), the results demonstrate that the MVTS-CNN model has higher accuracy, better generalization ability and superior robustness.

Digital twin of low dosage continuous powder blending - Artificial neural networks and residence time distribution models

In this paper we present a thorough description of the digital twin development for a continuous pharmaceutical powder blending process in accordance with the Process Analytical Technologies (PAT) and Quality by Design (QbD) guidelines. A low-dosage system of caffeine active pharmaceutical ingredient (API) and dextrose excipient was examined via continuous blending experiments. Near infrared (NIR) spectroscopy-based process analytics were applied; quantitative evaluation of spectra was achieved using multivariate data analysis. The blending system was represented with mechanistic residence time distribution (RTD) models and two types of recurrent artificial neural networks (ANN), experimental datasets were used for model training or fitting and validation. Detailed comparison of the two modelling approaches, the optimization of the model-based digital twin, and efficiency of the soft sensor-based process monitoring is presented through several validating simulations. Both RTD models and nonlinear autoregressive neural networks demonstrated excellent predictive power for the low dosage blending process. RTD models can prove to be more advantageous in industrial development as they are less resource-intensive to develop and prediction accuracy on low concentration levels lacks dependency from the precision of chemometric calibration. Reduced material costs and limited development timeframe render the digital twin an efficient tool in technological development.

Soft sensor for real-time estimation of tablet potency in continuous direct compression manufacturing operation

One of the critical quality attributes of the solid oral dosage forms produced in continuous direct compression operations is the tablet potency. A novel soft sensor comprising of a combination of first principle-based and empirical models has been developed to enable real-time monitoring of blend and tablet potency, and concentrations of other excipients at various stream levels along the direct compression line. The soft sensor model has only three adjustable parameters, primarily associated with the equipment design and operation, so the model is product agnostic which is key to enable flexible manufacturing. The estimation accuracy of the soft sensor is demonstrated through a series of real time experiments which include steady state and dynamic transitions of potency during the runs, compared with offline analytically tested tablet cores. The results indicate that the proposed soft sensor can be utilized as a robust tool for real-time monitoring of potency, suggesting an extension of its utilization to higher levels of control. Two potential applications of the soft sensor are: 1. An element of a control strategy for product diversion; 2. A predictive model for advanced process control strategy to minimize the variability in tablet composition.

Soft-sensor development for monitoring the lysine fermentation process

Monitoring cell growth and target production in working fermentors is important for stabilizing high level production. In this study, we developed a novel soft sensor for estimating the concentration of a target product (lysine), substrate (sucrose), and bacterial cell in commercially working fermentors using machine learning combined with available on-line process data. The lysine concentration was accurately estimated in both linear and nonlinear models; however, the nonlinear models were also suitable for estimating the concentrations of sucrose and bacterial cells. Data enhancement by time interpolation improved the model prediction accuracy and eliminated unnecessary fluctuations. Furthermore, the soft sensor developed based on the dataset of the same process parameters in multiple fermentor tanks successfully estimated the fermentation behavior of each tank. Machine learning-based soft sensors may represent a novel monitoring system for digital transformation in the field of biotechnological processes.

A hybrid NIR-soft sensor method for real time in-process control during continuous direct compression manufacturing operations

Near Infrared (NIR) spectroscopy is commonly utilized for continuous manufacturing as Process Analytical Technology (PAT) tool. This paper focus on a continuous direct compression manufacturing process, in which an NIR PAT probe is integrated into the tablet press feed frame and into the tablet diversion control system to ensure continuous monitoring of the potency and homogeneity of the blend within the process line. The quantification of NIR spectra is achieved through Partial Least-Squares (PLS) modeling, calibrated with offline analyzed tablet cores at different potency levels. Because the NIR measurements are often sensitive to sample physical properties caused by raw materials or process conditions, etc., adopting a data-driven approach will require a large amount of representative data throughout the method lifecycle. During the early stages of process development, whenever new uncaptured source of variability in the model space are encountered, the chemometric predictions can deviate from the offline reference, requiring frequent model updates. These deviations can be reduced by integrating process and physico-chemical knowledge in the on-line potency estimation. This paper presents a novel hybrid method combining the online NIR PLS and a potency soft sensor estimation, enabling a robust potency prediction whilst minimizing maintenance downtimes and facilitating cross-site method transfer.

coli concentration based on conventional monitoring parameters for wastewater disinfection control

Real-time acquisition of indicator bacteria concentration at the inlet of disinfection unit is a fundamental support to the control of chemical and ultraviolet wastewater disinfection. Culture-based enumeration methods need time-consuming laboratory analyses, which give results after several hours or days, while newest biosensors rarely provide information about specific strains and outputs are not directly comparable with regulatory limits as a consequence of measurement principles. In this work, a novel soft sensor approach for virtual real-time monitoring of E. coli concentration is proposed. Conventional wastewater physical and chemical indicators (chemical oxygen demand, total nitrogen, nitrate, ammonia, total suspended solids, conductivity, pH, turbidity and absorbance at 254 nm) and flowrate were studied as potential predictors of E. coli concentration relying on data collected from three full-scale wastewater treatment plants. Different methods were compared: (i) linear modeling via ordinary least squares; (ii) ridge regression; (iii) principal component regression and partial least squares; (iv) non-linear modeling through artificial neural networks. Linear soft sensors reached some degree of accuracy, but performances of the artificial neural network based models were by far superior. Sensitivity analysis allowed to prioritize the importance of each predictor and to highlight the site-specific nature of the approach, because of the site-specific nature of relationships between predictors and E. coli concentration. In one case study, pH and conductivity worked as good proxy variables when the occurrence of intense rain events caused sharp increases in E. coli concentration. Differently, in other case studies, chemical oxygen demand, total suspended solids, turbidity and absorbance at 254 nm accounted for the positive correlation between low wastewater quality and E. coli concentration. Moreover, sensitivity analysis of artificial neural network models highlighted the importance of interactions among predictors, contributing to 25 to 30% of the model output variance. This evidence, along with performance results, supported the idea that nonlinear families of models should be preferred in the estimation of E. coli concentration. The artificial neural network based soft sensor deployment for control of peracetic acid disinfectant dosage was simulated over a realistic scenario of wastewater quality recorded by on-line sensors over 2 months. The scenario simulations highlighted the significant benefit of an E. coli soft sensor, which provided up to 57% of disinfectant saving.

Novel manifold learning based virtual sample generation for optimizing soft sensor with small data

Due to the extremely complex mechanism and strong non-linear characteristics of industrial processes, data-driven soft sensor technologies play a key role in the intelligent measurement of process industries. However, the information of the collected process data in the steady stage is quite limited and unreliable, causing the small sample problem. As a result, it becomes an intractable challenge to catch the nature of the process and build accurate soft sensor models. To solve this problem, this paper proposes a novel manifold learning based virtual sample generation method (Isomap-VSG) to generate feasible virtual samples in the information gaps for supplementing the original small sample space. To find data sparse regions reasonably, one kind of manifold learning methods called Isomap is used to visualize process data with high dimension. Then virtual samples can be generated by the interpolation method and extreme learning machine. The simulation results on a standard dataset and a real-world application demonstrate that, compared with other advanced methods, the proposed Isomap-VSG method can achieve better performance in terms of generating feasible virtual samples and improving the accuracy of soft sensor models using limited samples.

Noninvasive online monitoring of Corynebacterium glutamicum fed-batch bioprocesses subject to spent sulfite liquor raw material uncertainty

In this study the use of a particle filter algorithm to monitor Corynebacterium glutamicum fed-batch bioprocesses with uncertain raw material input composition is shown. The designed monitoring system consists of a dynamic model describing biomass growth on spent sulfite liquor. Based on particle filtering, model simulations are aligned with continuously and noninvasively measured carbon evolution and oxygen uptake rates, giving an estimate of the most probable culture state. Applied on two validation experiments, culture states were accurately estimated during batch and fed-batch operations with root mean square errors below 1.1 g L^-1 for biomass, 0.6 g L^-1 for multiple substrate concentrations and 0.01 g g^-1 h^-1 for biomass specific substrate uptake rates. Additionally, upon fed-batch start uncertain feedstock concentrations were corrected by the estimator without the need of any additional measurements. This provides a solid basis towards a more robust operation of bioprocesses utilizing lignocellulosic side streams.

History and Evolution of Modeling in Biotechnology: Modeling & Simulation, Application and Hardware Performance

Biological systems are typically composed of highly interconnected subunits and possess an inherent complexity that make monitoring, control and optimization of a bioprocess a challenging task. Today a toolset of modeling techniques can provide guidance in understanding complexity and in meeting those challenges. Over the last four decades, computational performance increased exponentially. This increase in hardware capacity allowed ever more detailed and computationally intensive models approaching a “one-to-one” representation of the biological reality. Fueled by governmental guidelines like the PAT initiative of the FDA, novel soft sensors and techniques were developed in the past to ensure product quality and provide data in real time. The estimation of current process state and prediction of future process course eventually enabled dynamic process control. In this review, past, present and envisioned future of models in biotechnology are compared and discussed with regard to application in process monitoring, control and optimization. In addition, hardware requirements and availability to fit the needs of increasingly more complex models are summarized. The major techniques and diverse approaches of modeling in industrial biotechnology are compared, and current as well as future trends and perspectives are outlined.

Autoencoder-based nonlinear Bayesian locally weighted regression for soft sensor development

The framework of locally weighted learning (LWL) has established itself as a popular tool for developing nonlinear soft sensors in process industries. For LWL-based soft sensors, the key factor for achieving high performance is to construct accurate localized models. To this end, in this paper a nonlinear local model training algorithm called nonlinear Bayesian weighted regression (NBWR) is proposed. In the NBWR, the nonlinear features of process data are first extracted by the autoencoder; then, given a query sample a local dataset is selected on the feature space and a fully Bayesian regression model with differentiated sample weights is developed. The benefits of this approach, which include better consistency of correlation, stronger abilities to deal with process nonlinearities and uncertainties, overfitting and numerical issues, lead to superior performance. The NBWR is used for developing a soft sensor under the LWL framework, and a real-world industrial process is used to evaluate the performance of the NBWR-based soft sensor. The experimental results demonstrate that the proposed method outperforms several benchmarking soft sensing approaches.

Knowledge based recursive non-linear partial least squares (RNPLS)

Soft sensors driven by data are very common in industrial plants to perform indirect measurements of difficult to measure critical variables by using other variables that are relatively easier to obtain. The use of soft sensors implies some challenges, such as the colinearity of the predictor variables, the time-varying and possible non-linear nature of the industrial process. To deal with the first challenge, the partial least square (PLS) regression has been employed in many applications to model the linear relations between process variables, with noisy and highly correlated data. However, the PLS model needs to deal with the other two issues: the non-linear and time-varying characteristics of the processes. In this work, a new knowledge-based methodology for a recursive non-linear PLS algorithm (RNPLS) is systematized to deal with these issues. Here, the non-linear PLS algorithm is set up by carrying out the PLS regression over the augmented input matrix, which includes knowledge based non-linear transformations of some of the variables. This transformation depends on the system's nature, and takes into account the available knowledge about the process, which is provided by expert knowledge or emulated using software tools. Then, the recursive exponential weighted PLS is used to modify and adapt the model according to the process changes. This RNPLS algorithm has been tested using two case studies according to the available knowledge, a real industrial evaporation station of the sugar industry, where the expert knowledge about the process permits the formulation of the relationships, and a simulated wastewater treatment plant, where the necessary knowledge about the process is obtained by a software tool. The results show that the methodology involving knowledge regarding the process is able to adjust the process changes, providing highly accurate predictions.

A novel support vector machine ensemble model for estimation of free lime content in cement clinkers

Free lime (f-CaO) content is a crucial quality parameter for cement clinkers in rotary cement kiln. Due to lack of hardware sensors, f-CaO content in cement clinker is mostly obtained by offline laboratory measurement, making timely control rather difficult and even impossible. In this work, a soft sensor approach named as support vector machine ensemble (ESVM) model is proposed to estimate f-CaO content. The process data employed to train and test the model were collected from a cement plant in China, covering a time span of about 30 days. The raw data were preprocessed by filters and time-series matching. The processed data were then clustered by fuzzy c-means clustering algorithm to capture process features at different operating conditions. For each individual cluster, a base SVM regressor was trained to estimate f-CaO content. Finally, an ensemble model consisting of four base SVM regressors was established to estimate f-CaO content at multifarious process conditions. The effectiveness of the proposed ESVM model was investigated by comparing it with manual measurements and other models available in literature. The results demonstrate that the proposed ESVM model achieves improvements in model accuracy as well as generalization capability. The proposed ESVM model has a broad application space in cement production process for automatic monitoring of f-CaO content.

Computational method for multiphase flow characterization in the gas refinery

This paper presents a new computational method for the decentralized multiphase flow measurement based on the interconnections between the two subsystems to precisely estimate the states of the multiphase flow at the gas refinery. The states of the condensate and gas sub-systems were separately estimated using the Differential Mean Value Theorem by considering the relationship between two subsystems, designing an observer and converting the conditions to linear matrix inequality. To check the stability and performance of the system against the changes, the Lyapunov theory has been used. The states behavior investigated with and without disturbance in the system output and dynamics. Additionally, the Unscented Kalman Filter based on the simplified drift flux model was used to estimate the states. It is found that both observers are capable to identify the states with some differences in performance and drift flux model is sufficient for estimation of parameters and states.

Beyond signal quality: The value of unmaintained pH, dissolved oxygen, and oxidation-reduction potential sensors for remote performance monitoring of on-site sequencing batch reactors

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.

Online prediction of effluent COD in the anaerobic wastewater treatment system based on PCA-LSSVM algorithm

Since anaerobic wastewater treatment is a nonlinear and complex biochemical process, reasonable monitoring and control are needed to keep it operating stably and efficiently. In this paper, a least-square support-vector machine (LS-SVM) was employed to construct models for the prediction of effluent chemical oxygen demand (COD) in an anaerobic wastewater treatment system. The result revealed that the performance of the steady-state model based on LS-SVM for predicting effluent COD was acceptable, with the maximum relative error (RE) of 11.45%, the mean average percentage error (MAPE) of 0.79% and the root mean square error (RMSE) of 3.08 when training, and the performance fell slightly when testing. Even though, the correlation coefficient value (R) between the predicted value and the actual value of 0.9752 could be achieved, which means this model can predict the variation of effluent COD in general. The dynamic-state models under three kinds of shock loads, which were concentration, hydraulic, and bicarbonate buffer absent, showed good forecasting performance, the correlation coefficient values (R) all excelled 0.99. Among these three shocks, the dynamic LS-SVM model under bicarbonate buffer absent shock achieved the optimal performance and followed by the dynamic-state model under hydraulic shock. This paper provides a meaningful reference to improve the monitoring level of the anaerobic wastewater treatment process.

Predicting influent biochemical oxygen demand: Balancing energy demand and risk management

Ready access to comprehensive influent information can help water reclamation plant (WRP) operators implement better real-time process controls, provide operational reliability and reduce energy consumption. The five-day biochemical oxygen demand (BOD₅), a critical parameter for WRP process control, is expensive and difficult to measure using hard-sensors. An alternative approach based on a soft-sensor methodology shows promise, but can be problematic when used to predict high BOD₅ values. Underestimating high BOD₅ concentrations for process control could result in an insufficient amount of aeration, increasing the risk of an effluent violation. To address this issue, we tested a hierarchical hybrid soft-sensor approach involving multiple linear regression, artificial neural networks (ANN), and compromise programming. While this hybrid approach results in a slight decrease in overall prediction accuracy relative to the approach based on ANN only, the underestimation percentage is substantially lower (37% vs. 61%) for predictions of carbonaceous BOD₅ (CBOD₅) concentrations higher than the long-term average value. The hybrid approach is also flexible and can be adjusted depending on the relative importance between energy savings and managing the risk of an effluent violation.

Soft sensor for monitoring biomass subpopulations in mammalian cell culture processes

OBJECTIVES

Biomass subpopulations in mammalian cell culture processes cause impurities and influence productivity, which requires this critical process parameter to be monitored in real-time.

RESULTS

For this reason, a novel soft sensor concept for estimating viable, dead and lysed cell concentration was developed, based on the robust and cheap in situ measurements of permittivity and turbidity in combination with a simple model. It could be shown that the turbidity measurements contain information about all investigated biomass subpopulations. The novelty of the developed soft sensor is the real-time estimation of lysed cell concentration, which is directly correlated to process-related impurities such as DNA and host cell protein in the supernatant. Based on data generated by two fed-batch processes the developed soft sensor is described and discussed.

CONCLUSIONS

The presented soft sensor concept provides a tool for viable, dead and lysed cell concentration estimation in real-time with adequate accuracy and enables further applications with respect to process optimization and control.

On-line identification of fermentation processes for ethanol production

A strategy for monitoring fermentation processes, specifically, simultaneous saccharification and fermentation (SSF) of corn mash, was developed. The strategy covered the development and use of first principles, semimechanistic and unstructured process model based on major kinetic phenomena, along with mass and energy balances. The model was then used as a reference model within an identification procedure capable of running on-line. The on-line identification procedure consists on updating the reference model through the estimation of corrective parameters for certain reaction rates using the most recent process measurements. The strategy makes use of standard laboratory measurements for sugars quantification and in situ temperature and liquid level data. The model, along with the on-line identification procedure, has been tested against real industrial data and have been able to accurately predict the main variables of operational interest, i.e., state variables and its dynamics, and key process indicators. The results demonstrate that the strategy is capable of monitoring, in real time, this complex industrial biomass fermentation. This new tool provides a great support for decision-making and opens a new range of opportunities for industrial optimization.

A data-driven soft sensor for needle deflection in heterogeneous tissue using just-in-time modelling

Global modelling has traditionally been the approach taken to estimate needle deflection in soft tissue. In this paper, we propose a new method based on local data-driven modelling of needle deflection. External measurement of needle-tissue interactions is collected from several insertions in ex vivo tissue to form a cloud of data. Inputs to the system are the needle insertion depth, axial rotations, and the forces and torques measured at the needle base by a force sensor. When a new insertion is performed, the just-in-time learning method estimates the model outputs given the current inputs to the needle-tissue system and the historical database. The query is compared to every observation in the database and is given weights according to some similarity criteria. Only a subset of historical data that is most relevant to the query is selected and a local linear model is fit to the selected points to estimate the query output. The model outputs the 3D deflection of the needle tip and the needle insertion force. The proposed approach is validated in ex vivo multilayered biological tissue in different needle insertion scenarios. Experimental results in five different case studies indicate an accuracy in predicting needle deflection of 0.81 and 1.24 mm in the horizontal and vertical lanes, respectively, and an accuracy of 0.5 N in predicting the needle insertion force over 216 needle insertions.

Modeling and simulation of soft sensor design for real-time speed and position estimation of PMSM

This paper deals with the design of a speed soft sensor for permanent magnet synchronous motor. At high speed, model-based soft sensor is used and it gives excellent results. However, it fails to deliver satisfactory performance at zero or very low speed. High-frequency soft sensor is used at low speed. We suggest to use a model-based soft sensor together with the high-frequency soft sensor to overcome the limitations of the first one at low speed range.

Online monitoring and control of particle size in the grinding process using least square support vector regression and resilient back propagation neural network

Particle size soft sensing in cement mills will be largely helpful in maintaining desired cement fineness or Blaine. Despite the growing use of vertical roller mills (VRM) for clinker grinding, very few research work is available on VRM modeling. This article reports the design of three types of feed forward neural network models and least square support vector regression (LS-SVR) model of a VRM for online monitoring of cement fineness based on mill data collected from a cement plant. In the data pre-processing step, a comparative study of the various outlier detection algorithms has been performed. Subsequently, for model development, the advantage of algorithm based data splitting over random selection is presented. The training data set obtained by use of Kennard-Stone maximal intra distance criterion (CADEX algorithm) was used for development of LS-SVR, back propagation neural network, radial basis function neural network and generalized regression neural network models. Simulation results show that resilient back propagation model performs better than RBF network, regression network and LS-SVR model. Model implementation has been done in SIMULINK platform showing the online detection of abnormal data and real time estimation of cement Blaine from the knowledge of the input variables. Finally, closed loop study shows how the model can be effectively utilized for maintaining cement fineness at desired value.

Hybrid intelligent control of substrate feeding for industrial fed-batch chlortetracycline fermentation process

The lack of accurate process models and reliable online sensors for substrate measurements poses significant challenges for controlling substrate feeding accurately, automatically and optimally in fed-batch fermentation industries. It is still a common practice to regulate the feeding rate based upon manual operations. To address this issue, a hybrid intelligent control method is proposed to enable automatic substrate feeding. The resulting control system consists of three modules: a presetting module for providing initial set-points; a predictive module for estimating substrate concentration online based on a new time interval-varying soft sensing algorithm; and a feedback compensator using expert rules. The effectiveness of the proposed approach is demonstrated through its successful applications to the industrial fed-batch chlortetracycline fermentation process.