A calorimetric soft sensor for monitoring viscosity and estimating end-product properties in suspension polymerization

Melt viscosity is regarded as a key quality indicator of the polymer melt in polymer extrusion processes. However, limitations such as disturbances to the melt flow and measurement delays of the existing in-line and side-stream rheometers prevent the monitoring and controlling of this key parameter in real time. Soft sensors can be employed to monitor physical parameters that are difficult to measure using hardware sensing instruments. This study presents a grey-box soft sensing solution to predict the melt viscosity in real time, which combines physics-based knowledge with machine learning. A fine-tuned physics-based mathematical model is used to make melt viscosity predictions, and a deep neural network is employed to compensate for its prediction errors. The proposed soft sensor model reported a normalised root mean square error of 2.210−3 (0.22%), outperforming fully data-driven soft sensor models based on multilayer perceptron and long short-term memory neural networks. Furthermore, it exhibited an improvement of approximately 95% in terms of predictive performance, compared to a soft sensor based on a radial basis function neural network reported in a previous study. The proposed soft sensor can monitor viscosity changes caused by changes in operating conditions but not suitable for detecting viscosity changes due to changes in material properties. The findings of this study can aid in enhancing process monitoring and control in polymer extrusion processes.

This work introduces an approach to enhance the real-time monitoring of batch suspension polymerization processes through the integration of real-time calorimetry and state estimation techniques. The challenges inherent in monitoring dynamic and complex processes, particularly in the absence of direct measurements, are effectively addressed through the proposed approach. A dynamic model equation specific to batch operation is employed to assess conversion throughout the process, with timely updates of dynamic parameters in the model equation. A nonlinear high gain cascaded observer with calorimetric measurements estimates the overall heat transfer coefficient and reaction heat. Variations in other parameters such as heat capacity and density are derived from estimated conversion data and updated in the model equation. The study also accounts for the heat loss to the surroundings during isoperibolic batch processes. Validation of the proposed soft sensor model is carried out by comparing the estimated and experimental conversion data by conducting MMA suspension polymerization in a reaction calorimeter. Results demonstrate the potential of calorimetric state estimation as an alternative to direct conversion measurements, offering comprehensive and enhanced monitoring of the batch suspension polymerization process.

Advanced Process Control (APC) systems depend on inferential modeling to predict product quality in near real time, particularly in scenarios where analyzer data is unavailable or lab test results are delayed. Historically, such inferential models have relied on linear regression, which often fails to capture the dynamic and non-linear behavior typical of modern refining processes. As refining operations grow more non-linear and responsive, such conventional models often fall short in providing accurate, timely estimates—leading to suboptimal control and operational inefficiencies. To overcome these constraints, Hindustan Petroleum Corporation Limited (HPCL) developed a refinery-wide framework for machine learning (ML)-based soft sensors, aimed at replacing or augmenting traditional APC inferential. These soft sensors were built using supervised ML models trained on historical DCS and laboratory data, carefully validated to ensure first-principle consistency and explainability. The models predict critical quality parameters in near real time using streaming process data, ensuring high availability of quality estimates—even during analyzer downtimes or lab result delays. Over 40 soft sensors were developed and deployed across major refinery units such as CDU, VDU, FCC, CCR, DHT, SRU etc. Importantly, a subset of these models was integrated into closed-loop APC systems, where soft sensor outputs now serve as active feedback for control decisions—demonstrating their stability, accuracy, and reliability in live operations. The development and deployment lifecycle of these models was enabled by HPINSYT—HPCL’s in-house data analytics platform—which provided a no-code interface for model building, evaluation, deployment, and simulation. This enabled close collaboration between domain experts and data scientists, dramatically accelerating development cycles and improving adoption on the plant floor. This paper presents the methodology, model architecture, APC integration strategy, and field results from HPCL’s ML-based soft sensor deployments, highlighting their role in elevating APC systems beyond linear limitations.

Electronic skin (e‐skin) is an emerging technology with promising applications in various fields, including human–machine interfaces, prosthetics, and robotics. Soft and flexible sensors are vital components for the e‐skin that can mimic human skin's sensing capabilities. Among soft sensors, liquid‐metal‐based sensors have gained attention owing to their unique properties, such as high electrical conductivity, stretchability, and elasticity. Herein, a novel approach is presented that enables multidirectional pressure sensing with a machine‐learning approach from the transient response of the liquid‐metal‐based soft pressure sensor for the e‐skins. In this study, a soft sensor is developed that utilizes liquid metal and has an array of microchannels on a dome‐shaped structure to detect pressures from multiple directions. The transient response from six microchannels of the sensor is used as the input for a convolutional neural network (CNN) to predict the direction (classification accuracy of 99.1%) and magnitude (regression error of 20.13%) of the applied pressures in real time. Finally, a potential application of the developed liquid‐metal‐based soft sensor as a human–machine interface device is demonstrated by using it to control an RC model car through multidirectional predictions (pressure direction and magnitude) through machine learning in real time.

Real-time toxicity detection and monitoring using a microbial fuel cell (MFC) is often based on observing current or voltage changes. Other methods of obtaining more information on the internal state of the MFC, such as electrochemical impedance spectroscopy (EIS), are invasive, disruptive, time consuming, and may affect long-term MFC performance. This study proposes a soft sensor approach as a non-invasive real-time method for evaluating the internal state of an MFC biosensor during toxicity monitoring. The proposed soft sensor approach is based on estimating the equivalent circuit model (ECM) parameters in real time. A flow-through MFC biosensor was operated at several combinations of carbon source (acetate) and toxicant (copper) concentrations. The ECM parameters, such as internal resistance, capacitance, and open-circuit voltage, were estimated in real time using a numerical parameter estimation procedure. The soft sensor approach proved to be an adequate replacement for EIS measurements in quantifying changes in the biosensor internal parameters. The approach also provided additional information, which could lead to earlier detection of the toxicity onset.

This article deals with the determination of kinetic and thermodynamic data of free-radical polymerization by adiabatic reaction calorimetry. The polymerization of methyl methacrylate in solution, suspension, and emulsion were chosen as systems to be studied. From the measured temperature-time courses of the polymerizations the overall rate constants can be determined with and without gel effect. With knowledge of an appropriate mathematical model describing the kinetics of reaction it was also possible to estimate elementary reaction constants if the molecular weight distribution of the polymer formed was considered as well. The temperature rise and the self-heating rate can be modeled very well for polymerization in solution over the entire range of concentration and, in the case of polymerization in suspension and emulsion, up to a volume fraction of monomer of 20%. The modeling of molecular weight distribution of polymers produced by polymerization in solution and suspension is satisfactory. For emulsion polymerization, however, only the order of magnitude of the average molecular weight could be calculated with the model used. The average particle size of the polymer latex formed could be calculated rather well. © 1997 John Wiley & Sons, Inc. J Appl Polym Sci 66: 673–681, 1997

Composite particles constituted of fine AlN powder and polystyrene as the base polymer were prepared by suspension polymerization. To change the viscosity of the dispersed phase at the beginning of suspension polymerization, preliminary bulk polymerization of styrene containing AlN powder was carried out prior to suspension polymerization. Interfacial polycondensation was also carried out to form the polyamide film on the surface of the dispersed phase droplet. The AlN content of the composite particles increased with increase in the viscosity of the dispersed phase due to preliminary bulk polymerization and with decrease in the impeller speed during suspension polymerization. Formation of the polyamide film was found to effectively prevent AlN powder from escaping to the continuous phase, and this led to the increase in the AlN content. The dispersion state of AlN powder in the base polymer was found to be controlled by the preparation conditions.

This study investigated the usefulness of measurements from an agitator torque sensor in monitoring the dynamics of suspension polymerization. The main focus was to estimate the viscosity of the reaction mass during polymerization using the agitator torque as a secondary variable. Viscosity is a crucial parameter that plays a vital role in determining the efficiency of the process and the quality of the final product. Accurate viscosity monitoring is essential as it provides valuable insights into the progression of the polymerization process and its dynamic behaviour. This study developed a combined Kalman filter (KF) and fuzzy logic (FL) model to estimate viscosity in real time, addressing the challenges of noise in torque measurements. Experimental validation showed that the KF‐fuzzy model improved the accuracy and stability of viscosity predictions, particularly during the critical stages of polymerization. This approach enables better monitoring of reaction dynamics, thereby supporting process optimization and control.

Artificial Intelligence (AI) is relatively new neurocomputational science employed in oil industry to solve wide spectrum of non-linear problems with high parallelism, fault and noise tolerance. AI seems very attractive for its remarkable capabilities of processing-correlating data and learning attributes. This paper briefs a successful implementation of an ecosystem between AI and physics-based models in a smart field, that support accurate well allocation process during the project start-up when testing facilities were not available. This case study is presented in a smart field during the commissioning of a new phase-project incorporating 45 new strings at a time. Reconciliation between wells and fiscal meters vanished after incorporating the newly commissioned wells with no flow-tests. An innovative AI-model was developed to estimate well rates from real-time surface parameters (pressure, temperature, choke &gas lift rate) to assist the allocation process for those strings, utilizing extensive well test history and subsurface data from pre-existing nearby wells. Data mining and physics-based models were integrated to develop an ecosystem that can virtually measure daily oil, water and gas rates with reasonable accuracy enhancing well back-allocation process. Initial newly-strings models were built using only one commissioning test, consisting of diverting wells to a portable separator for few hours. This practice usually overestimates the productivity index since rates are captured in transient mode. Later, when wells are flowing for longer and cleaned up, production rates decrease after stabilization. Fortunately, these changes of performance over time are well-captured by the real time parameters. Therefore, this paper proposal can be extended to other digital fields to track the well performance and minimize the error in back allocated rates as it was observed in this field application, demonstrating how reconciliation factor was continuously enhanced day after day, once AI-virtual rate prediction was introduced. AI-model prediction was later verified against actual flow tests via portable separators with an average error below 20%. Virtual rates were predicted using machine learning (ML) techniques multilinear regression, Artificial Neural-Network (ANN) and Self-Organizing Maps (SOM). The results were used to update models for better well performance prediction. The performance of these ML techniques were improved with intelligent inputs and proper segregated and intuitive training to machine. As a practical application, an ecosystem was developed combining AI and physics to predict well-performance of newly-drill wells under gas lift and natural flow, using real-time measurement and the digital framework, to provides reasonably accurate in simulating both training and test time-dependent well performance inferred from pressure, temperature, gas-lift rate and choke. The novelty of this paper consists in the developed data cleansing and association technics that enable the implementation of successful AI-models in wells with no flow test data for training, levering Real-Time data usage.

Evolution of Particle Size Distribution During Suspension Polymerization: Theory and Experiments

A new parametric wake vortex transport and decay model is proposed that predicts probabilistic wake vortex behavior as a function of aircraft and environmental parameters in real time. The probabilistic two-phase wake vortex decay model (P2P) accounts for the effects of wind, turbulence, stable stratie cation, and ground proximity. The model equations are derived from the analytical solution of the spatiotemporal circulation evolution of the decaying potential vortex and are adapted to wake vortex behavior as observed in large-eddy simulations. Vortex decay progresses in two phases, a diffusion phase followed by rapid decay. Vortex descent is a nonlinear function of vortex strength. Probabilistic components account for deviations from deterministic vortex behavior inherently caused by the stochastic nature of turbulence, vortex instabilities, and deformations, as well as uncertainties and e uctuations that arise from environmental and aircraft parameters. The output of P2P consists of cone dence intervals for vortex position and strength. To assign a dee ned degree of probability to the predictions reliably, the model design allows for the continuous adjustment of decay parameters and uncertainty allowances, based on a growing amount of data. The application of a deterministic version of P2P to the Memphis wake vortex database yields favorable agreement with measurements.

Conventional sequences are static in nature, fixing measurement parameters in advance in anticipation of a wide range of expected tissue parameter values. We set out to design and benchmark a new, personalized approach-termed adaptive MR-in which incoming subject data is used to update and fine-tune the pulse sequence parameters in real time. We implemented an adaptive, real-time multi-echo (MTE) experiment for estimating T2 s. Our approach combined a Bayesian framework with model-based reconstruction. It maintained and continuously updated a prior distribution of the desired tissue parameters, including T2 , which was used to guide the selection of sequence parameters in real time. Computer simulations predicted accelerations between 1.7- and 3.3-fold for adaptive multi-echo sequences relative to static ones. These predictions were corroborated in phantom experiments. In healthy volunteers, our adaptive framework accelerated the measurement of T2 for n-acetyl-aspartate by a factor of 2.5. Adaptive pulse sequences that alter their excitations in real time could provide substantial reductions in acquisition times. Given the generality of our proposed framework, our results motivate further research into other adaptive model-based approaches to MRI and MRS.

Soft artificial skin sensors that can detect contact forces as well as their locations are attractive in various soft robotics applications. However, soft sensors made of polymer materials have inherent limitations of hysteresis and nonlinearity in response, which makes it highly difficult to implement traditional calibration techniques and yields poor estimation performance. In this paper, we propose intelligent algorithms based on machine learning and logics that can improve the performance of soft sensors. The proposed methods in this paper could be solutions to the aforementioned long-standing problems. They can also be used to simplify the system complexity by reducing the number of signal wires. Three machine learning techniques are discussed in this paper: an artificial neural network (ANN), the k-nearest neighbors (k-NN) algorithm, and a recurrent neural network (RNN). The Preisach model of hysteresis and simple logics were used to support these algorithms. We proved that classifying contact locations on a soft sensor is possible using simple algorithms in real time. Also, force estimation of a single contact was possible using an ANN with the Preisach method. Finally, we successfully estimated forces of multiple contact locations by predicting the outputs of mixed RNN results.

Commercial analyzers for measuring the aromatics in the Claus furnace exit gas are currently not available and this leads to sub-optimal energy efficiency and poses asset integrity concerns. To address this problem a high-fidelity model is developed to function as a real time analyzer. Objective of this work is to incorporate the soft sensor in the architecture of Real Time Optimizer (RTO) to monitor the presence of aromatics in the Claus furnace exit stream. The soft sensor is incorporated in the RTO server which provides the access to the plant operating data and the DCS (Distributed Control System). Soft sensor function in the RTO involves the following steps: Soft sensor accesses the plant data and collects the needful input data for simulation Simulation software available in the RTO executes the softs sensor model simulation and generates the aromatics composition data Aromatics composition data is written to the DCS interface as a soft measurement Operators monitor the aromatic composition and accordingly adjust the fuel gas firing Aromatic soft sensor is developed as a kinetic model, which is function of rate parameters of several key reactions of the Claus furnace. The kinetic model of the Claus furnace is incorporated in a process simulation model and catalytic convertors are simulated too. Model is validated with large plant data to show that model predicts furnace temperature within +/- 5% error and aromatics composition within +/- 5 ppm. Simulation analysis shows that the furnace temperature can be decreased by at least 5 °C while ensuring no BTEX slip. Such change in furnace temperature leads to a reduction in fuel gas flow by ~200 Nm3/h, which translates to a monetary benefit of 0.5 million $/yr. Deployment of the soft sensor is currently in progress through engagement with RTO licensor. To the best knowledge of authors, currently, there is no simulator in the market which can adequately model aromatics oxidation phenomena and predict the aromatic content in the furnace exit. This soft sensor being deployed is novel and first of its kind and expected to achieve a sustainable energy efficiency.

Monitoring both near and far field laser beam parameters is extremely helpful in understanding the quality of a laser process. The measurement of such parameters has not been practical to date due to the required disruption of the process beam in order to make the measurement. A significant quality control enhancement could be realized if one could monitor both near and far field patterns of the laser system during the process if it could be done with minimal loss or alteration of the process beam. A novel optical design is discussed that integrated into a laser process head with minimal power loss and disruption to the Laser beam. This in-line monitoring system provides focal spot size, Rayleigh length, focal position and M- squared values as well as all the other ISO beam profiling parameters in real time. An in-line laser beam monitoring system makes possible a higher level of quality control and reduced scrap thereby providing a higher level of reliability to laser processing.