Search

Deep learning models possess limited flexibility in computational burden in model adaptation owing to the conventional use of backpropagation for model training. To address this problem, we propose an alternate training methodology inspired by the forward–forward algorithm originally designed for classification tasks. We extend this concept through a kernel-based modification, enabling its application to regression tasks, which are commonly encountered in process system modeling. Our proposed Kernel-based Forward Propagating Neural Network (K-FP-NN) eliminates backpropagation, using layer-wise updates for better adaptability. We introduce a real-time (RT) updating framework, RT-K-FP-NN, to continuously refine model parameters with new data. Results indicate that when applied to model predictive control of a continuous stirred tank reactor (CSTR) system, our approach updates the model within 100 s, achieving better performance metrics compared to backpropagation-based real-time models, which require 326 s. This framework can be applied to various dynamic systems, enhancing real-time decision-making by improving predictive accuracy and system adaptability.

The present work aims at studying five Indian coals and their solvent extracted clean coal products using Py-GCMS analysis and correlating the characterization data using theoretical principal component analysis. The pyrolysis products of the original coals and the super clean coals were classified as mono-, di- and tri-aromatics, while other prominent products that were obtained included cycloalkanes, n-alkanes, and alkenes ranging from C10–C29. The principal component analysis is a dimensionality reduction technique that reduced the number of input variables in the characterization dataset and gave inferences on the relative composition of constituent compounds and functional groups and structural insights based on scores and loading plots which were consistent with the experimental observations. ATR-FTIR studies confirmed the reduced concentration of ash in the super clean coals and the presence of aromatics. The Py-GCMS data and the ATR-FTIR spectra led to the conclusion that the super clean coals behaved similarly for both coking and non-coking coals with high aromatic concentrations as compared to the raw coal. Neyveli lignite super clean coal was found to show some structural similarity with the original coals, whereas the other super clean coals showed structural similarity within themselves but not with their original coal samples confirming the selective action of the e,N solvent in solubilizing the polycondensed aromatic structures in the coal samples.

Begomoviruses have become the most potential plant pathogens and endanger livelihood of billions of farmers worldwide. Tomato leaf curl New Delhi virus(ToLCNDV) and Tomato leaf curl Gujarat virus(ToLCGV) are two destructive tomato-infecting begomoviruses that can infect diverse plant species including tobacco. Infection due to these begomoviruses induces a range of characteristics symptoms such as curling of leaves, puckering, leaf yellowing, and stunting of plants. Early detection of disease is desirable to contain the spread of the pathogens to manage the disease. Recent developments in the field of artificial intelligence and machine learning have resulted in various tools that enable in the accurate detection objects. In this study, inspiring from these, we have designed a deep learning model to help in early detection of the begemoviruses diseases. Specifically, we employ a deep learning architecture based on a convolutional neural network known as Visual Geometry Group 16 or VGG-16 to detect begomovirus infections in tobacco plants inoculated with ToLCNDV and ToLCGV. The outcome has been quite promising, as the modified model can detect symptomatic leaf curl plants with an accuracy of 97.211%.

A novel dynamic just-in-time (JIT) learning framework is proposed in this paper for the data driven modeling of batch process. In the proposed JIT framework, we employ a searching strategy based on “profile similarity” which takes into account the dynamicity of batch process instead of “sample similarity” measures as reported in previous literature. This is achieved by a “modified edit distance time warping” framework that is proposed in this work. In addition, a new weighting strategy that assigns space, time, and batch weights to data points is introduced to capture the complete dynamics of the batch process. Furthermore, the proposed method can detect and accommodate outliers in the historical data sets. To test the efficacy, we have validated the proposed approach with various simulation studies, namely, (i) a numerical example, (ii) the batch polymerization process, and (iii) the batch transesterification process.

This paper presents a formulation of Stochastic Model Predictive Control (SMPC) for control of batch processes modeled by nonlinear Differential algebraic Equations (DAEs). In presence of additive stochastic disturbances, states become stochastic variables and measurements are also corrupted with noise. Hence, to obtain refined state estimates we employ DAE-Extended Kalman Filter (EKF). Linear prediction model obtained by successive linearization is used in the SMPC formulation. Our proposed SMPC solves a chance constrained optimization problem in receding horizon window to calculate future control moves with pre-specified degree of constraint violation for trajectory tracking. Further, the probabilistic constraints are converted to hard constraints by accommodating necessary back-off. We present a simulation case study of batch transesterification reactor modeled by non-linear DAEs to validate the proposed SMPC framework.

Control of batch polymerization has been a challenging task. In this work, we have tried to use Reinforcement Learning (RL), and Deep Reinforcement Learning (DRL) based control on addressing the existing challenges. RL is a class of machine learning wherein an agent directly interacts with the environment and learns from its experience. The RL consist of an agent who takes an action, and the action changes the state of the environment. Based on old and new state agent gets a reward, which is reinforcement for its future actions. In this work, we have implemented RL and DRL based control for batch polymerization of Polymethyl methacrylate (PMMA). In both the controllers, the input variable considered was jacket temperature, while the reactor temperature was the output variable. Both the controllers have been found to achieve the given setpoint, while the DRL controller being faster than the RL controllers. Further, RL and DRL control with risk sensitivity were also carried out to accommodate the process constraints.

Designing new glasses requires a prioriknowledge of how the composition of a glass dictates its properties such as stiffness, density, or processability. Thus, accelerated design of glasses for targeted applications remain impeded due to the lack of composition–property models. To this extent, exploiting a large dataset of glasses comprising of up to 37 oxide components and more than 100 000 glass compositions, we develop high-fidelity deep neural networks for the prediction of eight properties that enable the design of glasses, namely, density, Young's modulus, shear modulus, hardness, glass transition temperature, thermal expansion coefficient, liquidus temperature, and refractive index. We demonstrate that the models developed here exhibit excellent predictability, ensuring its ability to capture the underlying nonlinear composition–property relationships. Using these models, we develop a series of new design charts, termed as glass selection charts. These charts enable the rational design of functional glasses for targeted applications by identifying unique compositions that satisfy two or more constraints, on both compositions and properties, simultaneously.

Developing a reinforcement learning (RL) framework that works satisfactorily in deterministic and stochastic environments is challenging. To address this problem, a twin agent RL framework is proposed in this work, wherein we amalgamate both stochastic and deterministic agents’ actions in a multiagent framework that works with a feedback mechanism that actively monitors the output. The proposed algorithm uses twin actor networks of different agents, corresponding to deterministic and stochastic agents, and an action selection critic network is used to choose the best action from both agents. Here, the algorithm blends the outcomes of two reinforcement learning (RL) agents, a stochastic agent and a deterministic agent, namely, Proximal Policy Optimization (PPO) and Twin Delayed Deep Deterministic Policy Gradient (TD3), respectively. We assess the effectiveness of the proposed algorithm by applying it to two case studies: (i) monoclonal antibody (mAb) production and (ii) production of propylene glycol (PG). The studies are conducted in the presence of parametric uncertainties, measurement noise, and nominal conditions. It is observed that for case study 1, the root-mean-square error (RMSE) value for the proposed algorithm is reduced by 40.9% when compared with TD3 and 27.57% when compared with PPO for the simulations. Similarly, for case study 2, the RMSE for the proposed algorithm is reduced by 8.87% when compared with TD3 and 5.8% with PPO. Based on extensive simulations, it is found that the proposed twin agent algorithm has faster convergence and better set-point tracking when compared to the agents operated individually.

Flare events in industrial processes are undesirable because of their negative economic and environmental impacts. Early warnings of flare events are needed in preventing occurrences of flaring, as corrective manual interventions can be performed by the plant operators. This article reports the successful application of a hierarchically distributed monitoring approach for flare event prediction. The industrial process considered here is a refinery with a flare gas recovery system installed, where flare events need to be monitored in real time. We use various data analysis tools, including principal component analysis, slow feature analysis, and wavelet transform, under a distributed process monitoring framework to achieve monitoring solutions. Results indicate that the application of the proposed hierarchically distributed monitoring framework is able to predict a majority of the flare events, with an acceptable false alarm rate for the data set we studied.

Two different types of measurements are often available for the key quality variables in process industries - (a) an accurate “slow-rate” laboratory measurements, and (b) a less accurate “fast-rate” online analyser measurements. Also, the analyser measurements are prone to fail due to hardware issues. Therefore, the main objective of this work is to present a novel approach for developing an accurate, fast-rate, inferential model of quality variables which is robust to outliers. For this purpose, we present a maximum likelihood based approach to integrate the multi-rate output data in the model building task, using Expectation Maximization algorithm. The efficacy of the proposed approach is demonstrated using a simulation example.

Outliers are one of the common factors that affect the quality of routine operational data. In this work, we propose a robust identification approach for the Switched Markov Autoregressive eXogeneous (SMARX) system to deal with outliers. The robust identification problem is formulated by imposing Student’s t-distribution to the noise model. The Expectation-Maximization algorithm is adopted to estimate the parameters of both the continuous dynamics described by local ARX models and discrete dynamics described by Hidden Markov Model. The advantages of the proposed approach are demonstrated through a numerical simulation.

The relatively weak tropical storms, known as low‐pressure systems (LPSs), contribute as much as 60% of the seasonal precipitation over the hugely populated central India. More than a third of LPSs are formed by the downstream amplification of the westward propagating disturbances from the Pacific. Here, we show that the downstream LPS genesis is associated with the tropical cyclones (TCs) in the Western North Pacific (WNP). Four major clusters of landfalling TCs that have a relatively lesser degree of recurvature account for about 83% of the downstream LPS genesis. Causality in the fluctuations of the sea level pressure over the Bay of Bengal (BoB) prior to the initiation of LPSs is attributed to the Rossby wave activity over WNP through transfer entropy analysis. Our results suggest a potential for the prediction of the downstream synoptic activity over the BoB at least seven days ahead. The monsoon low‐pressure systems (LPSs) are cyclonic vortices of diameter 1,000–2,000 km that are predominantly present over the Bay of Bengal (BoB) during the summer monsoon. These systems are responsible for more than half of the summer monsoon rainfall over the highly populated central India and Gangetic plains. The genesis mechanisms of monsoon LPSs are not fully understood but can be broadly classified into two types ‐ local processes (in situ) and remote forcing by the westward propagating atmospheric disturbances from the Pacific (downstream amplification). In this study, we show that the Rossby waves from four clusters of Western North Pacific (WNP) tropical cyclones might be responsible for triggering most of the downstream genesis of the synoptic‐scale storms over BoB during the summer monsoon season. Our results suggest that the downstream storm development over the BoB can be reliably predicted, possibly using deep learning models, by considering predictors from WNP. A causal relationship is found between Rossby waves over the Western North Pacific (WNP) and sea‐level pressure variation over the Bay of Bengal (BoB)Landfalling tropical cyclones over the South China Sea are associated with low‐pressure systems'(LPSs) genesis over the BoB83% of downstream LPSs over the BoB are linked to four tropical cyclone clusters over WNP A causal relationship is found between Rossby waves over the Western North Pacific (WNP) and sea‐level pressure variation over the Bay of Bengal (BoB) Landfalling tropical cyclones over the South China Sea are associated with low‐pressure systems'(LPSs) genesis over the BoB 83% of downstream LPSs over the BoB are linked to four tropical cyclone clusters over WNP

Fault or anomaly detection is one of the key problems faced by the chemical process industry for achieving safe and reliable operation. In this study, a novel methodology, spectral weighted graph autoencoder (SWGAE) is proposed, wherein, the problem of anomaly detection is addressed with the help of graphs. The proposed approach entails the following key steps. Firstly, constructing a spectral weighted graph, where each time step of a process variable in the multivariate time series dataset is modelled as a node in an appropriately tuned moving window. Subsequently, we propose to monitor the weights of the edges between two nodes that make a connection. The faulty instances are identified based on the discrepancy in the weight pattern compared to normal operating data. To this end, once the weights are determined, they are fed to the auto-encoder network, where reconstruction loss is calculated, and faults are identified if the reconstruction loss exceeds a threshold. Further, to make the proposed approach comprehensive, a fault isolation methodology is also proposed to identify the faulty nodes once the faulty variables are identified. The proposed approach is validated using Tennessee-Eastman benchmark data and pressurized heavy water nuclear reactor real-time plant data. The results indicate that the SWGAE method, when compared to the other state-of-the-art methods, yielded more accurate results in correctly detecting faulty nodes and isolating them. [ABSTRACT FROM AUTHOR]