SYSTEM FOR OPTIMIZING FATTY ACID PRODUCTION IN SOAP MANUFACTURING PROCESS

Abstract

A system (100) for optimizing fatty acid and steam production is disclosed. The system includes environment models (113a, 113b) derived from digital twins (112a, 112b) of an FSP (fatty acid splitting process) (114a) and an FADP (fatty acid distillation process) (114b). The reinforcement learning (RL) agent (202) receives training using the environment models (113a, 113b) and their predictions, and learns control strategies for optimizing actionable parameters through interaction with the environment models (113a, 113b). The RL agent (202) further outputs control strategy including the optimized actionable parameters, which when implemented or actuated simultaneously, maximize the fatty acid production and minimize the steam production. The system includes environmental models (113c) corresponding to evaporators, based on which an RL agent is trained to make it learn to optimize improvement of concentration of glycerol.

Specification

Description:THE FOLLOWING SPECIFICATION PARTICULARLY DESCRIBES THE INVENTION AND THE MANNER IN WHICH IT IS TO BE PERFORMED.
FIELD OF THE INVENTION
[0001] The present invention generally relates to the field of saponification process optimization, and more specifically, to a system for optimizing fatty acid production and steam production by optimizing actionable parameters associated with FSP (fatty acid splitting process) and FADP (fatty acid distillation process) of the saponification process.

BACKGROUND OF THE INVENTION
[0002] Fatty acids are crucial components in the production of soaps through the saponification process. Saponification is a process that involves the reaction of fats or oils with a base, typically sodium hydroxide (NaOH), to produce glycerol and soap. Generally, it has been observed that the existing solutions do not utilize optimization in fatty acid production for saponification, and thus face several disadvantages. Firstly, the quality of the soap can be inconsistent, leading to variations in texture, lathering properties, and overall user satisfaction. Without optimization, the process may not efficiently utilize raw materials, resulting in higher production costs and increased waste. This inefficiency can also lead to a larger environmental footprint due to higher energy consumption and emissions. Thus, there is a need for leveraging advanced technologies to address these challenges. Optimizing the production of fatty acids is essential for enhancing the efficiency and quality of the saponification process.

OBJECTS OF THE INVENTION
[0003] Some of the objects of the present invention are as follows:

[0004] An object of the present invention is to provide a system for maximizing fatty acid production and minimizing steam production by optimizing actionable parameters associated with FSP (fatty acid splitting process) and/or FADP (fatty acid distillation process) of the saponification process useful in the soap manufacturing process.

[0005] Another object of the present invention is to provide a system for adaptive learning and control. The reinforcement learning (RL) agent can adaptively learn from the process environment, continuously refining its control strategies for improved outcomes.

[0006] Another object of the present invention is to provide a system for facilitating precision in process control. The digital twin and RL agent may enable precise control over process variables, leading to more consistent and high-quality outputs.

[0007] Another object of the present invention is to provide a system that can process real-time operational data. The system effectively processes the real-time operational data for immediate adjustments, enhancing responsiveness to changing process conditions.

[0008] Another object of the present invention is to provide a system for facilitating enhanced predictive abilities. The digital twin's and the RL's capabilities allow for better forecasting and planning, reducing downtime and maintenance costs.

SUMMARY OF THE INVENTION
[0009] In accordance with one aspect of the present disclosure, a system and a method for optimizing fatty acid production and steam production are disclosed. The method includes one or more operations executable by one or more components or devices of the disclosed system. In an embodiment, the system includes a first digital twin that is configured to simulate an FSP (fatty acid splitting process) used in a saponification process. The system further includes a second digital twin that is configured to simulate an FADP (fatty acid distillation process) used in the saponification process. The first digital twin and the second digital twin may be simulated based on the respective collected operational data, relevant to the FSP and the FADP, respectively. Further, a first environment model corresponding to the FSP may be derived from the first digital twin to predict at least CFA (crude fatty acid) final flow. Further, a plurality of second environment models corresponding to the FADP may be derived from the second digital twin to predict at least DFA (distillate fatty acid) flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on a reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on a condenser, and distiller condenser flow. The system further includes a reinforcement learning (RL) agent. The RL agent may be configured to receive training using the derived environment models and their predictions and/or other operational data and constraints. Based on the received training, the RL agent may be further configured to learn control strategies for optimizing actionable parameters through interaction with the derived environment models. The actionable parameters may be optimized while maintaining the requisite constraints associated with the processes. The requisite constraints may include bottom product flow, DFA outlet temperature, DFA flow rate, and CFA final flow, which are maintained within their predefined ranges. The RL agent may be further configured to output at least one control strategy including the optimized actionable parameters. The optimized actionable parameters may include at least drier temperature, feed rate, precut feed temperature, flow rate oil, and high-pressure steam flow. Further, the optimized actionable parameters, when implemented or actuated simultaneously, maximize the fatty acid production and minimize the steam production.

[0010] In an embodiment, the CFA final flow may be predicted based on at least one of flow rate oil, flow rate water, high-pressure steam flow, and first steam injection temperature.

[0011] In an embodiment, the DFA flow rate may be predicted based on at least one of distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, and distiller reflux ratio.

[0012] In an embodiment, the DFA outlet temperature may be predicted based on at least one of distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, distiller reflux ratio, and DFA flow rate.

[0013] In an embodiment, the distiller steam flow rate may be predicted based on at least one of distiller heat load on the reboiler, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, and distiller reflux ratio.

[0014] In an embodiment, the distiller heat load on the reboiler may be predicted based on at least one of distiller steam flow rate, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, distiller reflux ratio, distiller reboiler temperature, distiller reflux temperature, and distiller vapor rates.

[0015] In an embodiment, the precut bottom product flow may be predicted based on at least one of feed rate and precut LFA (lower fatty acid) withdrawal.

[0016] In an embodiment, the distiller vapor rates may be predicted based on at least one of distiller reflux out, distiller steam flow rate, distiller heat load on the reboiler, distiller heat load on the condenser, distiller reflux in, and distiller reflux ratio.

[0017] In an embodiment, the distiller heat load on the condenser may be predicted based on at least one of distiller condenser flow, distiller steam flow rate, distiller heat load on the reboiler, distiller vapor rates, distiller reflux out, and distiller reflux temperature.

[0018] In an embodiment, the distiller condenser flow is predicted based on at least one of distiller heat load on the condenser, distiller steam flow rate, distiller heat load on the reboiler, and distiller reflux temperature.

[0019] In an embodiment, the system may further include a preprocessing module that is configured to process at least the operational data for cleaning and structuring at least the operational data.

[0020] In an embodiment, each environment model may be validated based on real test values and predicted values. Further, preprocessing may be performed and then hyperparameter tuning may be employed for retraining the environment model in cases where the validation does not meet the predefined error thresholds.

[0021] In an embodiment, the RL agent's training may further involve defining a reward function that is based on the requisite constraints that need to be maintained during the FSP and FADP processes. The RL agent may further learn by maximizing the received positive rewards for more desired outcomes and by minimizing the received negative rewards for less desired outcomes.

[0022] In an embodiment, the RL agent may be further configured to generate or create a set of policies corresponding to the control strategies for optimizing the actionable parameters to maximize the fatty acid production and minimize the steam production. The RL agent's policies may be continuously updated based on interactions with at least the one or more environment models and the real-time operational data associated with at least the FSP and FADP processes.

[0023] In an embodiment, the RL agent may be further configured to adapt to changing objectives, system and sub-system requirements, and new data post-deployment, and adjust its policies based on a continuous feedback loop from the real-time operational data associated with at least the FSP and FADP processes.

[0024] In an embodiment, each policy may be validated over the one or more digital twins or the one or more derived environment models. The validation may include simulating the one or more optimized actionable parameters and checking whether it maximizes the fatty acid production and minimizes the steam production while maintaining the requisite constraints.

[0025] In some embodiments, the system may further include a plurality of third environmental models for a plurality of evaporators. Further, each individual RL agent may be configured to optimize improvement of concentration of glycerol.

[0026] These and other features and advantages of the present disclosure may be appreciated from a review of the following detailed description of the present disclosure, along with the accompanying figures in which like reference numerals refer to like parts throughout.

BRIEF DESCRIPTION OF THE DRAWINGS
[0027] The novel features which are believed to be characteristic of the present invention, as to its structure, organization, use, and method of operation, together with further objectives and advantages thereof, will be better understood from the following drawings through which various preferred embodiments of the present invention will now be illustrated and described by way of example. It is expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the present invention. Embodiments of the present invention will now be described by way of example(s) in association with the accompanying drawings in which:

[0028] Figure 1a is a diagram that illustrates a system environment in which various embodiments of the present invention are practiced.

[0029] Figure 1b is a diagram that illustrates components of an application server of the system environment, in accordance with an embodiment of the present invention.

[0030] Figure 2 is a diagram that illustrates an optimization system for optimizing fatty acid production by utilizing an RL agent, in accordance with an embodiment of the present invention.

[0031] Figure 3 is a diagram that illustrates a flowchart of a method for optimizing actionable parameters for optimizing fatty acid production in a saponification process, in accordance with an embodiment of the present invention.

[0032] The figures depict various embodiments of the present invention for purposes of illustration only. One skilled in the art will readily recognize from the following discussion that alternative embodiments of the systems, structures, and methods illustrated herein may be employed without departing from the principles of the disclosure described herein. The present invention will now be explained in further detail, and by way of example, with reference to the accompanying drawings.

DETAILED DESCRIPTION OF THE INVENTION
[0033] The present invention will now be described more fully hereinafter with reference to the accompanying drawings, which form a part hereof, and which show, by way of illustration, specific exemplary embodiments by which the invention may be practiced. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the present invention to those skilled in the art. Among other things, the present invention may be embodied as methods, systems, or devices, or a combination thereof. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. The following detailed description is, therefore, not to be taken in a limiting sense.

[0034] As used in the specification and claims, the singular forms "a", "an", and "the" may also include plural references. For example, the term "an article" may include a plurality of articles. Further, those with ordinary skill in the art will appreciate that the elements in the figures are illustrated for simplicity and clarity and are not necessarily drawn to scale. For example, the dimensions of some of the elements in the figures may be exaggerated, relative to other elements, to improve the understanding of the present invention. There may be additional components described in the foregoing application that are not depicted in one of the described drawings. In the event such a component is described, but not depicted in a drawing, the absence of such a drawing should not be considered as an omission of such design from the specification.

[0035] References to "one embodiment", "an embodiment", "another embodiment", "yet another embodiment", "one example", "an example", "another example", "yet another example", and so on, indicate that the embodiment(s) or example(s) so described may include a particular feature, structure, characteristic, property, element, or limitation, but that not every embodiment or example necessarily includes that particular feature, structure, characteristic, property, element or limitation. Furthermore, repeated use of the phrase "in an embodiment" does not necessarily refer to the same embodiment.

[0036] The words "comprising," "having," "containing," and "including," and other forms thereof, are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items or meant to be limited to only the listed item or items. Unless stated otherwise, terms such as "first" and "second" are used to arbitrarily distinguish between the elements. Thus, these terms are not necessarily intended to indicate temporal or other prioritization of such elements. While various exemplary embodiments of the disclosed invention have been described below, they have been presented for purposes of example only, not limitations. They are not exhaustive and do not limit the invention to the precise form disclosed. Modifications and variations are possible considering the above teachings or may be acquired from practicing of the invention, without departing from the breadth or scope.

[0037] The present invention will now be described with reference to the accompanying drawings which should be regarded as merely illustrative without restricting the scope and ambit of the present invention.

[0038] Figure 1a is a diagram that illustrates a system environment 100 in which various embodiments of the present invention are practiced. The system environment 100 may include one or more servers such as an application server 102 and a database server 104. The system environment 100 may further include a computing device 106 for hosting a plurality of digital twins such as a digital twin 112a and a digital twin 112b. There is further shown a premises 108 including an FSP (fatty acid splitting process) 114a and an FADP (fatty acid distillation process) 114b used in a saponification process. The premises 108 may further include sensors and IOT device 116, a dashboard 118, and a control interface 120. The application server 102, the database server 104, the computing device 106, and one or more devices, sensors, or equipment installed in the premises 108 may be configured to communicate with each other via a communication channel such as a network 110.

[0039] The application server 102 is a framework, a type of software engine, that delivers applications to client computers or devices. The application server 102 may act as a middle layer between the end-user interface and the backend database and business applications, enabling the hosting and delivery of high-level applications, often web based. The application server 102 may be configured to primarily manage the logic of one or more applications. This means that the application server 102 may be configured to handle the execution of program commands, computation, data retrieval, and performance tasks. The application server 102 may be further configured to facilitate the connection between front-end interfaces and back-end databases and servers. For instance, IBM WebSphere application server can connect to various databases like Oracle, SQL Server, or DB2 to fetch and update data as required by the application. The application server 102 may be further configured to provide various middleware services like transaction management, messaging services, and authentication. The application server 102 may be further configured to enable applications to scale to handle increased loads and can manage sessions across multiple servers. The application server 102 may be further configured to include API gateways or management tools to control access to one or more APIs. In some applications, the application server 102 may also assist in delivering content to one or more users. The application server 102 may be further configured to provide a complete development and runtime environment for the one or more applications. For example, Google App Engine offers a fully managed platform as a service (PaaS) that includes an application server for developing and hosting web applications. The application server 102 may be further configured to offer features to secure the one or more applications, like SSL encryption, secure authentication mechanisms, and authorization services. The application server 102 may come with monitoring tools to help track the performance and health of the one or more applications and may provide logging and auditing services. Examples of the application server 102 may include, but are not limited to, Java EE Application Servers, Microsoft Windows-based Application Servers, Platform as a Service, SAP NetWeaver for enterprise applications, and Zend Server for PHP applications. Each application server 102 may offer a unique set of services and performance characteristics tailored to specific application requirements. The choice of the application server 102 may often depend on the needs of the application, the programming language used, the expected load, and the necessary scalability requirements.

[0040] In an exemplary embodiment, the application server 102 may serve as a robust platform for hosting, processing, and executing various models, which may include machine learning (ML) models, deep learning (DL) models, reinforcement learning (RL) models, and other computational and performance-based models. The application server 102 may be involved in a data processing pipeline that covers collection, preprocessing, transformation, and storage of data. The application server 102 may be capable of running different types of models. These could be algorithms or programs designed for specific tasks such as model training and learning, predictive analytics, simulations, or data processing and computation. The application server 102 may be designed to receive a variety of data inputs from multiple sources. For example, operational data may come from the sensors and Internet of Things (IoT) devices (as shown by 116 installed in the premises 108). Further, design data including at least one or more specifications of one or more variable parameters and/or equipment associated with the FSP 114a and the FADP 114b may come from the database server 104. In addition to the operational and design data, key performance indicator (KPI) data such as energy savings, CFA (crude fatty acid) flow, DFA (distillate fatty acid) flow, fatty acid production, steam production, and optimized savings may also be received from an appropriate and applicable source such as the database server 104. The data may be received in real-time or near real-time or based on a scheduled time as per the industry or application requirement or defined by an admin user.

[0041] Further, the application server 102 may be configured to handle unstructured data, which is data that does not have a predefined model or format. The application server 102 may be configured to perform preprocessing that includes cleaning the data (removing irrelevant or erroneous data), normalization (scaling data to a specific range), and transforming it into a format suitable for analysis or model input. The application server 102 may use one or more containers, which are isolated environments for running applications, to process data. This may help in managing dependencies and ensuring consistency across different computing environments. It applies data transformation functions to the processed data. This may involve converting raw data into features suitable for the models to understand or structuring the data according to a predefined schema. The structured data may then be stored in a predefined format, which could be in databases or data warehouses, making it easier for retrieval, processing, analysis, and reporting.

[0042] In an embodiment, the application server 102 may be configured to manage, regulate, optimize, and/or control at least one of the FSP 114a and the FADP 114b and the evaporators 114c for optimizing the fatty acid production, the steam production, and/or the glycerol production. Further, by analyzing and processing the data and running the models, the application server 102 or the RL model running on the application server 102 may optimize the actionable parameters while ensuring all constraints are met. With the optimized actionable parameters including at least optimized drier temperature, optimized feed rate, optimized precut feed temperature, optimized flow rate oil, and optimized high-pressure steam flow, when implemented or actuated simultaneously, may maximize the fatty acid production and minimize the steam production. In some embodiments, the application server 102 may act as an intelligent control system for the saponification process, leveraging data and computational power to maintain an optimal environment. It also ensures that the saponification process is responsive to both the immediate conditions and predictive insights, resulting in a smart, adaptive control and management system.

[0043] The database server 104 is a computer system that provides other computers and applications with services related to accessing and retrieving data stored in the databases. It is a key component in a client-server architecture, where it is dedicated to managing database resources, executing database operations, and ensuring data integrity and security. Examples of the database server 104 include, but are not limited to, Relational Database Servers (such as Oracle Database, Microsoft SQL Server, MySQL, and PostgreSQL), NoSQL Database Servers (such as MongoDB and Cassandra), and In-Memory Databases (such as Redis and SAP HANA). The database server 104 plays a crucial role in housing all the data that the application server 102 or other devices, simulators, controllers, and applications may need to function effectively. It is the repository of current and historical data, which is essential for monitoring the system's performance, training models, optimizing operations, maximizing fatty acid production, and minimizing the steam production and thereby ensuring energy or power savings.

[0044] In an embodiment, the database server 104 may be responsible for storing all the unstructured or structured data, such as the operational data, the design data, the predicted data, the optimized data, the constraint data, and the KPI data. In an embodiment, the database server 104 may be further configured to store the one or more control strategies and the corresponding one or more polices generated or created by the RL model or agent based on its training and learning. In an embodiment, the database server 104 may be further configured to handle one or more queries from the application server 102, the computing device 106, the sensors and IoT devices 116, and/or the control interface 120, retrieving data for further processing or analysis or execution. In an embodiment, the database server 104 may be further configured to manage updates to the database, ensuring that any new data sent from the application server 102, the computing device 106, the sensors and IoT devices 116, and/or the control interface 120 is accurately recorded and that existing data is updated or deleted according to the application's needs. In an embodiment, the database server 104 may be further configured to manage multiple requests at the same time, ensuring that requests are processed reliably and without conflict. In an embodiment, the database server 104 may be further configured to enforce security policies, controlling access to the data, and ensuring that only authorized applications and users can retrieve or modify the data. In an embodiment, the database server 104 may be further configured to regularly back up the data to prevent data loss and provide mechanisms for data recovery in case of server failure, which may be critical for maintaining the continuous operation of the saponification process.

[0045] The computing device 106 is an electronic device that is capable of processing, storing, and retrieving the required data to perform some kind of computations and operations. Such devices usually allow for input and output operations, making them interactive tools for a wide array of applications. One or more users may input data into the computing device 106 through various means, such as keyboards, touch screens, mice, voice recognition, sensors, and cameras. The computing device 106 may convey the results of their computing processes to the users via outputs like displays, speakers, printers, and actuators. Examples of the computing device 106 may include, but are not limited to, personal computer(s), desktop(s), laptop(s), smartphone(s), tablet(s), or wearable device(s).

[0046] In an exemplary embodiment, the computing device 106 may be used by a user to build, create, or simulate one or more digital twins such as the digital twins 112a and 112b that are configured to simulate the FSP 114a and the FADP 114b, respectively, useful in the saponification process. In an embodiment, the computing device 106 may be configured to run software or applications that will allow the user to design and simulate virtual representations (i.e., the digital twins 112a and 112b) of the FSP 114a and the FADP 114b and/or the equipment associated with the FSP 114a and the FADP 114b, creating multi-dimensional models that can be visualized in one-dimension (1D), two-dimension (2D), or three-dimension (3D) interfaces. The FSP and FADP 114a and 114b may be simulated based on the collected operational data and/or design data. For example, after preprocessing of at least the operational data, a first principles-based modelling software (running on the computing device 106 or the application server 102) may be used to create or simulate the first principles chemical-based model based on the processed operational data and/or the design data. Further, a ROM (reduced order model) may be generated from the first principles chemical-based model and simulation. The ROM may include at least design of experiments (DoE) data and outputs of the DoE data. Further, the one or more environment models 113a and 113b may be developed based on at least the DoE data and the outputs of the DoE data. Thus, in an embodiment, the digital twin 112a or 112b may include at least one of the first principles model, the ROM including at least the DoE data and outputs of the DoE data, and the one or more environment models 113a or 113b. The environment models 113a or 113b represent the various external and internal factors that can influence the FSP 114a or FADP 114b. These models are tailored to simulate different scenarios, conditions, and operational contexts in which various equipment of the FSP 114a or FADP 114b may operate. They are typically built using a combination of data-driven approaches, such as machine learning algorithms and empirical data. For instance, an environment model may simulate CFA flow rate. Other environment models may simulate DFA flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on condenser, and/or distiller condenser flow. These models are particularly useful for training the RL model, as they provide a diverse range of scenarios for the RL model to learn from. The RL model may use these environment models 113a or 113b to understand how different conditions affect the one or more processes and to develop strategies and policies for optimizing performance under various circumstances. The environment models 113a or 113b may be validated based on real test values and predicted values. Further, preprocessing may be performed and then hyperparameter tuning may be employed for retraining the one or more environment models 113a or 113b in cases where the validation does not meet predefined error thresholds.

[0047] Once the digital twin 112a or 112b has been established and validated, the computing device 106 may be used to simulate one or more scenarios, analyze system performance, and optimize operations based on the data received from the physical counterpart. In some embodiments, the user may use the computing device 106 to provide one or more inputs to adjust one or more parameters in the digital twin 112a or 112b and perform manual interventions when the automated systems do not suffice, or when unique or unforeseen circumstances arise. In some embodiments, the computing device 106 may be configured to trigger one or more operations in the real-world systems by generating and sending one or more commands from the digital twin's interface, effectively allowing the user to control one or more equipment operational aspects of the premises 108 remotely. In scenarios where the digital twin 112a or 112b is hosted by the application server 102, the application server 102 may trigger the one or more operations in the real-world systems by generating and sending the one or more commands, allowing remote control of the one or more equipment operational aspects of the premises 108.

[0048] The premises 108 refers to an environment or industrial plant where various systems and services are installed to make the space functional and ready for production. This can include infrastructure or plants such as a soap manufacturing plant. The premises 108 may include the FSP equipment associated with the FSP 114a, FADP equipment associated with the FADP 114b, one or more sensing and monitoring devices such as the sensors and IoT devices 116, one or more dashboards such as a dashboard 118, and one or more control interfaces such as a control interface 120 but should not be construed as limiting to the scope of the present disclosure.

[0049] The premises 108 is a soap manufacturing plant that involves a series of chemical and mechanical processes that convert raw materials into the final soap product. The primary processes may include saponification, mixing, plodding, milling, and finishing. Each stage may require specific equipment to ensure efficient and consistent production. The process begins with saponification, where fats and oils are heated and mixed with an alkali (e.g., sodium hydroxide or potassium hydroxide) in large reactors known as saponification tanks. This chemical reaction produces soap and glycerin. After saponification, the soap mixture may be allowed to cool and settle in a Crutcher, which is a large vessel where it can be stirred and blended with additives such as fragrances, colors, and preservatives. Once the soap has been homogenized, it moves to the plodding stage. In this stage, a soap plodder, which is a machine equipped with a screw conveyor, kneads and extrudes the soap into continuous bars. The soap may then be transferred to a milling machine, where heavy rollers may grind and refine the soap to improve texture and consistency. This step may be repeated multiple times to achieve the desired quality. After milling, the soap may be passed through a refining plodder to remove any air pockets and ensure a dense, uniform texture. The soap is then extruded again, cut into individual bars, and stamped with branding or decorative designs using a stamping machine. The stamped soap bars may then be conveyed to drying tunnels or chambers, where they are dried to the appropriate moisture content.

[0050] In the soap manufacturing process, the fatty acid splitting process (FSP) 114a and the fatty acid distillation process (FADP) 114b are two important steps. The FSP 114a, also known as hydrolysis, is an important step in the saponification process used in the soap manufacturing. This process involves breaking down fats and oils into their constituent fatty acids and glycerol through the application of water, heat, and pressure. The splitting process may occur in a vertical, cylindrical reactor called a hydrolyzer or splitting column. In the hydrolyzer, triglycerides (the main components of fats and oils) may be continuously fed into the column from the top, while high-pressure steam is injected from the bottom. The high temperature (usually around 200-260°C) and pressure (20-60 bars) may facilitate the hydrolysis reaction, where water molecules break the ester bonds in triglycerides, producing free fatty acids and glycerol. The splitting column may be designed to maximize the contact between the steam and the oil, ensuring efficient hydrolysis. Inside the column, trays or packing materials may help to disperse the oil and steam, increasing the surface area for the reaction. The free fatty acids, being lighter, rise to the top of the column and are collected, while the heavier glycerol settles at the bottom and is drawn off separately. The fatty acids produced through this process may then be purified and may be used directly in the saponification reaction with an alkali to produce soap. The glycerol by-product, after further purification, may be used as a commodity in various industries, including pharmaceuticals, cosmetics, and food.

[0051] The FADP 114b is an important purification step in the soap manufacturing, used to refine the fatty acids obtained from the hydrolysis or splitting of the fats and oils. This process ensures that the fatty acids used in the saponification process are free from impurities and have the desired purity and consistency. Fatty acid distillation may be performed in a distillation column under reduced pressure, referred to as a vacuum distillation column. The process may begin with the introduction of crude fatty acids into the distillation column. The column may operate under a high vacuum (low pressure) to lower the boiling points of the fatty acids, which prevents their thermal degradation and allows for more efficient separation. The column may be equipped with a series of trays or packing materials that facilitate the separation of fatty acids based on their boiling points. As the crude fatty acids are heated, they vaporize and ascend through the column. In the distillation column, the fatty acids may be separated into different fractions. Short-chain fatty acids, which have lower boiling points, rise to the top of the column and are collected first. Medium and long-chain fatty acids, which have higher boiling points, condense at various points along the column and are collected separately. This separation may be achieved by carefully controlling the temperature gradient along the column, ensuring that each type of fatty acid is condensed and removed at the appropriate level. To enhance the purity of the fatty acids, the process may include multiple distillation stages or the use of a rectification section at the top of the column. The rectification section may further refine the vapor by allowing the more volatile components to rise while the less volatile components return to the column for additional heating and separation. The purified fatty acids may be collected as distillates and may be directly used in the saponification process to produce high-quality soap.

[0052] The premises 108 of the soap manufacturing plant are further equipped with a network of sensors and IoT devices 116, which play a crucial role in monitoring, controlling, and optimizing the various processes involved in the production. These advanced technologies provide real-time data acquisition and communication capabilities, enabling seamless integration and efficient management of the plant's operations. In an embodiment, the sensors installed throughout the plant may be configured to measure one or more parameters such as temperature, pressure, composition, flow rates, and vibration across different stages of the saponification process. Further, the IoT-enabled system may be configured to facilitate remote monitoring and control, allowing the plant operators to access critical data and make informed decisions from anywhere, at any time. This connectivity also supports implementation of the digital twins and the RL agents, which simulate and optimize the fatty acid production and the steam production. By continuously analyzing the data from the sensors and IoT devices 116, these technologies may identify patterns, predict outcomes, and recommend control strategies to enhance productivity, reduce energy consumption, and lower emissions.

[0053] The premises 108 of the soap manufacturing plant are further enhanced by the inclusion of the dashboard 118 and the control interface 120, which serve as the central hubs for monitoring and managing the plant's operations. The dashboard 118 provides a comprehensive, real-time overview of the entire manufacturing process, displaying key performance indicators (KPIs) and critical data from the sensors and IoT devices 116. This visual representation includes real-time, predicted, and/or optimized metrics such as drier temperature, feed rate, precut feed temperature, flow rate oil, high-pressure steam flow, CFA flow rate, DFA flow rate, etc., allowing the plant operators to quickly assess the operational status and identify any potential issues. The control interface 120 is an interactive platform that enables the plant operators to adjust and optimize the process parameters. It allows for precise control over various aspects and stages of the soap production, including equipment power supply management, pressure management, temperature management, supply management, and more. Through this interface, the plant operators can implement control strategies recommended by the digital twins and RL agents, ensuring that the actionable parameters are continuously optimized for maximum efficiency and minimal environmental impact.

[0054] The network 110 is a communication channel between two or more computers or devices that are linked together to share resources, exchange files or commands, or allow electronic communications. The computers on the network 110 may be linked through cables, telephone lines, radio waves, satellites, or infrared light beams. Examples of wired networks may include, but are not limited to, ethernet networks, fiber optic networks, telephone networks, power line communications (PLC), and coaxial networks. Examples of wireless networks may include, but are not limited to, Wi-Fi networks, cellular networks, Bluetooth, satellite networks, near field communication (NFC), Zigbee and Z-Wave. The network 110, be they wired or wireless, plays a critical role in the connectivity of devices and servers of the system environment 100. In an installation like the soap manufacturing plants, the network 110 may be configured to facilitate communication between sensors, actuators, equipment controllers, security systems, and the various servers that manage data storage, data processing, model training, model learning, and control operations.

[0055] In operation, the system 100 for optimizing actionable parameters for maximizing fatty acid production in the soap manufacturing plant 108 has been disclosed. The operation involves a comprehensive and integrated approach utilizing advanced technologies and components such as the digital twins 112a and 112b, the reinforcement learning (RL) agents 202 (Figure 2), the sensors and IoT devices 116, the dashboards 118, and the control interfaces 120. This system 100 functions as a cohesive unit to enhance the efficiency, reliability, and sustainability of the soap production. The system 100 includes the digital twins 112a and 112b, virtual replica of the FSP 114a and the FADP 114b. The digital twins 112a and 112b may be developed or simulated based on the operational data and/or the design data collected from the equipment and the processes associated with at least the FSP 114a and the FADP 114b. The digital twins 112a and 112b simulate the behavior of the FSP 114a and the FADP 114b under various conditions, providing one or more dynamic environment models 113a and 113b that can predict the outcomes of different control strategies. In an exemplary embodiment, the environment model 113a corresponding to the FSP 114a may be derived from the digital twin 112a to predict at least the CFA final flow. Further, the environment models 113b corresponding to the FADP 114b may be derived from the digital twin 112b to predict at least one of the DFA flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on the reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, and distiller condenser flow.

[0056] The RL agent 202 (hosted on a cloud server such as the application server 102 or the database server 104) may be configured to interact with the digital twins 112a and 112b and/or the derived environment models 113a and 113b for its training to learn control strategies for optimizing the actionable parameters with the aim of maximizing the fatty acid production and minimizing the steam production. The RL agent 202 undergoes training by exploring different scenarios within the simulated environment models 113a and 113b, receiving feedback in the form of rewards based on the predefined objectives such as maximizing fatty acid production, minimizing the steam production, minimizing power consumption, reducing fuel consumption, and more. The reward function may be carefully designed to align with the constraints of the one or more processes associated with the soap manufacturing process. The requisite constraints may include at least one of the bottom product flow, DFA outlet temperature, DFA flow rate, and CFA final flow, which are maintained within their predefined ranges. The control strategies may be continuously updated based on real-time data from the plant's sensors and IoT devices 116, which monitor critical parameters throughout the FSP 114a and the FADP 114b and other relevant processes. In real time, based on the received real-time operational data, the RL agent 202 may be configured to output at least one control strategy including the optimized actionable parameters such as at least optimized drier temperature, optimized feed rate, optimized precut feed temperature, optimized flow rate oil, and optimized high-pressure steam flow.

[0057] Further, the processed data (such as real-time data, predicted data, the optimized data, and performance savings or improvements) may be displayed on the dashboard 118, providing the plant operators with a comprehensive, real-time overview of the system's performance. This dashboard 118 may further include visualizations of key performance indicators (KPIs) and alerts for any deviations from the optimal conditions. The control interface 120 may allow the plant operators to interact with the system, implementing the RL agent's 202 recommended optimized control strategies and making manual adjustments as necessary. In some embodiments, the control interface 120 may implement the optimized actionable parameters and/or actuate the related equipment by automatically implementing one or more actions in accordance with the optimized actionable parameters. The control interface 120 also supports fine-tuning of the actionable parameters to maintain the required constraints and respond to dynamic changes in the production environment. Further, the continuous feedback loop from the real-time operational data may allow the RL agent 202 to adapt to changing objectives, system requirements, and new data post-deployment. This adaptability will ensure that the control strategies remain effective over time, even as conditions within the plant 108 evolve. The system's capability for remote monitoring and control further enhances its operational flexibility, allowing the plant operators (e.g., admins, engineers, etc.) to oversee multiple production lines and make informed decisions from the offsite locations.

[0058] Figure 1b is a diagram that illustrates components of the application server 102 of the system environment 100, in accordance with an embodiment of the present invention. The application server 102 may include one or more components such as at least one of: a processor 102a, an RL model 102b, a preprocessing module 102c, and a user interface 102d but should not be construed as limiting to the scope of the present disclosure. The application server 102 may include other components and applications (not mentioned and shown in Figure 1b) but may be included to realize and implement one or more functionalities of the application server 102 in accordance with the intended objectives of the present invention. Each component may be intricately linked, ensuring the application server 102 functions as a cohesive unit. The processor 102a ensures swift and accurate execution of tasks, the RL model 102b adapts and optimizes operational strategies, the preprocessing module 102c guarantees the quality of data fed into the system, and the UI 102d provides a user-friendly platform for interaction with the system. Together, these components and others may form a robust framework for optimizing the saponification process of the soap manufacturing process, ensuring efficiency, precision, and ease of operation.

[0059] The processor 102a is the core computing unit of the application server 102, responsible for executing program instructions and managing the system's operations. Its primary function is to process complex calculations rapidly, handle multitasking between various components of the system, manage the flow of data, and process the data and applications. For instance, the processor 102a may be responsible for running simulations in the digital twins 112a and 112b or the environment models 113a and 113b, processing large datasets for the RL model 102b, training the RL model 102b on the environment models 113a and 113b to make it learn control strategies for optimizing the actionable parameters, and ensuring real-time responsiveness.

[0060] The RL model 102b is a specialized component that applies one or more types of learning techniques to optimize one or more processes and actionable parameters. It learns from the environments, developing one or more strategies for controlling one or more variables such as the drier temperature, feed rate, precut feed temperature, flow rate oil, high-pressure steam flow, CFA flow, DFA flow, DFA outlet temperature, distiller steam flow rate, distiller heat load on the reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, and distiller condenser flow. The RL model 102b may adapt its strategies based on ongoing feedback or real-time data processing, thereby continuously refining its approach to maintain the optimal process conditions.

[0061] In reinforcement learning, an agent is configured to learn to make decisions by performing one or more actions in the one or more environments to achieve a goal. The RL model 102b includes several key components such as the RL agent 202 (as shown in Figure 2). This is the decision-maker that learns from the environments (such as the environment models 113a and 113b) through interactions. The RL agent 202 may be configured to execute one or more RL algorithms and makes observations, takes actions, and receives feedback in the form of rewards. The environment is everything the RL agent 202 interacts with and is external to the RL agent 202. In the RL model 102b, the environment represents a space or context in which the RL agent 202 may operate. Further, a state is a concrete and immediate description of the situation in which the RL agent 202 finds itself. It is a specific condition or context within the environment at a particular time. Further, actions are a set of possible moves or decisions or strategies the RL agent 202 may make in any given state. Each action taken by the RL agent 202 may lead to a new state and a corresponding reward. The reward computation may be a feedback signal to the RL agent 202 that rates the success of an action. This is a mechanism by which the RL agent 202 assesses the success of its actions. A reward function may assign higher rewards for actions that can lead to reduced energy consumption, maximized fatty acid production, minimized steam production, or any other operational parameter optimization. Further, a policy is a strategy that the RL agent 202 employs to decide its actions at each state. It is essentially a mapping from one or more states to one or more corresponding actions, and the goal of the RL agent 202 is to learn or create a policy that maximizes the expected sum of rewards over time. The policy may be refined over time as the RL agent 202 learns from the experiences. The RL agent 202 may generate or create a set of policies corresponding to the control strategies for optimizing the actionable parameters, which when implemented or actuated simultaneously, may maximize the fatty acid production, minimize the steam production, minimize the power consumption, and reduce the fuel consumption. The RL agent's 202 policies may be continuously updated based on interactions with at least one of the digital twins 112a and 112b, the environment models 113a and 113b, the operational data, and the feedback data.

[0062] The preprocessing module 102c is essential for preparing and refining the data before it is used in the system. It cleans, normalizes, and structures the incoming data, ensuring its quality and consistency. This might involve filtering out irrelevant data points, correcting errors, or transforming data formats. For example, the preprocessing module 102c may take raw data and adjust it for any sensor inaccuracies or environmental factors. This preprocessing is crucial because the accuracy and reliability of the system's simulations and predictions depend on the quality of the input data.

[0063] The user interface (UI) 102d is the visual and interactive component through which the plant operators control and monitor the system. It is designed for clarity and ease of use, presenting complex data and system statuses in an understandable format. For example, the UI 102d may display a real-time dashboard with key metrics like the optimized operational parameters or behaviors. The UI 102d may also allow the plant operators to input new settings, respond to system alerts, or access historical data reports. The UI 102d may be crucial for bridging the gap between the system's advanced technical operations and the user, enabling efficient and informed decision-making.

[0064] Although various components of the application server 102 have been individually disclosed and described above, however, a person having ordinary skills in the art would understand that the processor 102a may be configured to execute every functionality of each of the RL model 102b, the preprocessing module 102c, and the UI 102d by itself without limiting the scope of the present disclosure.

[0065] Figure 2 is a diagram that illustrates an exemplary optimization system 200 for optimizing the fatty acid production by utilizing the RL agent 202, in accordance with an embodiment of the present invention. As shown, it illustrates a comprehensive schematic of the system's operation, highlighting the interaction between the server 204, the premises 108, the RL agent 202, and the environment models 113a and 113b in optimizing the fatty acid production in the soap manufacturing plant. The server 204 (such as the application server 102 or the database server 104) interfaces with various components, devices, or processes of the premises 108, from where it collects the real-time operational data associated with the FSP 114a, the FADP 114b, and the evaporator 114c. The data collected may include both operational metrics and design specifications, derived through the network of sensors and IoT devices 116 strategically installed throughout the premises 108. These sensors provide measurements of one or more parameters like the CFA flow rate, DFA flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on condenser, and distiller condenser flow. The sensors further provide measurements of other parameters like the flow rate oil, flow rate water, high-pressure steam flow, first steam injection temperature, feed rate, distiller reflux ratio, distiller reflux out, distiller reboiler temperature, distiller reflux temperature, precut LFA (lower fatty acid) withdrawal, and distiller reflux in. The sensors further provide measurements of other parameters like the sweet water flow, inlet temperature of sweet water, outlet temperature of sweet water, bottom liquid temperature at first vaporization chamber, vapor pressure in first vaporization chamber, temperature at first vaporization chamber, total dissolved solids in first vaporization chamber, bottom liquid temperature at second vaporization chamber, vapor pressure at second vaporization chamber, temperature at second vaporization chamber, total dissolved solids at second vaporization chamber, bottom liquid temperature at third vaporization chamber, pressure at vapor vacuum, temperature at vapor vacuum, total dissolved solids, mass flow rate of product, mass flow rate of product at first effect, mass flow rate of water evaporated at first effect, mass flow rate of product at second effect, mass flow rate of water evaporated at second effect, mass flow rate of product at final effect, mass flow rate of water evaporated at final effect, product conversion for first chamber, product conversion for second chamber, and product conversion for final chamber.

[0066] After collecting the requisite operational data, the server 204 may transmit the collected operational data to the RL agent 202, which may be residing on the same server or on a separate dedicated server. The RL agent 202 (a global or individual RL agent, or a combination thereof) may be configured to process the relevant operational data and output the optimized actionable parameters, including at least the optimized drier temperature, the optimized feed rate, the optimized precut feed temperature, the optimized flow rate oil, and the optimized high-pressure steam flow. These optimized parameters may be essential for maintaining efficient and stable operations while maximizing the fatty acid production, minimizing the steam production, and minimizing the energy consumption. Post the RL agent 202 determines the optimized actionable parameters, these may be communicated to the control interface 120 within the premises 108. The control interface 120 may be configured to process the optimized actional parameters, identify the corresponding actions, and then implement the actions in real-time, adjusting the saponification operations. This step may ensure that the plant continuously operates at optimal efficiency, meeting production goals and environmental standards.

[0067] Further, in some embodiments, the environmental models 113c corresponding to the one or more evaporators 114c may be derived or generated. Further, based on predictions by the environment models 113c, an individual RL agent (such as the RL agent 202) may be trained to make it learn to optimize the improvement of concentration of glycerol. For example, the environment models 113c may be derived for a series of evaporators 114c used in the process of concentrating glycerol, which is a valuable by-product of the fatty acid splitting process. These environment models 113c represent virtual simulations of each evaporator's operation, capturing the essential parameters and dynamics involved in the evaporation process, such as temperature, pressure, feed flow rate, vapor flow, and concentration of glycerol in the output stream. The purpose of these models 113c is to predict how changes in operational settings may impact the efficiency and effectiveness of the glycerol concentration. To optimize the concentration of glycerol, the RL agents (such as the RL agent 202, which can be an individual RL agent or a global RL agent) may be assigned to each evaporator 114c. Each agent may be trained independently using its corresponding environment model 113c to learn how to adjust the evaporator's operating parameters to achieve the highest possible glycerol concentration. The training process for each agent may involve simulating various scenarios in which the agent modifies key inputs, such as at least the steam flow rate, feed rate, and operating temperature, to observe the resulting effects on the glycerol concentration and overall energy consumption. For example, an RL agent might start by experimenting with increasing the steam flow rate to see if this can lead to a more rapid evaporation of water, thereby increasing the concentration of glycerol. The environment model 113c may simulate this scenario and provide feedback to the RL agent in terms of how much the glycerol concentration improved and the energy efficiency of the process. If the result is favorable, then the RL agent may receive a positive reinforcement signal, encouraging it to explore similar strategies. Conversely, if the increased steam flow leads to inefficiencies, such as excessive energy consumption or thermal degradation of glycerol, then the RL agent may receive negative feedback, and then it may try alternative approaches. Over time, each RL agent may refine its strategy through continuous interaction with its environment. The agent may learn to balance multiple objectives, such as maximizing glycerol concentration while minimizing energy use and preventing operational issues like foaming or scaling in the evaporator 114c. The objective for each agent is to develop a control policy that can be implemented in the evaporators 114c, ensuring optimal performance. In practice, these learned control strategies may include at least adjusting the feed rate into the evaporator 114c to maintain a steady state of concentration, fine-tuning the temperature to avoid glycerol degradation, and/or dynamically altering the steam input to respond to variations in the feed composition. For instance, if the glycerol concentration in the feed fluctuates, the RL agent might learn to adjust the evaporation rate to maintain a consistent output concentration, thus improving overall process efficiency. By training the individual agents for each evaporator 114c, the system can account for the specific characteristics and constraints of each unit, leading to the tailored optimization strategies. This approach ensures that each evaporator 114c may operate at peak efficiency, contributing to a higher overall yield of concentrated glycerol with the reduced operational costs and the improved product quality.

[0068] Further, a computing infrastructure (such as the computing device 106 or the application server 102 or the database server 104) may host the digital twins 112a and 112b and/or the environment models 113a, 113b, and 113c corresponding to the FSP 114a, the FADP 114b, and the evaporator 114c. In some embodiments, a single digital twin may be configured to simulate all the environment models 113a, 113b, and 113c. Before actual implementation, the digital twin(s) or the environment model(s) may first be implemented or simulated with the optimized actionable parameters to simulate and validate their effectiveness. This validation process checks whether the parameters will indeed minimize the power consumption, maximize the fatty acid production, and/or minimize the steam production while adhering to all the necessary constraints such as the bottom product flow, DFA outlet temperature, DFA flow rate, CFA final flow, and the like, which are maintained within their predefined ranges. The disclosed system not only enhances the operational efficiency and product quality but also significantly reduces the environmental impact and the operational costs.

[0069] Figure 3 is a diagram that illustrates a flowchart 300 of a method for optimizing the actionable parameters for optimizing fatty acid production in the saponification process, in accordance with an embodiment of the present invention.

[0070] At step 302, the FSP 114a and the FADP 114b are simulated. In an embodiment, the digital twin 112a may be configured to simulate the FSP (fatty acid splitting process) 114a useful in the saponification process. Further, another digital twin 112b may be configured to simulate the FADP (fatty acid distillation process) 114b useful in the saponification process. The digital twin 112a and the digital twin 112b may be simulated based on the collected operational data that are relevant to the FSP 114a and the FADP 114b, respectively. Further, the environment model 113a corresponding to the FSP 114a may be derived from the digital twin 112a to predict CFA final flow. In an embodiment, the CFA final flow may be predicted based on flow rate oil, flow rate water, high-pressure steam flow, and first steam injection temperature. Further, the environment models 113b corresponding to the FADP 114b may be derived from the digital twin 112b to predict DFA flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on the reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, and distiller condenser flow. In an embodiment, the DFA flow rate may be predicted based on distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, and distiller reflux ratio. In an embodiment, the DFA outlet temperature may be predicted based on distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, distiller reflux ratio, and DFA flow rate. In an embodiment, the distiller steam flow rate may be predicted based on distiller heat load on the reboiler, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, and distiller reflux ratio. In an embodiment, the distiller heat load on the reboiler may be predicted based on distiller steam flow rate, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, distiller reflux ratio, distiller reboiler temperature, distiller reflux temperature, and distiller vapor rates. In an embodiment, the precut bottom product flow may be predicted based on feed rate and precut LFA (lower fatty acid) withdrawal. In an embodiment, the distiller vapor rates may be predicted based on distiller reflux out, distiller steam flow rate, distiller heat load on the reboiler, distiller heat load on the condenser, distiller reflux in, and distiller reflux ratio. In an embodiment, the distiller heat load on the condenser may be predicted based on distiller condenser flow, distiller steam flow rate, distiller heat load on the reboiler, distiller vapor rates, distiller reflux out, and distiller reflux temperature. In an embodiment, the distiller condenser flow is predicted based on distiller heat load on the condenser, distiller steam flow rate, distiller heat load on the reboiler, and distiller reflux temperature.

[0071] At step 304, the RL agent 202 may be trained. In an embodiment, the RL agent 202 may receive training by using the derived environment models 113a and 113b. The RL agent's 202 training may further involve defining the reward function that is based on the requisite constraints that need to be maintained during the FSP and FADP processes 114a and 114b. The RL agent 202 learns by maximizing the received positive rewards for more desired outcomes and by minimizing the received negative rewards for less desired outcomes. The RL agent 202 may receive training to make it learn the one or more control strategies for performing the parameter optimization. For example, the RL agent 202 may learn the control strategies for optimizing the actionable parameters through interaction with the derived environment models 113a and 113b. The actionable parameters may be optimized while maintaining the requisite constraints. The requisite constraints may include bottom product flow, DFA outlet temperature, DFA flow rate, and CFA final flow, which are maintained within their predefined ranges.

[0072] At step 306, at least one control strategy is output. In an embodiment, the at least one control strategy including at least the optimized actionable parameters may be output. In one embodiment, the optimized actionable parameters including at least the optimized drier temperature, the optimized feed rate, the optimized precut feed temperature, the optimized flow rate oil, and the optimized high-pressure steam flow may be output. The at least one control strategy with the optimized actionable parameters may be output after receiving the real-time operational data from the sensors and IOT devices 116, and then processing the received operational data by the environment models 113a and 113b to generate the predicted actionable parameters, and then processing the predicted actionable parameters for their optimization by the RL agent 202 to provide the optimized actionable parameters. Post the optimization, the optimized actionable parameters, when implemented or actuated simultaneously, maximize the fatty acid production and minimize the steam production.

[0073] At step 308, the RL agent 202 adapts to changing objectives, system requirements, and new data post-deployment. In an embodiment, the RL agent 202 may adapt and adjust its policies based on a continuous feedback loop from the real-time operational data. Each policy may be validated over the digital twin 112a or 112b or the environment models 113a or 113b. The validation may include simulating the optimized actionable parameters over the digital twin 112a or 112b or the environment models 113a or 113b and checking whether it maximizes the fatty acid production and/or minimizes the steam production.

[0074] In association with Figures 1-3 of the present invention, the digital twins 112a and 112b (Figure 1) may be formed by creating virtual representations of physical processes such as the FSP 114a (Figure 1) and the FADP 114b (Figure 1). These digital twins 112a and 112b may be developed by collecting the operational data from the actual processes, including measurements of flow rates, temperatures, pressures, and other relevant parameters as discussed above. The collected data may be then used to create a mathematical or simulation model that captures the dynamics and interactions of the processes. For the FSP 114a and the FADP 114b, these models may include the chemical reactions, heat transfer, fluid dynamics, and/or other critical aspects of the processes. Further, the simulation software may be used to develop these models, incorporating both the physical laws governing the processes and the specific characteristics of the equipment used. Once the digital twins 112a and 112b are formed, they may be simulated to replicate the real-world behavior of the FSP 114a and the FADP 114b. The simulations may be run using the operational data as inputs, allowing the digital twins 112a and 112b to mimic the conditions and performance of the actual processes. Through this simulation, the digital twins 112a and 112b may predict outcomes such as the flow of CFA, the flow rate of DFA, and various other operational parameters under different conditions. These predictions are crucial for optimizing process performance and identifying potential areas for improvement. Further, the environment models 113a and 113b (Figure 1) may be derived from the digital twins 112a and 112b by using the virtual representations of the physical processes to simulate and analyze various operational scenarios. In context of the FSP 114a and the FADP 114b, the digital twins 112a and 112b, which are based on the collected operational data, serve as the foundation for generating these environment models 113a and 113b. To derive the environment models 113a and 113b, the digital twins 112a and 112b may be subjected to simulations that explore a wide range of operational settings, such as different flow rates, temperatures, and pressures. During these simulations, key process outputs such as the CFA final flow in the FSP 114a and various parameters in the FADP 114b like the DFA flow rate, outlet temperature, and steam flow rate are monitored and recorded. These outputs are influenced by the input parameters and process conditions, and by running numerous simulations, a comprehensive dataset may be generated that captures the relationships between the inputs and outputs. This dataset may then be used to create mathematical models that approximate the behavior of the physical processes. These environment models 113a and 113b encapsulate the cause-and-effect relationships observed in the simulations. For example, the environment models 113b for the FADP 114b might predict how changes in steam flow or condenser rates affect the DFA output and other critical parameters. Once derived, these environment models 113a and 113b may be utilized by the RL agent 202 (Figure 2) to experiment with different control strategies in a virtual setting environment.

[0075] The RL agent 202 may be trained to optimize the fatty acid production and the steam production by interacting with the environment models 113a and 113b derived from the digital twins 112a and 112b of the FSP 114a and the FADP 114b. The RL agent's 202 training may involve learning from the simulated interactions with these environment models 113a and 113b, where it experiments with different control strategies and receives feedback based on the predicted outcomes. The primary goal of this training is to identify the optimal set of actionable parameters that enhance the efficiency and yield of the production processes (such as fatty acid production, steam production, etc.) while adhering to the operational constraints. During the training phase, the RL agent 202 may repeatedly adjust parameters such as drier temperature, feed rate, precut feed temperature, flow rate of oil, and high-pressure steam flow. For each adjustment, the RL agent 202 may use the environment models 113a and 113b to simulate the potential outcomes, such as the flow rate of distillate fatty acids (DFA), the heat load on the reboiler, and other critical performance indicators. The environment models 113a and 113b provide the RL agent 202 with feedback in the form of rewards or penalties based on how close the simulated outcomes align with the desired objectives, such as maximizing production efficiency or minimizing energy consumption. Through this iterative process, the RL agent 202 may learn which combinations of parameters will produce the most favorable results. The RL agent's 202 learning may be guided by reinforcement signals that encourage strategies leading to optimal production and discourage those that result in inefficiencies or violations of the operational constraints. Over time, the RL agent 202 may refine its strategies, developing a set of control policies that can be applied to the actual production processes.

[0076] Once the RL agent 202 has been sufficiently trained, the RL agent 202 may receive the real-time operational data from the sensors and IoT devices 116, and then it may process the real-time operational data based on its training and learning and output at least one control strategy, which includes the optimized actionable parameters. These parameters may be implemented or actuated simultaneously in the real-world production environment. When applied, these optimized actionable parameters may enhance the overall performance of the fatty acid and steam production processes by maximizing output, improving energy efficiency, and ensuring consistent product quality. The implementation of the RL agent's 202 output may involve integrating the learned control strategies into the plant's automation and control systems. This may be done through direct feedback loops where the RL agent 202 continuously monitors the production processes and adjusts the parameters in real-time or by setting the parameters to predefined optimal values that the RL agent 202 has identified.

[0077] While the present system and method has been described with reference to exemplary embodiments, it should also be appreciated that numerous modifications and alternative embodiments may be devised by those having ordinary skill in the art without departing from the broader and intended scope of the present system as set forth in the claims that follow.
, Claims:We claim:
1. A system (100) for optimizing fatty acid and steam production, the system (100) comprising:
a first digital twin (112a) configured to simulate an FSP (fatty acid splitting process) (114a) used in a saponification process, and a second digital twin (112b) configured to simulate an FADP (fatty acid distillation process) (114b) used in the saponification process, wherein the first digital twin (112a) and the second digital twin (112b) are simulated based on collected operational data,
wherein a first environment model (113a) corresponding to the FSP (114a) is derived from the first digital twin (112a) to predict a least CFA (crude fatty acid) final flow, and a plurality of second environment models (113b) corresponding to the FADP (114b) are derived from the second digital twin (112b) to predict at least DFA (distillate fatty acid) flow rate, DFA outlet temperature, distiller steam flow rate, distiller heat load on reboiler, precut bottom product flow, distiller vapor rates, distiller heat load on condenser, and distiller condenser flow; and
a reinforcement learning (RL) agent (202) configured to:
receive training using the derived environment models (113a, 113b) and their predictions;
based on the received training, learn control strategies for optimizing actionable parameters through interaction with the derived environment models (113a, 113b), wherein the actionable parameters are optimized while maintaining requisite constraints; and
output at least one control strategy including the optimized actionable parameters including at least drier temperature, feed rate, precut feed temperature, flow rate oil, and high-pressure steam flow, wherein the optimized actionable parameters, when implemented or actuated simultaneously, maximize the fatty acid production and minimize steam production.

2. The system (100) as claimed in claim 1, wherein the requisite constraints include bottom product flow, DFA outlet temperature, DFA flow rate, and CFA final flow, which are maintained within their predefined ranges.

3. The system (100) as claimed in claim 1, wherein the CFA final flow is predicted based on flow rate oil, flow rate water, high-pressure steam flow, and first steam injection temperature.

4. The system (100) as claimed in claim 1, wherein the DFA flow rate is predicted based on distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, and distiller reflux ratio.

5. The system (100) as claimed in claim 1, wherein the DFA outlet temperature is predicted based on distiller steam flow rate, distiller heat load on the reboiler, feed rate, precut bottom product flow, distiller vapor rates, distiller heat load on the condenser, distiller condenser flow, distiller reflux ratio, and DFA flow rate.

6. The system (100) as claimed in claim 1, wherein the distiller steam flow rate is predicted based on distiller heat load on the reboiler, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, and distiller reflux ratio.

7. The system (100) as claimed in claim 1, wherein the distiller heat load on the reboiler is predicted based on distiller steam flow rate, distiller vapor rates, feed rate, precut bottom product flow, distiller heat load on the condenser, distiller reflux out, distiller reflux ratio, distiller reboiler temperature, distiller reflux temperature, and distiller vapor rates.

8. The system (100) as claimed in claim 1, wherein the precut bottom product flow is predicted based on feed rate and precut LFA (lower fatty acid) withdrawal.

9. The system (100) as claimed in claim 1, wherein the distiller vapor rates is predicted based on distiller reflux out, distiller steam flow rate, distiller heat load on the reboiler, distiller heat load on the condenser, distiller reflux in, and distiller reflux ratio.

10. The system (100) as claimed in claim 1, wherein the distiller heat load on the condenser is predicted based on distiller condenser flow, distiller steam flow rate, distiller heat load on the reboiler, distiller vapor rates, distiller reflux out, and distiller reflux temperature.

11. The system (100) as claimed in claim 1, wherein the distiller condenser flow is predicted based on distiller heat load on the condenser, distiller steam flow rate, distiller heat load on the reboiler, and distiller reflux temperature.

12. The system (100) as claimed in claim 1, comprising a preprocessing module (102c) configured to process at least operational data for cleaning and structuring at least the operational data.

13. The system (100) as claimed in claim 1, wherein each environment model (113a or 113b) is validated based on real test values and predicted values, and
wherein preprocessing is performed and then hyperparameter tuning is employed for retraining the environment model (113a or 113b) in cases where the validation does not meet predefined error thresholds.

14. The system (100) as claimed in claim 1, wherein the RL agent's (202) training involves defining a reward function based on the requisite constraints that need to be maintained during the FSP and FADP processes (114a, 114b), and
wherein the RL agent (202) learns by maximizing the received positive rewards for more desired outcomes and by minimizing the received negative rewards for less desired outcomes.

15. The system (100) as claimed in claim 1, wherein the RL agent (202) is configured to generate or create a set of policies corresponding to the control strategies for optimizing the actionable parameters to maximize the fatty acid production and minimize the steam production, and
wherein the RL agent's (202) policies are continuously updated based on interactions with at least the environment models (113a, 113b) and real-time operational data associated with the FSP and FADP processes (114a, 114b).

16. The system (100) as claimed in claim 15, wherein the RL agent (202) is configured to adapt to changing objectives, system and sub-system requirements, and new data post-deployment, and adjust its policies based on a continuous feedback loop from the real-time operational data associated with the FSP and FADP processes (114a, 114b).

17. The system (100) as claimed in claim 15, wherein each policy is validated over the digital twin (112a or 112b) or the derived environment models (113a or 113b), and
wherein the validation includes simulating the optimized actionable parameters and checking whether it maximizes the fatty acid production and/or the minimize steam production while maintaining the requisite constraints.

18. The system (100) as claimed in claim 1, comprising a plurality of third environmental models (113c) for a plurality of evaporators,
wherein an individual RL agent is trained to make it learn to optimize improvement of concentration of glycerol.

Documents