FORM 2 THE PATENTS ACT, 1970 (39 of 1970) & THE PATENT RULES, 2003 COMPLETE SPECIFICATION (See Section 10 and Rule 13) Title of invention: METHOD AND SYSTEM FOR DEVELOPING PRETRAINED MODELS FOR INDUSTRIAL DIGITAL TWINS Applicant Tata Consultancy Services Limited A company Incorporated in India under the Companies Act, 1956 Having address: Nirmal Building, 9th floor, Nariman point, Mumbai 400021, Maharashtra, India Preamble to the description: The following specification particularly describes the invention and the manner in which it is to be performed. TECHNICAL FIELD [001] The disclosure herein generally relates to a data analytics, and, more particularly, to a method and system for developing pretrained models for industrial digital twins. BACKGROUND [002] Industrial systems, assets, and equipment are constrained by limited real-time sensor measurements, often operating at different time scales. Performance monitoring relies heavily on laboratory and experimental data. Due to the limited real-time sensor measurements, monitoring of key performance indicators (KPI) in real time is difficult which may lead to sub optimal operations of these systems. The acquisition of experimental data is both costly and subject to uncertainty, owing to limited sample sizes, leading to potential errors in predictions and estimations. Additionally, the irregular availability of experimental data compounds these issues. Therefore, to assess the current state of equipment, to enable predictive and prescriptive maintenance, models that capture the underlying physics of the processes in detail are imperative. For developing data-driven models capable of capturing this level of process physics demands an extensive amount of data, a resource often lacking in this context. Furthermore, data-driven models alone fall short in capturing the complete physics of the process, limiting their predictive power and utility. [003] Further, physics-based models provide a comprehensive representation of the process physics but are hindered by high computational costs and accuracy contingent on numerical methods that are implemented. Their unsuitability for real-time monitoring due to extended simulation and prediction times further limits their practicality. Reduced-order models, designed as substitutes for high-fidelity physics-based models, introduce simplifications that compromise precision during estimation. In response to these challenges, physics-based data-driven models emerge as a remedy to address the computational expenses of physics-based models, the interpretability, and physical consistency shortcomings of purely data-driven models. The physics-based data-driven models permit real-time monitoring, diagnosis, predictive, and prescriptive analyses due to their minimal prediction times. However, developing such models for industrial systems confronts various challenges. Designing the model architecture to achieve good predictive capability, which often requires extensive trial and error. Ensuring that the model captures the necessary physics of the process, however, while validating against experimental/lab measurements, a task that poses convergence challenges. Addressing the difficulty of achieving physical consistency in the models, especially when process parameters are unknown or require tuning. However, most of the times even if the approximate of physics of the process is known, parameters of the process are unknown or their exact formulation for given process needs to be tuned. Due to high number of unknown parameters or their formulations incorporating the physics in the models is tough. Handling challenges related to limited scenarios covered in experimental/sensor measurements, uncertainty and reliability issues in experimental data, and potential errors in sensor measurements, and operational shifts or faults. [004] In the current landscape, multiple modeling techniques are available, including the physics-based models, the data-driven models, and hybrid models that amalgamate both approaches. Computational intensity of the high-fidelity physics-based models, limited quality, and explication of measurement data, which hinder widespread deployment of data-driven models within industrial settings. Hybrid models, which combine merits of both approaches, offer enhanced comprehensibility and reliability while mitigating computational overhead. However, in industrial contexts, obscurity of the underlying physics often leads to the unavailability of critical parameters, impeding the seamless integration of the physics-based approach with data-driven methods. Furthermore, due to the scarcity and noise in sensor measurements, predicting the parameters of the physics-based approach from data becomes unfeasible. Moreover, the iterative selection and search for a suitable model design for hybrid models to deliver high-fidelity process parameter solutions are complicated by ill-posed nature of physics model definitions and scarcity of high-quality measurement data. SUMMARY [005] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one aspect, a method of developing one or more pretrained models for industrial digital twins is provided. The processor implemented method includes receiving, via one or more hardware processors, a data associated with one or more entity as an input; preprocessing, via the one or more hardware processors, the data associated with the one or more entity to obtain a preprocessed data; generating, via the one or more hardware processors, a physics-based mathematical formulation (M) for the one or more entity; developing, via the one or more hardware processors, one or more n data-driven models using one or more machine learning techniques based on one or more known parameter or one or more unknown parameter to select one or more top m models; selecting, via the one or more hardware processors, one or more top k unknown parameters based on a highest value of a parameter attribution score for the one or more selected top m models; determining, via the one or more hardware processors, an estimated value for the one or more top k unknown parameters using an optimization-based technique; developing, via the one or more hardware processors, one or more physics informed data driven (PIDD) models for each of the one or more selected top m models using the estimated value for the one or more top k unknown parameter; and iteratively training, via the one or more hardware processors, the one or more physics informed data driven (PIDD) models by minimizing a loss function L, which combines a data-based error (Ldata()). [010] In an embodiment, the sensor measurement data pertains to (i) a temperature measurement data, (ii) a pressure measurement data, (iii) a flow measurement data, and (iv) a vibration measurement data. In an embodiment, the experimental measurement data pertains to a property or a quality estimation of a fluid or a solid in processing of the one or more entity. In an embodiment, the data associated with the one or more entity is preprocessed to obtain the preprocessed data, includes: (a) an aggregated dataset () is obtained at uniform intervals, (b) the outlier data, and the missing data in the aggregated dataset () is imputed to obtain the cleaned imputed dataset (). In an embodiment, the aggregated dataset () include the outlier data and the missing data which are to be identified by (i) a defining threshold method, or (ii) a statistical method, and (iii) one or more imputation methods respectively. In an embodiment, the one or more parameters are marked as the one or more unknown parameters, if the one or more parameters in the physics-based mathematical formulation (M) is not present in the comparison mappings generated for the one or more entity using the mapping model. In an embodiment, the physics-based mathematical formulation (M) pertains to a set of algebraic equations, or a set of differential equations, or combination thereof. In an embodiment, the one or more trained n data-driven model are ranked based on one or more evaluated model performance metrics. In an embodiment, the one or more evaluated model performance metrics pertains to one or more error metrics. In an embodiment, the one or more error metrics are computed for each n data-driven model , and a rank is assigned using one or more ranking methods. In an embodiment, the one or more top m models are selected based on the corresponding ranks . In an embodiment, one or more ranges are estimated for the one or more unknown parameters by querying a reference literature and a database. In an embodiment, the sensitivity () is calculated as a partial derivative of a residue with respect to each parameter . In an embodiment, the residue of the physics-based mathematical formulation (M) is calculated using one or more data-driven model predictions for each element in a set of parameter combinations (_, ). [011] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed. BRIEF DESCRIPTION OF THE DRAWINGS [012] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles: [013] FIG. 1 illustrates a system to develop one or more pretrained models for industrial digital twins, according to an embodiment of the present disclosure. [014] FIG. 2 is an exemplary block diagram of the system of FIG. 1 to develop the one or more pretrained models for the industrial digital twins, according to an embodiment of the present disclosure. [015] FIG. 3A through FIG. 3C are exemplary flow diagrams illustrating a method of developing one or more physics-informed data-driven models for the industrial digital twins, according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF EMBODIMENTS [016] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. [017] There is a need for an efficient approach to address the challenges associated with hybrid model development and difficulty in achieving physical consistency in models, especially when process parameters are unknown or require tuning. Embodiments of the present disclosure provide a method and system for developing pretrained models i.e., one or more physics-informed data-driven models for industrial digital twins. A data associated with one or more entity is received as an input. The data associated with the one or more entity pertains to (i) a sensor measurement data, (ii) an experimental measurement data, (iii) a design data, and (iv) one or more operational parameters. The one or more entity pertains to a system, or an equipment, or a process. The data associated with one or more entity is preprocessed to obtain a preprocessed data, and the preprocessed data pertains to a cleaned imputed dataset (). The data associated with one or more entity is mapped with one or more known parameters or one or more unknown parameters of the one or more entity to generate comparison mappings for the one or more entity using a mapping model. A physics-based mathematical formulation (M) is generated for the one or more entity. The one or more data-driven models are developed using a machine learning technique based on the one or more identified known parameters or the one or more identified unknown parameters to select one or more top m models. One or more top k unknown parameter is selected based on a highest value of a parameter attribution score for each selected top m model. The parameter attribution score is determined based on a degree of deviation or a sensitivity () of a residue of the physics-based mathematical formulation (M) for one or more changes in the one or more unknown parameters. An estimated value is determined for the top k unknown parameter using an optimization-based technique. One or more physics informed data driven (PIDD) models are developed for each of the one or more selected top m models using the estimated value for the one or more top k unknown parameters. The one or more physics informed data driven (PIDD) model is iteratively trained by minimizing a loss function ℒ, which combines a data-based error and a physics-based error. The physics-based error and the data-based error must be less than a pre-defined physical discrepancy threshold and a data discrepancy threshold respectively. The one or more trained physics informed data driven (PIDD) models associated with the one or more top k unknown parameters associated with the one or more entity are iteratively updated. [018] Referring now to the drawings, and more particularly to FIGS. 1 through 3C, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments, and these embodiments are described in the context of the following exemplary system and/or method. [019] FIG. 1 illustrates a system 100 to develop one or more pretrained models for industrial digital twins, according to an embodiment of the present disclosure. In an embodiment, the system 100 includes one or more processor(s) 102, communication interface device(s) or input/output (I/O) interface(s) 106, and one or more data storage devices or memory 104 operatively coupled to the one or more processors 102. The memory 104 includes a database. The one or more processor(s) processor 102, the memory 104, and the I/O interface(s) 106 may be coupled by a system bus such as a bus 108 or a similar mechanism. The one or more processor(s) 102 that are hardware processors can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the one or more processor(s) 102 is configured to fetch and execute computer-readable instructions stored in the memory 104. In an embodiment, the system 100 can be implemented in a variety of computing systems, such as laptop computers, notebooks, hand-held devices, workstations, mainframe computers, servers, a network cloud, and the like. [020] The I/O interface device(s) 106 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like. The I/O interface device(s) 106 may include a variety of software and hardware interfaces, for example, interfaces for peripheral device(s), such as a keyboard, a mouse, an external memory, a camera device, and a printer. Further, the I/O interface device(s) 106 may enable the system 100 to communicate with other devices, such as web servers and external databases. The I/O interface device(s) 106 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, local area network (LAN), cable, etc., and wireless networks, such as Wireless LAN (WLAN), cellular, or satellite. In an embodiment, the I/O interface device(s) 106 can include one or more ports for connecting number of devices to one another or to another server. [021] The memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random-access memory (SRAM) and dynamic random-access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes. In an embodiment, the memory 104 includes a plurality of modules 110 and a repository 112 for storing data processed, received, and generated by the plurality of modules 110. The plurality of modules 110 may include routines, programs, objects, components, data structures, and so on, which perform particular tasks or implement particular abstract data types. [022] Further, the database stores information pertaining to inputs fed to the system 100 and/or outputs generated by the system (e.g., data/output generated at each stage of the data processing) 100, specific to the methodology described herein. More specifically, the database stores information being processed at each step of the proposed methodology. [023] Additionally, the plurality of modules 110 may include programs or coded instructions that supplement applications and functions of the system 100. The repository 112, amongst other things, includes a system database 114 and other data 116. The other data 116 may include data generated as a result of the execution of one or more modules in the plurality of modules 110. Herein, the memory for example the memory 104 and the computer program code configured to, with the hardware processor for example the processor 102, causes the system 100 to perform various functions described herein under. [024] FIG. 2 is an exemplary block diagram of the system 100 of FIG. 1 to develop the one or more pretrained models for the industrial digital twins, according to an embodiment of the present disclosure. A system 200 may be an example of the system 100 (FIG. 1). In an example embodiment, the system 200 may be embodied in, or is in direct communication with the system, for example the system 100 (FIG. 1). The system 200 is configured to develop the one or more pretrained models i.e., one or more trained physics-informed data-driven models for the industrial digital twins. The system 200 includes a sensor data preprocessing unit 202, a parameter identification unit 204, a physics-based mathematical formulation (M) generation unit 206, a data driven model search unit 208, a physical 5 discrepancy explainer unit 210, a physical discrepancy minimizer unit 212, and a physics informed data driven model development unit 214. A data associated with one or more entity is received as an input. The data associated with the one or more entity pertains to (i) a sensor measurement data, (ii) an experimental measurement data, (iii) a design data, and (iv) one or more operational parameters. The one or 10 more entity pertains to a system, an equipment, and a process. The sensor measurement data pertains to but is not limited to (i) a temperature measurement data, (ii) a pressure measurement data, (iii) a flow measurement data, and (iv) a vibration measurement data. The sensor measurement data are obtained using one or more physical sensor instruments, and are stored as per design and capacity of 15 the sensor instruments. The experimental measurement data pertains to a property or a quality estimation of fluid or solid in processing of the one or more entity. The measurements are carried out in an experimental laboratory at a predetermined interval. The pre-determined interval herein may pertain to (i) once every eight-hour measurement, or (ii) once everyday measurement intervals, or (iii) similar 20 measurement intervals. [025] The sensor data preprocessing unit 202 pre-processes the data associated with the one or more entity to obtain a preprocessed data. The preprocessed data pertains to a cleaned imputed dataset (). The step of preprocessing of the data associated with the one or more entity includes (a) an 25 aggregated dataset () is obtained at uniform intervals, and (b) an outlier data, and a missing data is imputed to obtain the cleaned imputed dataset () based on the aggregated dataset (). The uniform intervals pertains to measurements recorded per minute, or per hour, or any pre-determined measurement frequency. The parameter identification unit 204 maps the data associated with the one or more 30 entity with one or more known parameters or one or more unknown parameters of the one or more entity using a mapping model. 16 [026] For example, considering the one or more pretrained models to be developed for a rotary kiln as an equipment which produces materials that require high-temperature treatment. The sensor measurement data and the experimental measurement data are collected as the input which includes the temperature 5 measurement data at various locations along with length of the rotary kiln, and gas composition measurements. The temperature measurement data is represented as mentioned below: _ (): Temperature of gas off a wall at location in Kelvin. _(): Temperature of gas off a bed at location in Kelvin. 10 (): Temperature of the solid bed at location in Kelvin. (): Temperature of the kiln wall at location in Kelvin. : Position along the kiln length (in meters). [027] Similarly, the gas composition measurement data is represented as mentioned below: 15 (): Concentration of component in a gas phase at time . Components: 2, 2, , 2, , , 4, . : Time in hours. [028] The temperature measurement data is spatially distributed along the length of the rotary kiln's, recorded by a temperature measurement sensor at time 20 interval of 1 min and stored in the database, and the gas composition measurement data is temporally distributed, recorded at specific time intervals of every 8 hours with experimental sample analysis. In an embodiment, location of the temperature measurement sensor is captured in a piping and instrumentation diagram (P&ID) of the rotary kiln. A mapping of the sensor measurement data and corresponding 25 location are created using the mapping model. The mapping model considers input image of the P&ID diagram and a sensor measurement schema as an input to create the mapping for the sensor measurement data and the corresponding location as an output. The mapping model can be a pre-trained multimodal model which take input in an image and a text format. The mapping model return a parameter list associated 30 with the equipment that is measured and corresponding measurement sensor ids. The parameter list pertains to type, design capacity, dimensions, material, material 17 physical, and chemical properties, and setpoint of operation. The parameter list may be in a string format. [029] The temperature measurement data and the gas composition measurement data are recorded at different time intervals and are aggregated to 5 analyze performance of the rotary kiln. A common time interval is chosen for aggregation. e.g., a 15-minute interval is proposed. For the temperature measurement data, which is more frequent, are interpolated or extrapolated to match the 15-minute intervals. Similarly, for the gas composition measurement data, which is less frequent, summarize the data over each of 15-minute interval. 10 Let be the set of all temperature measurement types (i.e., __ , __ , , ). Let be the set of all gas composition components (i.e., 2, 2, , 2, , , 4, ). : Original time interval for temperature measurements (1 min). 15 : Original time interval for gas composition measurements (8 hours). : Aggregated time interval (15 min). : Number of original intervals in one aggregated interval () for temperature and () for gas composition. [030] An aggregated temperature measurement data model is obtained. 20 Consider, for each measurement type ∈ , and at each location , the aggregated temperature measurement data model ̅(x , tagg) is calculated as: (,) = { } ∑(,) Where are time points within the aggregated interval . [031] An aggregated gas composition measurement data model is 25 obtained. Consider, for each component ∈ , an aggregated composition () over an interval is approximated as: () = () Where is a closest time point to within the 8-hour measurement interval. 18 [032] A model is inputted with (a) a time series of the temperature measurement data at different locations along the rotary kiln, recorded every minute, and (b) a time series of gas composition measurement data, recorded every 8 hours respectively to obtain the aggregated dataset () i.e., (i) an aggregated 5 temperature measurement data averaged over 15-minute intervals, and (ii) an aggregated gas composition measurement data summarized to align with the 15- minute intervals respectively. The aggregated dataset () may include the outlier data and the missing data which are identified by defining threshold or a statistical method, and one or more imputation methods respectively. 10 [033] An outlier detection model is generated by implementing the statistical method or an interquartile range (IQR) method. The statistical method is implemented to identify the outlier data in the aggregated dataset () by computing a mean () and a standard deviation () of the aggregated dataset (). An outlier threshold is defined as e.g., data points outside ± are considered outliers, 15 where is a predefined constant. The outlier data identified by the statistical method are as depicted below: = { ∈ : < - > + } Where : Set of identified outliers, : A single data point in the . 20 ℳ: Set of missing data. [034] The interquartile range (IQR) method is implemented to identify the outlier data in the aggregated dataset (). The outlier data is identified by calculating a first quartile (Q1) and a third quartile (Q3) of the , IQR = Q3 - Q1, 25 [035] The outlier threshold is defined as data points outside i.e., [1 - × ,3 + × ]. = { ∈ : < 1 - × > 3 + × } [036] The identification of missing data is represented as mentioned below: 30 ℳ = { ∈ : }. 19 [037] An imputation model is generated by one or more methods but is not limited to a mean or median imputation method, an interpolation method, and a machine learning-based imputation method. In the mean or median imputation approach, the missing data, or the outlier data are replaced with a mean or median 5 of the aggregated dataset (B). dvmputed = mean(t)ormedian(t)for diEMUO [038] In the interpolation method, considering a time-series data, a linear or a spline interpolation are used to estimate missing data based on neighboring data points. The aggregated dataset (B) is considered as an input to obtain the 10 cleaned imputed dataset (Vcleaned) or the imputed model i.e., each missing data or outlier data, the dt is replaced with the dfnputed. [039] The physics-based mathematical formulation (M) generation unit 206 generates a physics-based mathematical formulation (M) for the one or more entity. The parameter identification unit 204 includes a text retrieval model which 15 retrieves the physics-based mathematical formulation (M) available in a text or a document which are stored in the database. In an embodiment, one or more large language models (LLMs) are implemented by appropriate query to generate the physics-based mathematical formulation (M). The parameter identification unit 204 identifies one or more parameters in the physics-based mathematical formulation 20 (M) by comparison mappings generated for the one or more entity. The one or more parameters pertains to one or more known parameters, or one or more unknown parameters. The one or more parameters are marked as the one or more unknown parameters, if the one or more parameters in the physics-based mathematical formulation (M) is not present in the comparison mappings generated for the one or 25 more entity using the mapping model. [040] For example, considering the one or more pretrained models to be developed for the rotary kiln. The physics-based mathematical formulation (M) which is generated includes a set of algebraic equations, or a set of differential equations, or a combination thereof. The physics-based mathematical formulation 30 is represented as mentioned below: 20 = () where represents function that generates the physics-based mathematical formulation () based on the given process context, . herein can be rule based programs or artificial intelligence (AI), i.e., the one or more large language 5 models (LLMs) or a deep learning-based model. The context pertains to a process description and an operational context of the process which is inputted to the model to obtain the set of algebraic or the set of differential equations describing the process. The physics-based mathematical formulation () is considered as an input to identify a source for one or more model parameters i.e., one or more plant design 10 documents, a maintenance history, and a research literature. The one or more known parameters or the one or more unknown parameter are identified using the mapping model based on the physics-based mathematical formulation (M) for the one or more entity. : Set of parameters required for the physics-based mathematical 15 formulation (), = { 1,2 }. : Set of sources for information, = {1( ),2( ℎ ),3()} (, ): Query function for parameter and source . 20 [041] The one or more parameters in the M is identified and categorized based on availability in the . For parameters with known values: = {(,)|∈ , ℎ }. For parameters without known values: 25 = { | ∈ , }. [042] One or more query functions are generated for the using (,). [043] For example, a sample implementation for heat transfer in the rotary Kiln used in the cement industry is considered. The documents required as an input 30 are plant design specification document, a kiln design specification document, a maintenance history document/database, operational settings/control guideline 21 document. The sample model output is (a) the set of differential equations representing heat transfer dynamics within the rotary kiln, and (b) the one or more parameters i.e., (i) the one or more known parameters e.g., Kiln dimensions, material thermal properties from plant design specifications and the mapping 5 obtained using the mapping model, (ii) the one or more unknown parameters e.g., queries generated for specific heat transfer coefficients, emissivity values (requiring research in literature). [044] The data driven model search unit 208 develops one or more data-driven models i.e., n data-driven models using one or more machine learning 10 techniques based on the one or more identified known parameters or the one or more identified unknown parameters (i.e., one or more neural networks with varying architectures and varying parameters) to select one or more top m models. The one or more neural networks with varying architectures pertains to but is not limited to a deep neural network (DNNs), a convolutional neural network (CNNs), 15 a recurrent neural network (RNNs). The varying parameters pertains to number of layers, neurons per layer, types of activation functions, for each model. In an embodiment, a supervised learning algorithm is implemented to train each n data-driven model to obtain one or more trained n data-driven models based on the cleaned imputed dataset (), and an error metrics is calculated. The one or 20 more trained n data-driven models are ranked based on one or more evaluated model performance metrics. The one or more evaluated model performance metrics pertains to the one or more error metrics. The one or more error metrics are computed for each n data-driven model and a rank is assigned using the one or more ranking methods. The one or more top m models are selected based on the 25 corresponding ranks. [045] For example, the data-driven models are developed based on variations of the DNN by considering one or more temperature predictions (, , , ℎ) of the rotary kiln. A model structure for ℎ DNN Model includes: (i) an input layer which receives a sensor location (); (ii) hidden layers i.e., 30 multiple layers, each with the varying number of neurons, e.g., specify hidden layers with neurons in ℎ layer; (iii) activation functions i.e., each layer may 22 use different activation functions, e.g., a rectified linear unit (ReLU), a Sigmoid, or a Tanh; and (iv) output layer which produces a vector = [,,,ℎ], representing the predicted temperatures. : Number of DNN models. 5 : Input data (i.e., the sensor location). : output prediction (i.e., the sensor measurements: , , , ℎ). : Function of the ℎ DNN model. : Parameter set for the ℎ model. [046] There are one or more variations in DNN Architectures (a) a grouped 10 network i.e., sub-networks where each predicts a subset of outputs e.g., one sub¬ network for [, ] and another for [, ℎ]; (b) a sequential network: i.e., multiple sub-networks where output of one is passed as an input to the next, along with the sensor location. Each sub-network predicts one temperature component, e.g., first sub-network predicts , the second uses the prediction and to predict . 15 [047] For example, a mathematical formulation for training and evaluation for deep neural networks for a rotary kiln heat transfer model is considered. The input considered is DNN models with varying architectures and the parameters . During the training phase, each model is trained using the supervised learning algorithm on the cleaned imputed dataset () containing and . The 20 models are ranked based on one or more error metrics e.g., a Root Mean Squared Error (RMSE), a Mean Absolute Error (MAE), and a Mean Absolute Percentage Error (MAPE). For each metric, models are ordered from a lowest score to a highest score. Error metrics is derived from (,̂) = 1 ∑( - ) } 25 }(,̂)= ∑ | - ̂| 100 ∑| - ̂ (,)=— 23 [048] The models are ranked also by (i) a composite score ranking, (ii) a Pareto ranking, (iii) a cross-validation ranking. The composite score ranking in which a composite score is developed by combining the RMSE, the MAE, and the MAPE. For example, a weighted average of three metrics are used, where weights 5 reflect importance of each metric in the specific context. The Pareto ranking in which models are ranked based on multiple metrics simultaneously. A model is considered Pareto optimal if no other model performs better in all the metrics which results in a set of top-performing models, each optimal in different aspects. The cross-validation ranking use a cross-validation to evaluate each model's 10 performance. The models are ranked based on the average performance across different folds. [049] The one or more error metrics are computed for each model , and a rank is assigned based on the chosen ranking method. The one or more top ‘m’ models are selected based on the corresponding ranks . 15 : Number of data points. : True values of sensor measurements. : Predicted values from the ℎ model. : Number of top-performing models to select. : Ranking of the ℎ model. 20 [050] The accurate prediction of the physics-based mathematical formulation () hinges on estimation of the one or more unknown parameters. A subset of unknown parameter from the one or more unknown parameters is determined by a parameter attribution score, which is derived from a degree of deviation between the predictions of the n data-driven models and the physics- 25 based mathematical formulation (). The physical discrepancy explainer unit 210 estimates one or more ranges for unknown parameters by querying a reference literature and a database. The reference literature may include research papers or book on topic of equipment/system mathematical formulation. The databases include one or more material properties database, empirical relations database. 30 For each ∈ , query literature and databases to estimate (). () = (, ). 24 Where Punkn0Wn is a set of unknown parameters, and R(p) is an estimated range for parameter p. [051] For each p G Punknown, sample values within the range R(p). For example, Monte Carlo or Latin Hypercube Sampling can be used for efficient 5 exploration of a parameter space. A set of parameter combinations (P_known, P^lZwn) are generated. [052] The physical discrepancy explainer unit 210 includes a parameter attribution determination unit 210A which determines the parameter attribution score based on a degree of deviation or a sensitivity S(p) of a residue of the physics-10 based mathematical formulation (M) for one or more changes in the one or more unknown parameters. The residue of the physics-based mathematical formulation (M) is calculated using the n data-driven models prediction for each element in the set of parameter combinations (P_known, P^^wn) . For example, the residue is computed for each combination: 15 Residue(Pknown, Punknown, Yt ): Residue function evaluated for the physics-based mathematical formulation (M) with the n data-driven model predictions. [053] The sensitivity S(p) is calculated as a partial derivative of the residue with respect to each parameter p. The sensitivity S(p) measures how changes in p affect the discrepancy between the physics-based mathematical 20 formulation (M) and the n data-driven models: Where S(p): Sensitivity of the residue with respect to parameter p. Pknown: Set of known parameters. Punknown: Set of unknown parameters. 25 R (p) : Range for each unknown parameter p. ?: Prediction from the ith data-driven model. [054] The previously calculated sensitivity values S(p) for each parameter V e Punknown. The score for parameters in Punknown is assigned based on corresponding sensitivity value S(p) in a descending order. The physical 30 discrepancy explainer unit 210 selects one or more top k unknown parameters based 25 on a highest value of the parameter attribution score for the one or more selected top m models. For example, the mathematical formulation is mentioned as below: _ = Top elements from sorted {()| ∈ }. Where : The number of top parameters to be selected, and _: Set of 5 top unknown parameters based on sensitivity. [055] The physical discrepancy minimizer unit 212 determine an estimated value for the one or more top k unknown parameters using an optimization-based technique. The optimization-based technique pertains to a particle swarm optimization technique, or a genetic technique, or an expectation 10 maximization technique. An objective function (()) is constructed, which measures the discrepancy between the physics-based mathematical formulation (M) and the n data-driven models using both known parameters and estimated values of _ . (): Function quantifying the discrepancy for the 15 physics-based mathematical formulation M calculated using the n data- driven model predictions is same as the (,, ̂ used in the parameter attribution determination unit 210A. Note _ is subset of . Where : Set of top unknown parameters selected based on 20 parameter attribution determination unit output. : Vector representing the values of parameters in . [056] The optimization-based technique iteratively adjusts the to find an optimal set of parameter values , thereby the discrepancy (()) is minimized. 25 Where : Optimized values of . [057] The physics informed data driven model development unit 214 develops one or more physics informed data driven (PIDD) models for each of the one or more selected top m models using the estimated value for the one or more top k unknown parameters. The one or more physics informed data driven (PIDD) 30 models are iteratively trained by minimizing a loss function ℒ, which combines a 26 data-based error (£data(0)) and a physics-based error (Lphysics((p)). The physics-based error i.e., quantified as a L2 norm of the mathematical equations at collocation points, and the data-based error i.e., the Root Mean Squared Error (RMSE) and/or the Mean Squared Error (MSE). The one or more top k unknown 5 parameters are iteratively identified by the inner iteration loop (iterk) in the range of the Tk times, if any of the one or more selected top m models does not meet the pre-defined threshold value for the data-based error (£data(0)), and the physics-based error (Lphysics((p)). The one or more new set of top m models are selected and continued by the outer iteration loop (iterm) until the range of the Tm times, if 10 any of the one or more selected top m models does not meet the pre-defined threshold value for the data-based error (£data(0)), and the physics-based error (£Physics(