Description:Complete Specification:
The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed

Field of the invention
[0001] The present disclosure relates to the field of digital twin and predictive diagnostics. In particular the present invention discloses a method of training an AI model to determine working parameters of an industrial asset and a corresponding control unit thereof.

Background of the invention
[0002] A digital twin is a digital representation of a physical object, person, or process, contextualized in a digital version of its environment and is updated from real-time data. Digital Twins use simulation models based in numerical analysis. Numerical analysis has taken an edge in simulating the real time complex phenomenon or physics. Enhancement in high power computing facility has helped in getting a data for these numerical simulations in very shorter time. Advanced numerical modeling techniques have also brought down the numerical prediction error to the greater extent.

[0003] Sensors play an important role in understanding health status or in predictive maintenance of any industrial asset or process. However Physical sensors have their limitations. They cannot be placed at all desired locations because of plant constraints and hazardous operating conditions. They have their limitation in measuring the complex and require regular calibration and maintenance which involves high operational cost.

[0004] In order to overcome these limitations, one can look for an alternative called virtual sensors. These virtual sensor measure working parameters of an industrial asset by leveraging the advantages of advanced AI algorithms and utilizing simulation data from numerical analysis. Virtual sensors are built based on the data from advanced numerical analysis-based simulation models and trained by advanced artificial intelligence.

[0005] Patent Document EP3525055 A1 titled “Diagnostic system with a virtual sensor” discloses a first real sensor (10) for detecting a plurality of first real sensor values (12), at least one further real sensor (20) for capturing a plurality of further real sensor values (22), and at least one virtual sensor (30) for the at least one first real sensor (10) for the purpose of diagnosing the at least one first real sensor (10) by determining a plurality of virtual sensor values (32) on the basis of a simulation on the basis of the plurality of further real sensor values (22); comparing the plurality of virtual sensor values (32) with the plurality of first real sensor values (12), and detecting a deviation between the plurality of virtual sensor values (32) and the plurality of first real sensor values (12). Furthermore, the invention relates to a corresponding virtual sensor, a corresponding method and a corresponding computer program product.

Brief description of the accompanying drawings
[0006] An embodiment of the invention is described with reference to the following accompanying drawings:
[0007] Figure 1 depicts a framework for training AI model (M) to determine working parameters of an industrial asset (102);
[0008] Figure 2 illustrates method steps to train the AI model (M) to determine working parameters of the industrial asset (102).

Detailed description of the drawings
[0009] Figure 1 depicts a framework for training AI model (M) to determine working parameters of an industrial asset (102). The framework comprises the industrial asset (102), a control unit (103), a simulation model (104) and at least an AI model (M). The control unit (103) is in communication with a control unit (103) and at least an AI model (M). The industrial asset (102) is a machinery placed in an industrial facility. The industrial asset (102) has a plurality of input (s) such as electrical, hydraulic and mechanical. The industrial asset (102) has a plurality of sensors that measure the operating conditions of the asset (102). For example: atmospheric pressure, ambient temperature, humidity etc.

[0010] The industrial asset (102) has a well-defined simulation model (104) for extensive range of input signals for all possible operating conditions. The simulation model (104)s are built using numerical analysis. Numerical analysis use Physics based mathematical modelling or 1D simulation enables to build digital twins of actual industrial asset (102) and analyze overall performance of the asset (102). Physics based Numerical analysis are used to build the entire industrial asset (102) with bottom-up approach. This involves discretizing the asset (102) into multiple blocks without compromising the fidelity of the systems inside the asset (102) and connecting them in orderly manner. Asset (102) is validated and finetuned against measurement data available from test bench.

[0011] The control unit (103) is logic circuitry and software programs that respond to and processes logical instructions to get a meaningful result. It may be implemented in the system as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, one or more microchips or integrated circuits interconnected using a parent board, hardwired logic, software stored by a memory device and executed by a microprocessor, firmware, an application specific integrated circuit (ASIC), and/or a field programmable gate array (FPGA), and/or any component that operates on signals based on operational instructions. The control unit (103) is in communication with an AI model (M).

[0012] An AI model (M) with reference to this disclosure can be explained as a component which runs a model. A model can be defined as reference or an inference set of data, which uses different forms of correlation matrices. Using these models and the data from these models, correlations can be established between different types of data to arrive at some logical understanding of the data. A person skilled in the art would be aware of the different types of AI model (M)s such as linear regression, naïve bayes classifier, support vector machine, neural networks and the like.

[0013] Some of the typical tasks performed by AI model (M)s are classification, clustering, regression etc. Majority of classification tasks depend upon labeled datasets; that is, the data sets are labelled manually in order for a neural network to learn the correlation between labels and data. This is known as supervised learning. Some of the typical applications of classifications are: face recognition, object identification, gesture recognition, voice recognition etc. Clustering or grouping is the detection of similarities in the inputs. The cluster learning techniques do not require labels to detect similarities. Learning without labels is called unsupervised learning. The AI model (M) is trained using modelled value of working parameters received from a simulation model (104) of the industrial asset (102) trained in a supervised manner in accordance with method steps 200.

[0014] The control unit (103) is adapted to determine working parameters of the industrial asset (102). The control unit (103) is configured to: receive a plurality of input signals fed to the industrial asset (102); receive measured values of a plurality of operating conditions of the industrial asset (102) from the plurality of sensors; feed the plurality of input signals and the plurality of operating conditions as inputs to the trained AI model (M); executing the trained AI model (M) to obtain working parameters of the industrial asset (102). The input signals comprise electrical, hydraulic and mechanical signals. The operating conditions refer to the external environmental operating conditions of the industrial asset (102). The control unit (103) is configured to perform method steps in accordance with figure 2.

[0015] It should be understood at the outset that, although exemplary embodiments are illustrated in the figures and described below, the present disclosure should in no way be limited to the exemplary implementations and techniques illustrated in the drawings and described below.

[0016] Figure 2 illustrates method steps to train an AI model (M) to determine working parameters of an industrial asset (102). The industrial asset (102) and the framework used to train the AI model (M) for the industrial asset (102) has been elaborated in accordance with figure 1. The method steps are explained with an exemplary embodiment of the present invention, where the industrial asset (102) is a pump and the AI model (M) is a neural network. For the pump the working parameters that need to be determined are but not limited to cavitation volume, cavitation number, Net Positive Suction Head.

[0017] Method step 201 comprises recording a plurality of input signals fed to the industrial asset (102). Input signals comprise electrical, hydraulic and mechanical signals. For a pump they would only be electrical signal i.e. the current and voltage. Other derivative parameters may be extracted by processing these raw current and voltage signals.

[0018] Method step 202 comprises measuring a plurality of operating conditions of the industrial asset (102) by means of the plurality of sensors. The operating conditions here refer to the external environmental operating conditions of the industrial asset (102) for example atmospheric pressure, ambient temperature and the like.

[0019] Method step 203 comprises processing the input signals and operating conditions by the AI model (M) to determine working parameters. In this method step the AI model (M) is executed with the input signals to obtain the output. The output working parameter obtained in the first iterations will not be accurate.

[0020] Method step 204 comprises defining a loss function for the AI model (M) based on modelled value of working parameters received from the simulation model (104) and the determined working parameters. The modelled value of working parameters act as ground truth and assist the supervised learning for the AI model (M). The loss function is defined as the difference between modelled value of working parameters vis-à-vis the working parameters obtained by executing the AI model (M).

[0021] Method step 205 comprises minimizing the loss function to train the AI model (M). Minimizing the loss function comprises optimizing the network parameters and hyperparameters of the AI model (M) i.e. a neural network. Multiple iterations for method steps 203, 204, 205 are performed until the value of working parameters obtained by executing the AI model (M) is same or close to the modelled value of working parameters. This means that the neural network has learnt to correlate a value of the working parameters for specific input signals in the specific operating conditions.

[0022] Neural networks are inspired by the biological neural network or brain cell i.e. neurons. The network parameters include but are not limited to a layers, filter and the like. For simplicity, in computer science, a network of neurons are represented as a set of layers. These layers are categorized into three classes which are input, hidden, and output. Every network has a single input layer and a single output layer. Different layers perform different kinds of transformations/operations on their inputs. Data flows through the network starting at the input layer and moving through the hidden layers until the output layer is reached. Hyperparameter is a parameter whose value is used to control the learning process. While networks parameters are learned during the training stage, hyper parameters are given/chosen. Hyper parameters are typically characterized by the learning rate, learning pattern and the batch size. In method step 205 we basically tune or configure the hyperparameters and network parameters until we reach a minimal loss function.

[0023] Once trained, the AI model (M) is configured to receive real-time data from sensors and input signals and give the desired working parameter as the output. Although the method is explained with an exemplary model of a pump, the same technique can be applied for any working parameter of any industrial asset (102). For example in an alternate embodiment of the present invention , the asset cold be a motor wherein the input signals would be current, voltage, power factor and the like. The working parameter that needs to determined could be flow rate etc.

[0024] The proposed methodology and framework of the present invention minimizes dependency on physical sensors for each and every working parameter, thereby reducing cost and increased efficiency. The proposed methodology to determine working parameters turns out like a virtual sensor for measuring these working parameters. The virtual sensors improve reliability, reduce maintenance costs, and enhance asset (102) performance. For example, in case of the pump the working parameters cavitation volume, cavitation number, Net Positive Suction Head are indicative of health of the pump. This helps in predictive maintenance and diagnostics.

[0025] A person skilled in the art will appreciate that while these method steps describes only a series of steps to accomplish the objectives, these methodologies may be implemented with modification and customizations to the framework. It must be understood that the embodiments explained in the above detailed description are only illustrative and do not limit the scope of this invention. Any ancillary modification to the method of training AI model (M) to determine working parameters of industrial asset (102)s and the framework thereof is envisaged. The scope of this invention is limited only by the claims.
, Claims:We Claim:

1. A method (200) of training an AI model (M) to determine working parameters of an industrial asset (102), the industrial asset (102) having a well-defined simulation model (104) for extensive range of input signals for all possible operating conditions, the method comprising;
Recording (201) a plurality of input signals fed to the industrial asset (102);
Measuring (202) a plurality of operating conditions of the industrial asset (102) by means of a plurality of sensors;
processing (203) the input signals and operating conditions by the AI model (M) to determine working parameters;
defining (204) a loss function for the AI model (M) based on modelled value of working parameters received from the simulation model (104) and the determined working parameters;
minimizing (205) the loss function to train the AI model (M).

2. The method (200) of training an AI model (M) to determine working parameters as claimed in claim 1 wherein the input signals comprise electrical, hydraulic and mechanical signals.

3. The method (200) of training an AI model (M) to determine working parameters as claimed in claim 1 wherein operating conditions refer to the external environmental operating conditions of the industrial asset (102).

4. The method (200) of training an AI model (M) to determine working parameters as claimed in claim 1 wherein minimizing the loss function comprises optimizing the network parameters and hyperparameters of the AI model (M).

5. A control unit (103) adapted to determine working parameters of an industrial asset (102), the control unit (103) in communication with a trained AI model (M), the control unit (103) configured to:
receive a plurality of input signals fed to the industrial asset (102);
receive measured values of a plurality of operating conditions of the industrial asset (102) from a plurality of sensors;
feed the plurality of input signals and the plurality of operating conditions as inputs to the trained AI model (M);
executing the trained AI model (M) to obtain working parameters of the industrial asset (102).

6. The control unit (103) adapted to determine working parameters of an industrial asset (102) as claimed in claim 5, wherein input signals comprise electrical, hydraulic and mechanical signals.

7. The control unit (103) adapted to determine working parameters of an industrial asset (102) as claimed in claim 5, wherein operating conditions refer to the external environmental operating conditions of the industrial asset (102).

8. The control unit (103) adapted to determine working parameters of an industrial asset (102) as claimed in claim 5, wherein the trained AI model (M) is trained using modelled value of working parameters received from a simulation model (104) of the industrial asset (102).