Temperature profiling method of a gas turbine enclosure and system using such method

Abstract

A temperature profiling method for a machine or component in the industrial plant involves receiving bidimensional images from thermal cameras (31,32,33,34), converting these 2D frames into a 3D thermal model, and associating each surface point on the model with its coordinates and a temperature prediction based on a machine learning model (42). This method allows for real-time temperature monitoring via a user interface and video module (43), allowing operators to determine temperatures at specific points within the enclosure. The system includes thermal cameras (31,32,33,34) to acguire images of the machine or component in the industrial plant, generating 2D images with temperature signals for each pixel, and a processing unit to execute the method, providing real-time temperature data to the user interface for monitoring purposes.

Description

Temperature profiling method of a Gas Turbine Enclosure and System using such MethodDescriptionTECHNICAL FIELD

[0001] The present disclosure concerns temperature profiling method of a gas turbine enclosure and system using such method, wherein the method uses artificial intelligence. Particularly, the method is a non-contact method of temperature profiling based on artificial intelligence.BACKGROUND ART

[0002] As is well known, the gas turbines are complex turbomachines. Inside the gas turbines the chemical reactions generate high temperatures. The temperature profile within a gas turbine enclosure is an important information that enables the accurate design of instrument placement and the monitoring of their performance.

[0003] The temperature profile inside the gas turbine enclosure provides information about the effectiveness of ventilation system. It helps to monitor the operating ambient temperature of instrumentation system inside the gas turbine enclosure, which is vital for prevention of instrumentation system and associated cabling failure prevention.

[0004] A gas turbine comprises an enclosure. Access to the internal parts of the enclosure of a gas turbine during machine running condition is not feasible due to safety risks. Currently, the temperature inside the enclosure is measured indirectly using the temperature of the ventilation exhaust air. For investigation of any anomaly due to temperature inside the enclosure not limited to instrument failure, the machine needs to be shut down and operating personal has to enter after a cooldown period, which is established according to operation and maintenance (O&M) recommendations and procedures. The absence of direct continuous real time direct temperature measurement under machine running condition is hampering plant operation with increased downtime.

[0005] This indirect measurement approach and the mandatory shutdown lead to increased downtime and limited insight into the actual temperature conditions within the enclosure during operation.

[0006] Currently, this design relies on computational tools like CFD analysis to predict the temperature profile inside the enclosure. However, there is no realistic means to determine the actual temperatures during operation. This lack of accurate information reduces the ability to assess the performance or identify failures of the instrumentation used within the enclosure, creating a significant technical challenge in ensuring optimal functionality and reliability.

[0007] Accurately determining the temperature at any specific coordinate within a gas turbine enclosure would be useful for understanding the thermal distribution and identifying potential instrument, cable, or component failures caused by elevated ambient temperatures. This knowledge provides a more realistic assessment of the enclosure's temperature profile, offering a clearer understanding of how heat is distributed. Such precise information would enhance the safety of operating personnel entering the enclosure by ensuring thorough temperature verification before access, ultimately reducing potential hazards and improving maintenance decision-making.

[0008] The prior art patent application number US20230168146A1 discloses a method and apparatus for autonomously detecting thermal anomalies, primarily in the aerospace, power, transportation, and medical industries. The system comprises a plurality of infrared cameras that capture a baseline image set, including at least two thermal images. Emissivity data is generated based on the baseline image set and provided to an artificial intelligence model. This model generates a reconstructed image set. The system determines a difference between the baseline image set and the reconstructed image set, and generates an alert indicating detection of an engine anomaly when the difference exceeds a threshold.

[0009] The patent application number US8724976 discloses a real-time temperature monitoring and control using an infrared camera. The apparatus and method are designed to monitor and control the temperature of a substrate during processing. Theapparatus leverages an infrared camera to obtain temperature profiles of multiple regions or the entire surface of the substrate. This data is then used by a system controller to calculate and coordinate an optimized strategy in real time for reducing any potential temperature non-uniformity on the substrate during processing. The disclosure enhances the accuracy of temperature measurement and control, thereby improving the overall quality and efficiency of semiconductor manufacturing processes.

[0010] In light of the above, it is, therefore, an object of the present disclosure to provide a realistic means of determining the actual temperature experienced during gas turbine operation to accurately assess the performance of the instrumentation and promptly identify any failures.

[0011] Another object of the disclosure is to enable the precise placement of instruments based on accurate, real-time temperature data, thus improving monitoring and enhancing reliability.

[0012] It is also an object of the present disclosure to optimize the design process by offering accurate insights into the temperature profile, minimizing reliance on computational predictions alone, and ultimately leading to more efficient and effective instrumentation system design within the gas turbine enclosure.SUMMARY

[0013] In an aspect, the subject-matter disclosed herein is a temperature profiling method of a machine or component in an industrial plant, particularly within a gas turbine enclosure. This method comprises the steps of receiving bidimensional images of the machine or component from at least one thermal camera, converting these 2D frames into a three-dimensional thermal model, and associating each surface point of the 3D model with coordinates and a temperature prediction based on a machine learning model. The method also involves showing a thermal image via a user interface and video module to enable real-time temperature monitoring by an operator.

[0014] A further aspect of the present disclosure is drawn to a method wherein theconversion step is performed by an image processing machine learning model, specifically a convolutional neural network (CNN). This method includes a training phase for the image processing machine learning model.

[0015] In another aspect, disclosed herein is a training step that comprises providing an image processing algorithm for a 3D thermal processing model, training the machine learning model with images of the machine or component, and testing the trained model. If validation fails, further training is conducted; if successful, the model is applied.

[0016] In some aspects, the validation evaluates if the training of the CNN has arrived to an error below a certain threshold.

[0017] A further aspect of the present disclosure is drawn to a method wherein the training step involves comparing current thermal data against historical temperature distribution patterns within the machine or component.

[0018] This is a self-learning concept, which means that when convolutional neural network (CNN) sees a new feature it adds up to the training data both the feature and the predicted value.

[0019] In another aspect, disclosed herein is an association step that involves receiving the 3D thermal model, converting it to an array of grayscale pixel values, associating these values with relevant coordinates, generating output data as temperature at each surface point, and transmitting this data to a video module with a graphic user interface.

[0020] A further aspect of the present disclosure is drawn to a method that comprises establishing a relation between temperature and grayscale pixel values. This involves determining a dataset range of temperatures associated with pixel values and providing the best fit regression model to this dataset.

[0021] In another aspect, disclosed herein is a step for determining a selected dataset range, involving identifying a reference object, acquiring industrial temperature sensordata, connecting to the thermal camera, executing a pixel value extractor algorithm, and determining a dataset range of temperatures associated with pixel values.

[0022] A further aspect of the present disclosure is drawn to a method for providing the best fit regression model, which includes training multiple machine learning models, choosing the best fit model based on R2and RMSE parameters, and validating the selected model.

[0023] In another aspect, disclosed herein is a method that further includes notifying a user of unusual temperature fluctuations or potential equipment failures within the machine or component.

[0024] A further aspect of the present disclosure is drawn to a method wherein the thermal camera can be movable to acquire frames from multiple angles, and wherein the system comprises multiple thermal cameras positioned at defined viewing angles to cover the entire volume of the machine or component.

[0025] In another aspect, disclosed herein is a method wherein the machine or component is located inside an enclosure, specifically a gas turbine.

[0026] In another aspect, disclosed herein is a system for profiling the temperature within a gas turbine enclosure, comprising at least one thermal camera configured to acquire images and a processing unit connected to the camera. The processing unit includes a module configured to execute the temperature profiling method and a user interface to receive thermal image signals for real-time monitoring. A further aspect of the present disclosure is drawn to a system wherein the thermal cameras are movable and configured to be positioned at defined viewing angles, and the processing module includes a convolution neural network trained by conversion data sets of images taken by the thermal cameras.

[0027] In another aspect, disclosed herein is a system wherein the machine or component is located inside an enclosure, specifically a gas turbine.BRIEF DESCRIPTION OF THE DRAWINGS

[0028] A more complete appreciation of the disclosed embodiments of the disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:Fig. 1 illustrates a schematic of a temperature profiling system according to an embodiment;Fig. 2 illustrates a scheme, according to embodiment of Fig. 1;Fig. 3 illustrates an additional schematics of the operations of the microprocessor;Fig. 4 illustrates a flow charts of a method for profiling the temperature of the enclosure of a gas turbine according to the present disclosureFig. 5 illustrates a flow charts of a step of the method of Fig. 4; and Fig. 6 illustrates a flow charts of another step of the method of Fig. 4.DETAILED DESCRIPTION OF EMBODIMENTS

[0029] According to one aspect, the present subject matter is directed to a method for temperature profiling within gas turbine enclosures, using artificial intelligence to provide accurate real-time data. A thermal image-based Al network gathers data and generates a reliable temperature profile inside the gas turbine enclosure using four thermal cameras that comprehensively covers the enclosure volume at specific viewing angles. The data collected is processed through a processing unit, where the pixel grayscale values are mapped to the object's temperature and subsequently converted into temperature data via specific algorithms. The thermal image is then streamed in real time to the operator, enabling immediate monitoring of temperature variations. A machine learning algorithm continually refines its predictions to improve the accuracy of the prediction capability. This ensures safe and reliable operations, and reduces maintenance downtime. Additionally, operators can accurately measure the temperature inside the enclosure, including on machine surfaces, to detect failures in instrumentation, cables, or other components, allowing for realistic assessments and prompt corrective actions.

[0030] The system helps design engineers and users to validate instrument placements, monitor operational performance, and investigate system failures. It also provides flexible temperature readings at any point inside the enclosure using 3D coordinates or 2D frames.

[0031] Finally, by verifying actual temperatures during the cool-down phase, the system safeguards operating personnel entering the enclosure after shutdown, reducing potential risks. Operating personal has, therefore, a realistic means to verify actual temperature inside the enclosure of all hot components before entering into the enclosure, thus increasing the operating personal safety.

[0032] Referring now to the drawings, Figures 1, 2, and 3 show, respectively, a high level scheme of a profiling apparatus 1 and a technical scheme of a temperature profiling system 1 of an embodiment.

[0033] The temperature profiling system 1 is specifically designed for profiling the temperature within an enclosure 21 of a gas turbine 2, although it can be used in principle for other applications. The temperature profiling system 1 comprises a plurality of thermal cameras 3, a processing unit 4, for processing the signals acquired by the cameras 3, and a user interface 5.

[0034] However, the system 1 can be applied for the temperature monitoring of machine or a component in an industrial plant, like turbomachines in general.

[0035] The thermal cameras 3 are arranged at predetermined viewing angles to cover the entire enclosure volume of the enclosure 21. Each image captured by one of the cameras 3 is composed by a matrix of pixels. Each pixel is associated with a signal that represents the temperature value at the given point in the image. The thermal data composed of the pixel signals from the images reflect the thermal distribution of the enclosure 21 in two dimensions.

[0036] The cameras are configured to capture a comprehensive thermal profile within the gas turbine 2 enclosure 21.

[0037] In the present embodiment, the cameras are four, and are labeled by the reference numbers 31, 32, 33, and 34.

[0038] In some embodiments the system 1 comprises one or more cameras 3 that are movable. This movability allows each camera 3 to acquire frames from multiple angles, enhancing the coverage and flexibility of the monitoring process. This feature ensures that critical areas within the gas turbine enclosure 21 can be inspected from various perspectives, facilitating detailed analysis and maintenance.

[0039] The thermal camera 3 operates by detecting infrared radiation (heat) emitted from objects and converting it into an electronic signal that can be visualized as an image. As known, every object emits a certain amount of infrared radiation, depending on its temperature. Thermal cameras have sensors sensitive to this wavelength range, typically in the long-wave infrared region (8-14 fim). The camera's lens focuses the infrared radiation onto these sensors. A thermal camera is equipped with a thermographic sensor, which converts the infrared radiation into an electronic signal. These sensors are often based on microbolometers, which change resistance with temperature or use quantum well detectors sensitive to infrared light.

[0040] The electronic signals from the cameras 3 are processed and 2D matrix of pixel values are provided. Then, a dedicated module, with a Al Algorithm is used, to determine the temperature at each point in the scene. This data forms a temperature map or matrix, where each pixel corresponds to a specific temperature value in 2D.

[0041] In some embodiments, the signal is greyscale-based. Specifically, for each pixel, to a different degree of gray, corresponds a different temperature. However, in some other embodiments, the different temperature can be associated to specific colors.

[0042] The processing unit 4 is connected to each of the thermal cameras 31, 32, 33, and 34. The processing unit 4 comprises a processing module 41, a machine learning module 42, and a video module 43, having a graphic user interface (GUI).

[0043] The processing module 41 is configured to receive thermal data from the cameras and correlate each pixel signal, grayscale-based signals value, in the present embodiment, with the corresponding temperatures of the volume points and objects within the enclosure 21, as better explained below. The processing module 41 processes signals to build an accurate temperature profile.

[0044] The processing module 41 is usually implemented by a microprocessor. In some embodiment, the microprocessor with dedicated GPU which is programmable to handle image processing tasks and capable of running Convolution Neural Network (CNN).

[0045] Any commercially available multicore processor preferably with clocks speed of more than 3 GHz and GPU with a graphic memory of greater than 6GB can be used along with multiple USB connector provision for interface with camera and display unit is used to realize the concept

[0046] Other processors or architectures can be used to implement processing module 41.

[0047] The machine learning module 42 refines the temperature profile of the enclosure 21 by applying a prediction algorithm to the signals received from the thermal cameras 31, 32, 33, and 34. This prediction algorithm is executed by the processing module 41 and is optimized over time by analyzing historical data patterns and refining predictive capabilities.

[0048] Specifically, the processing unit 4 is based on three algorithms.

[0049] A first algorithm, better disclosed in the in Fig. 4, from a structural point of view uses a Convolution Neural network model (CNN), which is trained using conversion data set comprising of sufficient number of pictures of the enclosure captured using thermal camera placed at different angles. This conversion data set is used to train the CNN-based model and enables the CNN to perform feature extraction and reconstruct the 2D image in to a 3D image by predicting the depth value from 2D image, he CNN model does minimum following functions, by:a. receiving continuous feed of minimum four frames from four different cameras at any instant of time; b. identifying frames, frame stitching and depth information estimation; c. generating a 3D images; and d. the algorithm uses feature extraction, for self-learning. From the image set when the features are extracted, it compares with existing feature, if there new features found, it is added a new learning in the CNN memory.

[0050] Regarding the depth information estimation, 2D images are estimated in all directions as boundary conditions and moving from the interior toward the center. Since the pixel value of the 2D image of a surface is related to the temperature of the surface, while estimating the pixel value in 3D, through depth estimation, it is estimated the 3D temperature of the surface.

[0051] Alternatively, from the 2D thermal image, which will be the boundary condition, Physics informed Neural Network (PINN) can be trained. The Physics informed Neural Network (PINN) is trained on temperature of source of heat, thermal conductivity of all surrounding surface both metal and non-metal up to the boundary condition (The surface that is visible in thermal image). So, when from a 2D thermal image temperature is estimated, it can be provided to PINN to estimate the temperature for any hidden surface in the machine volume

[0052] The second algorithm, better disclosed in the in Fig. 5, basically carries out two main operations. According to one operation, it uses the output of the first algorithm which is a 3D image to covert it in a 3D array of pixels and assigns a temperature value to it using the operation of the third algorithm. Specifically, the second algorithm associates to the 3D array of grayscale values at each point of the 3D image of the enclosure 21 of a gas turbine 2, a temperature, by means of the map provided by the third algorithm. The first algorithm converts 2D to 3D is a CNN algorithm that has selflearning capability.

[0053] As a second operation, the second algorithm converts the 3D thermal image from the first algorithm, after being suitably processed, in to step file that can be loaded in to any 3D software commercially available for viewing by operator. The operator selection to view either 3D image or temperature values in different units will be managed by this module and corresponding output will be sent to operator for view.

[0054] Advantageously, once the 3D thermal image is created, operator has the possibility to view the 3D thermal image, 3D visual image, and array of pixel values. All these solutions are realizable by means of a backend, with user interface to enhance user experience.

[0055] Finally, the mentioned third algorithm is a Machine Learning (ML)-based algorithm that is trained using two source of physical data. In particular, ML-training data set comprise a temperature data set coming from a physical sensor applied to a hot body, namely a temperature source, and the data derived by thermal cameras of the same object, extracting pixel values, so that the ML-based third algorithm is trained to associate the temperature data to each extracted pixel of the images.

[0056] Therefore, the third algorithm establishes best fit relation between the thermal image pixel value and the actual temperature measured. Once the correlation is established the algorithm is able to predict the temperature outside the trained data set using the mathematical function established during ML training. The third ML algorithm once established is able to predict the temperature outside the trained data set using the correlation function established during the training.

[0057] The video module 43 converts, then, the thermal data processed by the machine learning module 42 into a thermal image, which is streamed to the user interface 5 to allow the operator or user to monitor the temperature at specific points within the enclosure 21 based on either three-dimensional or bidimensional coordinates. This feature facilitates real-time monitoring and helps operators visualize the temperature distribution inside the enclosure 21, ensuring immediate identification of potential issues.

[0058] As mentioned, the pixel signal value of each image acquired by the thermalcameras 31, 32, 33, and 34 is typically a grayscale signal, which the processing module 41 correlates with the corresponding temperatures at various volume points and objects inside the enclosure 21 thanks to the above mentioned third algorithm. This correlation is a mapping between the camera 3 images and the thermal distribution within the enclosure 21. This operation ensures that the temperature values are assigned to each pixel signal value.

[0059] The machine learning module 42 further comprises a training module 44, which trains the prediction algorithm to interpret temperature values from grayscale signals. The machine learning module 42 also comprises a thermal simulation model 45 of the thermal cameras to simulate the pixel signal values, and a temperature simulation model 46 to simulate the temperature values within the enclosure 21 of the gas turbine 2.

[0060] The models or blocks of Fig. 2 carry out different steps of the above mentioned algorithms. In other embodiments, the structure of the modules can be different.

[0061] In particular, with reference to the functional relationship of Fig.2, the spectrum of colour emission is function of the temperature of a hot body. The thermal cameras captures this spectrum of rays which is correlated to temperature of the hot surface. For training purpose, hot surface with known temperature is correlated to the thermal image pixel value, and ML algorithm is developed and trained.

[0062] As mentioned above, the ML-training data set comprises of the pixel value of the thermal image of temperature source and actual measured temperature of a heat source using an industrial sensor with predefined accuracy are data that are obtained from lab experiments. These simulation models feed such ML-training data into the training ML module 44 to refine the predictive capabilities of the machine learning module 42, improving the algorithm's ability to map temperature profiles when acquired in real time.

[0063] Pixel value to temperature conversion is not an Al algorithm but a ML algo-rithm that doesn't have self-learning capability. Once the algorithm is trained, it is capable of predicting temperature values from pixel outside the trained data set.

[0064] The processing module 41 executes the prediction algorithm using at least one convolutional neural network 421, applied for image processing, which is actually a machine learning model. The convolutional neural network 421 enables frame identification and stitching, feature extraction and depth mapping for converting between two-dimensional and three-dimensional data.

[0065] The processing module 42 reads the grayscale values from the thermal images generated by the training module 44 to establish relationships between the trained dataset's temperature and pixel grayscale values. This process helps the processing module 4 Ito assign temperature values based on the grayscale data.

[0066] The processing unit 4 also comprises a data storage unit 46, to store previous thermal data for use by the machine learning (ML) module 42. This ensures that the machine learning module 42 is continuously trained and improves the prediction accuracy along the time. By comparing current thermal data against historical temperature distribution patterns within the enclosure 21, the machine learning module 42 adapts its algorithms to identify unusual temperature fluctuations or equipment failures.

[0067] The processing module 41 may notify the operator or user U of these anomalies, improving safety and reducing maintenance costs.

[0068] The user interface 5 can be a display or a computer setup. It also provides flexible temperature readings at any point inside the enclosure using 3D coordinates or 2D frames. In particular, the user may pick a point of the volume of the enclosure and detect the temperature value.

[0069] The operations of the machine learning module 42, the training module 44, thermal simulation model 45, and the temperature simulation model 46 may be executed by the processing module 41. Such operations may be instructions of a computer program.

[0070] The temperature profiling system 1 operates as follows. In particular, the flowcharts of the above mentioned three algorithms and their coordination is disclosed.

[0071] Referring to Fig. 4 a flowchart of the temperature profiling method 100 of a gas turbine 2 enclosure 21 is shown. More specifically, Fig. 4 illustrates mainly the operating steps of the first algorithm. In fact, the temperature profiling method 100 comprises several steps, to generate an accurate temperature profile for monitoring purposes.

[0072] The temperature profiling method 100 comprises training a machine learning model 421 (i.e., the first algorithm) for processing data from at least one thermal camera 3 to achieve or receive two-dimensional (2D) thermal image data and process them as follows.

[0073] More specifically, this step involves receiving 120 bidimensional frames from the at least one thermal camera 3 and converting (step 130), by the image processing machine learning model (421), the bidimensional frames from the at least one thermal camera 3 into a three-dimensional (3D) thermal model.

[0074] In some embodiments the camera can be more than one, e.g., four. In addition, as mentioned above, in some embodiments the camera 3 can be movable, to take more than one frame in a short time interval.

[0075] On each surface point of the three-dimensional (3D) thermal model, the method 100 associates (step 140) the coordinate of the point, e.g., referred to a Cartesian frame taken from a reference point, and the relevant detected temperature.

[0076] The training step 110 further comprises providing 111 the algorithm of the image processing machine learning model 421 for a three-dimensional (3D) thermal processing model configured to process the two-dimensional (2D) thermal image data taken by the cameras 3.

[0077] The image machine learning model 421 is trained (step 112), as mentioned above, by the conversion data set, followed by testing (step 113) the trained machinelearning model 421. If the validation of the trained machine learning model 421 fails, the training step 112 is executed again; else, if the machine learning model 421 is successful, it is then applied.

[0078] The machine learning model 421 comprises a convolutional neural network (CNN).

[0079] In addition to convolutional neural networks (CNN), other machine learning models can be utilized for the temperature profiling method. These alternatives include recurrent neural networks (RNN), which are suitable for sequential data processing and can handle temporal dependencies in thermal data. Another option is support vector machines (SVM), which are used for classification and regression tasks with highdimensional spaces. Decision trees and random forests can also be employed for handling diverse data types and their ability to model complex decision boundaries. Additionally, k-nearest neighbors (k-NN) algorithms can be considered for their simplicity and efficiency in classifying temperature profiles based on proximity to known data points.

[0080] The conversion data set is based on images of the gas turbine 2 enclosure 21 taken before and in real time of the conversion process.

[0081] The association step 140 (see Fig. 5) comprises the sub-steps of receiving 141 the three-dimensional thermal model as generated by the machine learning module 431, and converting in step 142 this three-dimensional thermal model to an array of grayscale pixel values of the surface points of the images acquired.

[0082] This steps ease the processing and the association of the data to the simulated temperatures.

[0083] Subsequently, the three-dimensional array of grayscale pixel values of the surface points is associated (see step 143) with the relevant coordinates of the points of the surfaces of the images acquired. This generates (step 144) output data as the temperature at each surface point of the coordinate in three-dimensional space of the volume of the gas turbine enclosure 21 to be monitored.

[0084] The output data is then transmitted in step 145 to a video module 43, which includes a graphic user interface (GUI), whereby the user can see on a monitor the temperature distribution into the gas turbine enclosure 21 to be monitored. To ensure such association, the method 100 comprises establishing the relation between the temperature and each grayscale pixel value (step 150), allowing the association of the three-dimensional (3D) array of grayscale values at different coordinates on the surface of the images acquired

[0085] The sub-steps of the establishing step 150 is shown in the flowchart of Fig. 6. The establishing step 150 comprises the sub-steps of determining 151 a selected dataset range of temperatures associated with the pixel value and providing (see step 152) the best fit regression model to this selected dataset range of temperatures.

[0086] Determining the dataset range 151 involves identifying a reference object 1511, such as a thermal profile within the gas turbine 2 enclosure 21, and acquiring industrial temperature sensor data 1512 of the reference object. This step also comprises connecting to the at least one thermal camera 3, carrying out a pixel value extractor algorithm 1514, and determining a dataset range of temperatures associated with pixel values 1515.

[0087] Providing the best fit regression model 152 comprises training regressionbased multiple machine learning models 1521 and choosing the best fit model based on the R2value and Root-Mean-Square Deviation (RMSE) parameters 1522. These are statistical methods for achieving the best fit.

[0088] The validity of the selected machine learning model is then checked in step 1523; if the selected model is not validated, the training step 1521 is repeated, otherwise, the best fit regression model is identified.ADVANTAGES

[0089] An advantage of the present disclosure is that it offers a more accurate method for operators to determine the temperature within the enclosures of the gas turbines, including the temperature of machine surfaces, ensuring better monitoring, enhancedsafe and reliable operation and control over the equipment.

[0090] Another advantage of the present disclosure is that it allows the measurement of temperatures in the surroundings where instrument cables or other components are installed, which enables to ensure that instruments and associated accessories ambient temperature is within the manufacturer suggested ambient temperature range, enabling a more realistic identification of any failures induced by higher operating ambient temperature. This leads to prompt corrective actions, reducing downtime and improving reliability.

[0091] It is also an advantage of the present disclosure that it ensures the operator's safety before entering the enclosure after a machine trip, as the system verifies that temperature levels have safely decreased during the cool-down period, mitigating any potential risks.

[0092] It is noted that the system can be installed in any enclosed, un-accessible, confined space temperature profiling and thermal inspection area. As interlocks can be placed in enclosure door opening if high temperature is predicted anywhere inside the enclosure of the gas turbine.

[0093] While aspects of the disclosure have been described in terms of various specific embodiments, it will be apparent to those of ordinary skill in the art that many modifications, changes, and omissions are possible without departing form the spirt and scope of the claims. In addition, unless specified otherwise herein, the order or sequence of any process or method steps may be varied or re-sequenced according to alternative embodiments.

[0094] Reference has been made in detail to embodiments of the disclosure, one or more examples of which are illustrated in the drawings. Each example is provided by way of explanation of the disclosure, not limitation of the disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made in the present disclosure without departing from the scope or spirit of the disclosure. Reference throughout the specification to "one embodiment" or "an embodiment"or “some embodiments” means that the particular feature, structure or characteristic described in connection with an embodiment is included in at least one embodiment of the subject matter disclosed. Thus, the appearance of the phrase "in one embodiment" or "in an embodiment" or "in some embodiments" in various places throughout the specification is not necessarily referring to the same embodiment(s). Further, the particular features, structures or characteristics may be combined in any suitable manner in one or more embodiments.

[0095] When elements of various embodiments are introduced, the articles “a”, “an”, “the”, and “said” are intended to mean that there are one or more of the elements. The terms “comprising”, “including”, and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements.

[0096] The subject matter described herein can be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structural means disclosed in this specification and structural equivalents thereof, or in combinations of them. The subject matter described herein can be implemented as one or more computer program products, such as one or more computer programs tangibly embodied in an information carrier (e.g., in a machine-readable storage device), or embodied in a propagated signal, for execution by, or to control the operation of, data processing apparatus (e.g., a programmable processor, a computer, or multiple computers). A computer program (also known as a program, software, software application, or code) can be written in any form of programming language, including compiled or interpreted languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file. A program can be stored in a portion of a file that holds other programs or data, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers at one site or distributed across multiple sites and interconnected by a communication network.

[0097] The processes and logic flows described in this specification, including the method steps of the subject matter described herein, can be performed by one or more programmable processors executing one or more computer programs to perform functions of the subject matter described herein by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus of the subject matter described herein can be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).

[0098] Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read-only memory, or a random access memory, or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto-optical disks, or optical disks. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including by way of example semiconductor memory devices, (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks, (e.g., internal hard disks or removable disks); magneto-optical disks; and optical disks (e.g., CD and DVD disks). The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.

[0099] To provide for interaction with a user, the subject matter described herein can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, (e.g., a mouse or a trackball), by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well. For example, feedback provided to the user can be any form of sensory feedback, (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user can be received in any form, including acoustic, speech,or tactile input.

[0100] The techniques described herein can be implemented using one or more modules. As used herein, the term “module” refers to computing software, firmware, hardware, and / or various combinations thereof. At a minimum, however, modules are not to be interpreted as software that is not implemented on hardware, firmware, or recorded on a non-transitory processor readable recordable storage medium (i.e., modules are not software per se). Indeed “module” is to be interpreted to always include at least some physical, non-transitory hardware such as a part of a processor or computer. Two different modules can share the same physical hardware (e.g., two different modules can use the same processor and network interface). The modules described herein can be combined, integrated, separated, and / or duplicated to support various applications. Also, a function described herein as being performed at a particular module can be performed at one or more other modules and / or by one or more other devices instead of or in addition to the function performed at the particular module. Further, the modules can be implemented across multiple devices and / or other components local or remote to one another. Additionally, the modules can be moved from one device and added to another device, and / or can be included in both devices.

[0101] The subj ect matter described herein can be implemented in a computing system that includes a back-end component (e.g., a data server), a middleware component (e.g., an application server), or a front-end component (e.g., a client computer having a graphical user interface or a web browser through which a user can interact with an implementation of the subject matter described herein), or any combination of such back-end, middleware, and front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.

Claims

Temperature profiling method of a Gas Turbine Enclosure and System using such MethodCLAIMS1. A temperature profiling method (100) of a machine or a component in an industrial plant, comprising the following steps: receiving (120) bidimensional images of the machine or component in the industrial plant from at least one thermal camera (3); converting (130) the bidimensional (2D) frames of said bidimensional images into a three-dimensional (3D) thermal model; associating (140) on each surface point of the three-dimensional (3D) thermal model the coordinates of the point and a temperature prediction based on a machine learning model, said machine learning model being adapted to predict a temperature of each point of the three-dimensional (3D) thermal model of the machine or component in the industrial plant; and showing by a user interface (5) and a video module (43) a thermal image derived from the bidimensional images of the machine or component in the industrial plant, to allow an operator to determine the temperature at specific points within the machine or component in the industrial plant, for real-time monitoring.

2. The method (100) of claim 1, wherein the conversion step (130) is carried out by an image processing machine learning model (421), wherein the image processing machine learning model (421) is a convolutional neural network (CNN), and wherein the method (100) comprises a training (110) step for training the image processing machine learning model (421).

3. The method (100) of the preceding claim, wherein the training (110) step comprises the following sub-steps:providing (111) an image processing machine learning model algorithm for three-dimensional (3D) thermal processing model configured to process the two-dimensional (2D) thermal image data; training the image machine learning model (421) by means of the images of the machine or component in the industrial plant; and testing (113) the trained machine learning model (421), such that if the validation fails, the training (112) step is executed, else if the machine learning model (421) is successful, the machine learning model (421) is applied.

4. The method of the preceding claim, wherein the machine training (110) step comprises the sub-step of comparing current thermal data against historical patterns of temperature distribution within the machine or component in the industrial plant.

5. The method (100) of any one of the preceding claims, wherein the association (140) step comprises the sub-steps of: receiving (141) the three-dimensional thermal model; converting (142) the three-dimensional thermal model to an array of grayscale pixel value of the surface points of the images acquired; associating (143) the three-dimensional array of grayscale pixel value of the surface points at relevant coordinates of the points of the surfaces of the images acquired; and generating (144) output data as temperature at each surface point coordinates in three-dimensional space; transmitting (145) the output data of the generating (144) step to a video module (43), having a graphic user interface (GUI); and wherein the method (100) also comprises the step of establishing (150) the relation between the temperature and each grayscale pixel value, wherein said establishing (150) step is executed for allowing the association (143) step of the three-dimensional array of grayscale values at different coordinates on the surface of the images acquired.

6. The method (100) of claim 5, wherein the three-dimensional thermal model received in the receiving (141) sub-step is carried out by a machine learning model.

7. The method (100) of any one of claims 5 or 6, wherein the relation establishment (150) step comprises the following sub-steps: determining (151) a selected dataset range of temperatures associated to the pixel value; and providing a best fit the regression model (152) to the selected dataset range of temperatures associated to the pixel value.

8. The method (100) of claim 7, wherein the determining (151) step comprises the following sub-steps: determining the reference object (1511), such as thermal profile within the machine or component in the industrial plant; acquiring the industrial temperature sensor data (1512) of the reference object; connecting (1513) to the at least one thermal camera (3); carrying out a pixel value extractor algorithm (1514); and determining a dataset range of temperatures associated to pixel values (1515).

9. The method (100) of any one of claims 7 or 8, wherein the step of providing the best fit the regression model (152) comprises the sub-steps: training the regression based multiple machine learning models (1521); choosing (1522) the best fit model based on the R2value and Root-Mean- Square Deviation (RMSE) parameters; checking if the selected machine learning model is valid (1523), so that if the selected machine learning model is not validated, go to the training the regression based multiple machine learning models (1521) step, else the best fit regression model is identified.

10. The method (100) of any one of the preceding claims, further comprising the step of notifying a user (U) of unusual temperature fluctuations or potential equipment failures within the machine or component in the industrial plant.

11. The method (100) of any one of the preceding claims, wherein the at least one camera (3) can be movable, to acquire frames from multiple angles.

12. The method (100) of any one of the preceding claims, wherein the at least one camera (3) comprises a plurality of thermal cameras (3) configured to be positioned at defined viewing angles to cover the entire volume of the machine or component in the industrial plant.

13. The method (100) of any one of the preceding claims, wherein the thermal cameras (31, 32, 33, 34) are four.

14. The method (100) of any one of the preceding claims, wherein the thermal cameras (31, 32, 33, 34) are installed inside an enclosure (21).

15. The method (100) of any one of the preceding claims, wherein machine or a component in an industrial plant is located inside an enclosure (21).

16. The method (100) of the preceding claim, wherein the machine is a turbomachine such as a gas-turbine (2).

17. A system (1) for profiling the temperature of a machine or component in the industrial plant, comprising: at least one thermal camera (3) configured to acquire images of a volume of the machine or component in the industrial plant to be monitored, wherein for each image acquired by the at least one camera (3), and to generate a two-dimensional image composed by a plurality of pixels, wherein to each pixel is associated a respectivesignal representative of the temperature value of the images acquired of the internal surface of the volume of the machine or component in the industrial plant; and a processing unit (4), connected to the at least one thermal camera (3), comprising a processing module (41), configured to execute the method of any one of claims 1-16; a user interface (5), connected to the processing unit (4), configured to receive the thermal image signal to allow an operator to determine the temperature at specific points within the machine or component in the industrial plant, for real-time monitoring.

18. The system (1) of the preceding claim, wherein the at least one camera (3) is movable, to acquire frames from multiple angles.

19. The system (1) of any one of claims 17 or 18, comprising a plurality of thermal cameras (3) configured to be positioned at defined viewing angles to cover the entire volume of the machine or component in the industrial plant.

20. The system (1) of any one of claims 17-19, wherein the thermal cameras (31, 32, 33, 34) are four.

21. The system (1) of any one of claims 17-20, wherein the thermal cameras (31, 32, 33, 34) are installed inside an enclosure (21).

22. The system (1) of any one of claims 17-21, wherein the processing module (41) comprises a convolution neural network (CNN) trained by conversion data set that comprise pictures of the machine or component in the industrial plant taken by the at least one thermal camera (3).

23. The system (1) of any one of claims 17-22, wherein machine or a component in an industrial plant is located inside an enclosure (21).

24. The system (1) of the preceding claim, wherein the machine is a tur- bomachine such as a gas-turbine (2).

Images