Detection system and method of operation thereof

Abstract

A system for detecting a gas, such as hydrogen. The system comprises a gas sensor having at least one chemical color-changing pigment capable of changing its color upon contact with the gas to be detected. The system further comprises a camera for detecting the color chemical color-changing pigment. The gas sensor is functionally coupled to a control logic unit equipped with an artificial intelligence-based algorithm residing in a processor trained for recognizing color changes of the chemical color-changing pigment of the sensor. A method for detecting a computer-implemented method for detecting a gas leak is also disclosed.

Description

[0001] manual Technical Field

[0002] This disclosure relates to a system for detecting gases such as hydrogen. This disclosure also relates to an operating system and a method for visualizing potential leaks. Background Technology

[0003] Modern factories typically employ complex systems to achieve optimal functionality. A fundamental component used in some of these processes is hydrogen (H2). As an energy carrier and reactive gas, hydrogen can pose significant risks when leaked, especially in confined spaces. Leaks not only reduce process efficiency but, more critically, can create serious safety hazards. Therefore, timely detection of any H2 leaks is essential to ensuring the safe and efficient operation of machinery and the plant.

[0004] The primary challenge in current systems is detecting the precise location of hydrogen leaks. The physical properties of hydrogen, such as its small molecular size and low viscosity, allow it to diffuse rapidly, making leak detection challenging. Furthermore, areas where leaks may occur are often located deep within machinery or plant infrastructure, making them difficult to access. Consequently, human operators find it nearly impossible to approach these locations and directly identify the source of the leak, leaving significant gaps in maintenance and safety protocols. Moreover, given the duration of plant downtime, locating the exact site of a hydrogen leak is time-consuming and expensive.

[0005] In fact, existing methods and technologies for leak detection primarily focus on identifying the presence of leaks, but are insufficient in detecting their exact location and / or determining their quantity. The inability to locate the leak source and quantify it in real time can delay corrective action, leading to prolonged downtime (as mentioned above), operational inefficiencies, and potential safety incidents.

[0006] Therefore, there is an urgent need for an innovative solution that not only detects the presence of hydrogen leaks but also identifies their exact location and / or quantity, enabling immediate and targeted corrective action. This solution should be further integrated with control systems to autonomously trigger safety measures, ensuring the system transitions to a safe state without manual intervention. Furthermore, the need for effective detection mechanisms / systems is not limited to hydrogen. Other gases used in machines and plants similarly require robust leak detection and location systems. For example, gases such as methane (CH4), ammonia (NH3), and carbon dioxide (CO2) are indispensable in various industrial sectors. Accidental or other releases of these gases can lead to environmental damage, operational disruptions, and, in some cases, serious safety hazards. Therefore, as with hydrogen, the rapid and accurate resolution of leaks in these gases (especially in their identification, location, and / or quantification) is crucial. Thus, for a variety of gases across different industrial landscapes, the need for sophisticated and reliable detection systems is widely recognized.

[0007] Related prior art also includes patent applications US 2023 / 177726 A1, US 2020 / 124579 A1 and US2022 / 307981 A1. Summary of the Invention

[0008] In one aspect, the subject matter disclosed herein relates to a leak detection system for detecting gas leaks from a factory or machine to be monitored. The system includes one or more gas sensors adapted to corresponding parts of the factory or machine for detecting potential gas leaks. Each gas sensor contains at least one chemichromic pigment that, in the event of a leak, changes color upon contact with the gas to be detected. The system also includes one or more image acquisition detectors, each arranged to acquire an image of a region of interest where a part of the factory or machine to which the sensor is applied is located. Each image acquisition detector is adapted to generate a signal corresponding to the region of interest. A control logic unit operatively connected to the image acquisition detectors is configured to receive and process the signal using an artificial intelligence-based algorithm trained to identify color changes of the chemichromic pigment in the event of a gas leak and generate a safety signal indicating the area where the gas leak has occurred.

[0009] Another aspect of this disclosure relates to a gas detection strip made of chemically color-changing pigments, which can be applied to mechanical parts or surfaces of a plant or machine to be monitored. The image acquisition detector includes one or more cameras, each adapted to detect one or more gas sensors.

[0010] Another aspect of this disclosure relates to one or more gas sensors (3) comprising at least one gas detection strip made of a chemically color-changing pigment and suitable for use in a plant or machine, located within a housing of one or more corresponding parts (P) of the machine. In a further embodiment, the sensor may also be located within a mechanical part or surface, thereby allowing the acquisition of signals within the mechanical part or surface. In another embodiment, one or more cameras (4) are located within the housing of the machine.

[0011] On the other hand, this paper discloses a system in which the control logic unit and the image acquisition detector communicate via, for example, Bluetooth. ® Wireless system connections such as Wi-Fi, infrared, and radio frequency are used. The gas to be detected in the system is hydrogen (H2). The artificial intelligence-based algorithm is implemented by a neural network, preferably a convolutional neural network (CNN).

[0012] On the other hand, the subject matter disclosed herein relates to a system in which an artificial intelligence algorithm includes a discriminator neural network. This network is configured to predict whether an input image acquired by an image acquisition detector was captured under specific aging and environmental conditions in a leak-free state, and to identify signs of gas leaks.

[0013] On the other hand, this paper discloses a system in which an artificial intelligence algorithm includes a generator neural network. This network is capable of generating realistic images of pigments under specific environmental conditions in the absence of gas leaks. The images generated by the generator neural network are used to train a discriminator neural network.

[0014] Another aspect of this disclosure relates to a control logic unit operatively connected to a safety module arranged on a gas supply line of a plant or machine to be monitored. Safety signals are used to control the safety module of the plant or machine to interrupt the supply of gas through the gas supply line. The control logic unit includes: a database storing signals generated by optical detectors; and a display for signaling possible gas leaks and displaying the location of such leaks within the plant or machine to be monitored.

[0015] On the other hand, this paper discloses a computer-implemented method for detecting gas leaks, comprising a control logic unit having a processor and a memory storing a computer program implementing an artificial intelligence-based algorithm. The computer program includes instructions for: determining the correct acquisition of a generated signal; rectifying the signal; processing the signal to detect a color difference in a chemically chromogenic pigment of at least one gas sensor; and generating a safety signal if the chemically chromogenic pigment of the gas sensor changes color due to a gas leak.

[0016] On the other hand, the method includes obtaining a 3D model of the piping system and connection locations (such as critical welded joints, threaded connections, fittings, accessories, and connections between gas supply pipes, manifolds, valves, seals, joints, vents, and orifices); determining strip locations in the 3D environment; determining strip lighting conditions; and determining the optimal location for the image acquisition detector. Signal processing steps include: detecting noise or environmental constraints; determining a baseline by noise removal; determining operational constraints for the region of interest; and determining operational thresholds that define the color difference threshold between the expected and actual values ​​for various alarm levels.

[0017] Another aspect of the invention relates to adjusting a safety signal based on the possible refraction of an image of a part of a factory or machine on which a sensor is applied, which includes the steps of receiving another optical signal by another optical sensor to detect different refractions and correlate the refraction values ​​with gas leaks.

[0018] On the other hand, the determination step or signal processing may include: determining the spectrum of pixels in an image acquired by a camera, and analyzing the spectral data of each pixel and comparing it with the spectrum of pixels within the same image or from a previous image to identify spectral differences to determine changes in the physical or chemical properties of the observed part in order to determine a gas leak.

[0019] Another aspect of this disclosure relates to modulating a safety signal by having one or more sensors receive signals for detecting different refractions and correlating the refraction values ​​with a gas leak, wherein such sensors are located on or within a mechanical part or surface, thereby allowing the acquisition of signals on or within the mechanical part or surface. Such positioning of one or more sensors on or within a mechanical part or surface can be initiated based on noise or environmental constraints (such as camera vibrations that blur the image, or factors such as radiation, diffraction, reflection, refraction, etc., over a region). In another aspect, this disclosure relates to modulating a safety signal by having one or more cameras located outside or inside a housing detect the signal. Such one or more cameras may be selected and located outside or inside the housing, depending on noise or environmental constraints (such as objects covering the area, reflections, etc.). Attached Figure Description

[0020] When considered in conjunction with the accompanying drawings, the embodiments disclosed in this invention and their many accompanying advantages will become better understood by referring to the following detailed description, thereby readily providing a more comprehensive understanding of them, wherein:

[0021] Figure 1 A schematic diagram of a leak detection system according to this disclosure is illustrated;

[0022] Figure 2 A schematic diagram illustrating the structure of an artificial intelligence-based algorithm is shown.

[0023] Figure 3 A flowchart illustrating a method for operating a leak detection system according to this disclosure is shown;

[0024] Figure 4 A flowchart of a generator neural network is illustrated; and

[0025] Figure 5 A table illustrating different aging processes of chemically chromogenic pigment strips is provided. Detailed Implementation

[0026] The proposed technical solution, detailed below, involves the use of special pigments that change color upon contact with a specific gas (i.e., hydrogen (H2)). These pigments can be placed at points, connections, or areas where leaks are most likely to occur. Such areas can be, for example, parts of a factory or machine; and more specifically, one or more gas supply lines, manifolds, valves, seals, joints, vents, orifices, critical welded connections, threaded connections, pipe fittings, and accessories. Cameras are used to monitor these pigments. One or more cameras can be used, specifically, as stationary or fixed cameras and / or cameras mounted on drones or autonomous robots. When a leak occurs, the pigments change color, and the camera detects this change. The camera can capture one or more, or a series of, options such as: one or more single images, one or more still images, one or more frames extracted from video, and one or more video images. The camera is adapted to generate a signal corresponding to the area of ​​interest, and especially a video signal.

[0027] Once the monitoring system (e.g., the camera mentioned above) identifies a color change indicating a potential hydrogen (or gas) leak, it communicates with a control logic unit equipped with a trained artificial intelligence program based on two neural networks: one network is trained using artificially generated colors to simulate possible color changes in the pigment under conditions of a gas leak or simple aging (false positives); and the other network processes the image captured by the camera. The control logic unit not only identifies the color change but also automatically takes steps to ensure safety. Thus, by using simple visual signals, such as color changes in the pigment, the control logic unit can activate safety devices.

[0028] Now refer to the attached diagram, Figure 1A schematic diagram of a system for detecting gas leaks in factories, machines, or similar equipment is shown. This system provides advanced capabilities compared to traditional gas detection methods by utilizing chemically-based color-changing pigments that change color upon contact with a specific gas.

[0029] The leak detection system referred to by reference numeral 1 in the attached figure is specifically designed for use in factories, machines, gas supply lines, manifolds, valves, seals, joints, discharge points, orifices, critical welded connections, threaded connections, pipe fittings, accessories, or other such equipment that requires the detection of gas leaks.

[0030] System 1 includes multiple gas sensors 3, which are placed in different locations within the system to detect potential gas leaks.

[0031] Specifically, in the described implementation scheme, the detection of hydrogen (H2) gas is considered. However, other gases, such as helium, nitrates, methane, ammonia, carbon dioxide, etc., may be considered, as leaks of these gases in the plant could be hazardous due to their flammability or explosive risks. Specific chemical color-changing pigments must be used for different gases.

[0032] Gas sensors 3 are placed on one or more corresponding parts P in a controlled machine or plant. Each gas sensor 3 includes at least one chemically color-changing pigment. This pigment undergoes a color change when it encounters a specific gas that the system is designed to detect.

[0033] Examples of chemichromic pigments used for detecting hydrogen leaks include titanium-free metal oxide particles and platinum group metal (PGM) compounds. Examples of chemichromic pigments for detecting ammonia are alkali metals and metal oxide compounds.

[0034] System 1 also includes one or more image acquisition detectors 4. The image acquisition detectors 4 are oriented in a manner that enables them to capture image or video images of the area of ​​interest, including parts P of a plant or machine in which the gas sensor 3 is installed. This orientation of the image acquisition detectors 4 is crucial because any color change in the chemically discoloring pigment of the gas sensor indicates a possible gas leak. Therefore, each image acquisition detector 4 is technically configured to generate a signal or video signal reflecting the monitored area.

[0035] In other implementations, the acquisition detector 4 (i.e., the camera) may also be non-fixed. For example, the camera 4 may be mounted on a drone or autonomous robot that can move in space.

[0036] The leak detection system 1 also includes a control logic unit 2. The control logic unit 2 includes a processor 21 and is functionally linked to the image acquisition detector 4. The control logic unit 2 is designed to receive video signals from the image acquisition detector 4.

[0037] The control logic unit 2 includes a processor 21 configured to process the video signals using an artificial intelligence (AI) driven algorithm 100 implemented as a computer program. The AI-driven algorithm 100 is trained to identify color changes in the chemical color-changing pigment in the gas sensor 3 in the presence of a gas leak originating from the video signal detected by the image acquisition detector 4. Furthermore, upon detecting such a color change, the control logic unit 2 generates a safety signal highlighting the specific area in the machine or factory where the leak has occurred.

[0038] The AI-driven algorithm 100 aims to accurately verify images captured by an image acquisition detector 4 (which is typically a camera 4) to monitor parts P in a plant of interest by employing a generative model (specifically called a Generative Adversarial Network (GAN)). The model is designed to detect deviations from the expected aging of pigments under certain environmental conditions, which could indicate the presence of gas leaks.

[0039] AI-based Algorithm 100 (see...) Figure 2 The GAN framework consists of two neural networks: a generator neural network 110 and a discriminator neural network 120.

[0040] The generator neural network 110 is a neural network trained to produce realistic images of pigments under specific environmental conditions and in the absence of gas leakage. It is then designed to train another neural network.

[0041] The discriminator neural network 120 is a more advanced neural network structure compared to traditional GANs, characterized by a "three-way" discriminator neural network 120. In fact, the discriminator neural network 120 is optimized not only to distinguish real images from those generated by the generator 110 (used for its training, as mentioned above), but also to identify whether the real image shows the presence of a gas leak. Specifically, the discriminator 120 predicts whether the input image:

[0042] - It is captured under specific aging and environmental conditions without leakage;

[0043] -Generated by generator network 110; or

[0044] - Indicates signs of a gas leak.

[0045] It is worth noting that the optimization procedure of the discriminator neural network 120 follows the guidelines proposed in relevant literature (such as Valvano, G., Leo, A. and Tsaftaris, SA, 2021, “Re-using adversarial mask discriminators for test-time training under distribution shifts”).

[0046] Furthermore, during the test, images of the pigment were captured by one of the detector cameras in detector camera 4 and evaluated by discriminator neural network 120. If discriminator neural network 120 determined that the image was not realistic, this indicated that the visual changes in the pigment could not be attributed to the natural aging process of part P. In this case, the system marked the image as having undergone abnormal changes, which may be associated with a gas leak.

[0047] The discriminator neural network 120 also has the ability to go beyond binary classification typically seen in GANs, enabling it to distinguish between real images with and without leakage, thereby reducing false positives with greater confidence compared to other models that do not implement a two-order neural network approach.

[0048] Back Figure 1 The control logic unit 2 also includes a storage memory 23 for storing video signals generated by the optical detector 4 and computer programs for AI algorithms used to control the operation of the control logic unit 2 itself. The control logic unit 2 also includes a display 22 that provides the user with a visual representation of a potential gas leak and its precise location.

[0049] One of the exemplary processors 21 or processing components typically used to run AI-based algorithms is a specialized computing unit called a graphics processing unit (GPU). Their inherent architecture allows for parallel processing, making them well-suited for matrix operations frequently used in neural networks, a common AI methodology. In the context of implementing the leak detection system 1, the GPU can rapidly process large amounts of data from the image acquisition detector 4, thus ensuring real-time detection.

[0050] In other implementations, processor 21 may be a field-programmable gate array (FPGA). FPGAs offer the flexibility of post-manufacturing programmability. FPGAs are reconfigurable, which ensures they can be adapted to a variety of AI models or updated as algorithms evolve.

[0051] A dedicated AI processor (often referred to as a neural processing unit (NPU)) represents a third possible implementation of processor 21 or processing element. The NPU is designed to accelerate machine learning tasks. Furthermore, the NPU is optimized for the lower-precision computations common in neural network operations, enabling them to deliver faster performance with reduced power consumption. When integrated into leak detection system 1, the NPU can effectively handle real-time data streams, processing them through a trained AI model to identify patterns indicating potential leaks, faults, or hazards.

[0052] In some embodiments, the gas sensor can be implemented as a gas detection strip 3. This strip, together with a chemically color-changing pigment, is laminated onto an outer surface and can adhere to a mechanical component or surface of the observed plant or machine. Figure 1 In the embodiment shown, the gas detection strip 3 is applied to the joint P.

[0053] The image acquisition detector may include one or more cameras 4. Each camera 4 has the capability to detect one or more gas sensors 3.

[0054] The control logic unit 2 and the image acquisition detector 4 can be connected via cable or wirelessly. In the latter case, this can be via, for example, Bluetooth. ® Various means such as Wi-Fi, infrared, or radio frequency are used to achieve this, making the system ultimately very flexible and easy to install or replace components, especially the camera 4.

[0055] Camera 4 is a high-definition (HD) type specifically designed for security monitoring systems. In the first embodiment, the HD camera 4 is equipped with a wide-angle lens, enabling it to capture a wider area of ​​interest, thereby capturing more sensors 3. The lens is also coated with an anti-glare solution, which prevents image distortion due to sudden changes in light, thus ensuring clear visibility during both day and night.

[0056] The second implementation emphasizes the HD camera's adaptability to changing lighting conditions. An advanced night vision mode utilizing infrared (IR) LEDs is incorporated into camera 4. When ambient light drops below a certain threshold, camera 4 has a built-in control system to automatically switch to this mode. The transition between normal mode and night vision mode is seamless, ensuring continuous monitoring.

[0057] Another implementation may include the motion detection capabilities of an HD camera. Advanced algorithms analyze captured video footage in real time.

[0058] In another implementation, the HD camera 4 incorporates zoom features to allow focusing on a specific part or object or gas sensor in the area of ​​interest.

[0059] To complement the hardware features, another implementation introduces cloud-based storage integration. When connected to a secure network, the HD camera 4 can automatically store the video signal of captured video clips in the cloud.

[0060] In many cases, AI-driven algorithms use neural networks to perform their tasks. Convolutional Neural Networks (CNNs) are the preferred choice for this task due to their exceptional performance in image recognition.

[0061] In some implementations, the neural network algorithm is trained on a large dataset encompassing various types of gas leak scenarios, ranging from small leaks to major leaks, using different colors. This comprehensive dataset enables the algorithm to identify even the most subtle gas leak signatures, thus improving the sensitivity of GLDS. This extensive training ensures the system can instantly identify real-time leak patterns, making timely detection possible and thereby preventing potential hazards.

[0062] Another implementation scheme explores in detail the convolutional neural network (CNN)-based model used in the gas leak detection system 1. Leveraging the inherent spatial hierarchy within CNNs, the model analyzes image or video feeds from sensors that can reveal the physical manifestations of gas leaks, such as changes in chemical color-changing pigments. The layered approach of CNNs ensures that the model can recognize patterns at varying levels of complexity, thereby capturing gas leaks that might be missed by conventional detection systems.

[0063] Another implementation focuses on recurrent neural network (RNN) models ideal for time-series data. Considering that gas leaks may exhibit temporal patterns, an RNN-based gas leak detection system 1 can analyze continuous data points from gas sensor 3 over time. This is particularly useful in detecting slow leaks that may develop over extended periods. By understanding past patterns and predicting future patterns, this implementation contributes to proactive leak management.

[0064] The implementation plan emphasizes the integration of feedback loops in the neural network algorithm. Upon detecting a gas leak, the system records the event, and this data is continuously fed back into the neural network for retraining. This dynamic learning process ensures that the gas leak detection system 1 continuously improves its detection capabilities, thereby adapting to new scenarios and environments.

[0065] Continue to refer to Figure 1 The control logic unit 2 can be operatively connected to the safety module 5 of the machine or plant. The safety module 5 can be mounted on the supply pipe 6. In a specific embodiment, the supply pipe 6 is a gas supply pipe, and in an even more preferred embodiment, the supply pipe is a hydrogen supply pipe. A hydrogen gas flow 7 is shown in the same figure.

[0066] In the illustrated embodiment, safety module 5 includes a shut-off valve 51 positioned on the gas supply line 6. Upon detection of a leak, control logic unit 2 generates a safety signal to immediately command shut-off valve 51 to close. This immediate closure ensures that gas flow is stopped, thereby preventing any further leaks and associated hazards. Shut-off valve 51 can be designed to respond within milliseconds, ensuring rapid action to mitigate any potential risks.

[0067] In addition to the shut-off valve 51, the safety module 5 may also integrate an exhaust valve 52. The exhaust valve 52 serves a dual purpose. First, it releases any overpressure that may have accumulated in the supply line 6 after the shut-off valve 51 is closed, and second, it vents any residual gas present in the supply line 6, thereby ensuring the safety of the surrounding environment.

[0068] Vent valve 52 can be modified to open automatically in response to a predetermined pressure threshold or upon receiving a command from control logic unit 2. Vent valve 52 complements shut-off valve 51, thereby providing a comprehensive safety mechanism that not only stops gas flow but also manages any aftereffects of a leak.

[0069] In addition to its primary function of indicating gas leaks, the generated safety signal is also helpful in controlling safety module 5. When activated, the safety signal can stop the gas supply via hydrogen supply pipe 6.

[0070] The aforementioned system operates using method 8, which is based on a discriminator neural network 120. Method 8 ensures accurate gas leak detection. Specifically, refer to... Figure 3 The flowchart of gas detection method 8 is shown below.

[0071] Method 8 utilizes a control logic unit 2 with an integrated processor 21 and a storage memory 23 containing an AI-driven algorithm. This algorithm completes several steps when executed by the processor 21.

[0072] At the outset, method 8 provides the following steps: the control logic unit 2 acquires (step 81) the video signal generated by the image acquisition detector 4. After this, the signal undergoes rectification (step 82). The refined signal is then processed (step 83) to identify any color change in the chemically chromogenic pigment of the gas sensor 3. Any detected color change indicates a gas leak, prompting the generation of a safety signal (step 84) specifying the location of the leak.

[0073] Method 8 provides enhanced accuracy in detecting hydrogen (H2) gas leaks. Furthermore, it encompasses multiple sub-steps to ensure leak detection.

[0074] Specifically, the acquisition step 81 includes the following: acquiring a 3D model of the pipeline system 6, determining the strip position 3 in such a 3D environment, assessing the lighting conditions, and optimizing the position and focus of the image acquisition detector 4.

[0075] The acquisition step 81 also includes step 812, in which the position of the stripe in the 3D environment is determined using the pigment being framed, based on the position and orientation of the camera, making full use of the knowledge of where various colored areas are located.

[0076] The acquisition step 81 also includes a step 813 to determine the strip lighting conditions, wherein the region of interest is actually illuminated (light intensity, white balance, etc.) based on environmental conditions such as time of day, date of year (to know the position of the sun), the location of the region of interest (whether it is shaded or not, determined based on the position of the sun and the 3D model), and the average brightness detected in the frame.

[0077] Acquisition step 81 also includes step 814 of determining the optimal camera position. In this step, system 1 determines whether the region of interest is correctly framed, and then checks whether the focus of camera 4's lens and the position of the region of interest within the frame are optimal or can be improved.

[0078] The rectification step 82 includes noise or environmental constraint detection 821, wherein, based on the analysis of the frames, it is determined, based on the location of the region of interest, the presence of "noise" that hinders or reduces the accuracy of color measurements (for open environments, this could be objects covering the area, reflections, etc.; for enclosed environments, this could be camera vibrations that blur the image, radiation, diffraction, reflections, refractions, etc.). In a specific embodiment, noise or environmental constraint detection 821 is performed by acquiring one or more frames of the chemically colored stripe under factory or machine shutdown conditions. It is not excluded that typical problems in open environments do not exist in closed environments, and vice versa. It is also not excluded that typical problems inside a housing do not exist outside the housing. Such a housing could be, for example, the housing of a gas turbine. The acquired data is sent to a database for comparison with a predetermined set of data, and this data is also stored for use in the next run. In even more specific embodiments, the predetermined set of data may represent a certain environment, and thus can be identified as a first subset of data representing a first environment and a second subset of data representing a second environment. Such subsets of data that can be associated with a specific environment can be defined as different measurement ranges of the data, such as occlusion by objects covering the area, light reflection in open spaces and / or camera vibrations that blur the image, and light or elemental radiation, diffraction, reflection, and refraction on areas within enclosed spaces. Following such acquisition of frames and comparison with sets or subsets of data, a baseline is determined in baseline determination step 822. Rectification step 82 also includes a working constraint definition step 823. In the latter step 823, working conditions for the region of interest are defined, such as pressure, temperature, and / or the percentage of the element to be detected actually present in the working fluid, which are extrapolated from the plant's operating conditions. In a specific embodiment, the working constraint is represented by the pressure and / or partial pressure of hydrogen, based on which the exposure time can be defined. It is well known that hydrogen in a fuel mixture has a lower partial pressure than pure hydrogen; therefore, the acquisition time for color changes will be slower in hydrogen in a fuel mixture.

[0079] More specifically, in step 823, the images acquired by camera 4 are further processed by determining the spectrum of each constituent pixel of each acquired image.

[0080] In some implementations, the spectrum can be calculated using a Fast Fourier Transform (FFT). The spectral data for each pixel is then analyzed and compared to the spectral profiles of other pixels within the same image or from previous images. This comparison aims to identify any significant spectral differences that could suggest a change in the physical or chemical properties of the observed part P. The differential spectrum associated with each pixel is then evaluated to identify any anomalies or deviations from the expected spectrum (step 824). Such deviations are then assessed to determine whether they indicate the presence of a gas leak or other abnormal conditions.

[0081] Following the comparison and storage sub-process 824, the rectification step 82 further includes a threshold determination sub-step 825, where thresholds for the expected and actual color difference are defined for various alarm levels based on the evaluations made in previous steps. In some embodiments, between the baseline determination step 822 and the operational constraint definition 823, there is a step 826 for eliminating ambient noise, where possible measures are taken to eliminate noise by rescanning. If it is impossible to effectively frame the region of interest after two or three attempts, System 1 sends a notification and continues the scanning sequence of each region of interest. If it is impossible to verify the region of interest after several consecutive traversals, System 1 sends an alarm message to resolve the problem hindering operation. System 1 learns the optimal camera position to avoid noise based on external parameters (time of day, brightness level, etc.).

[0082] The process of processing the refined signal 83 includes a sub-step 831 that defines the expected pigment color due to aging, where the expected color to be detected after scanning is determined. The expected color depends on environmental conditions (lighting conditions) and historical factors (color from the previous reading, the duration of pigment exposure to external factors such as temperature, UV radiation, dust, and precipitation). To calibrate the prediction algorithm, virtual pigments (exposed to external factors like real pigments but mounted so that they never encounter the product they are intended to detect) and information derived from laboratory tests on the same pigments can be used. Furthermore, processing the refined signal 83 includes a sub-step 832 that determines the pigment color, where the actual color of the pigment is evaluated and the color difference is assessed. Then, at sub-step 833, the color difference is correlated with the signal based on a defined threshold. Specifically, System 1 decides what actions to take (e.g., no action, warning message, danger message, or security system lockout).

[0083] Signal processing also includes sub-steps such as identifying ambient noise, establishing a baseline by removing the noise, determining operational constraints in the observed area, and defining a color difference threshold that triggers changes in alarm intensity.

[0084] The processing steps also include: determining the expected pigment color change due to aging, establishing the current pigment color, and associating any color difference with a specific safety signal. The safety signal is modulated based on the potential refraction in the video image of the part (P) with the attached sensor (3).

[0085] In summary, this system and method represent a significant advancement in gas leak detection, ensuring both safety and efficiency.

[0086] In various implementations, the leak detection system 1 employs a camera 4 with a rotating head, supplemented by a zoom lens. This combination ensures that the camera 4 maximizes its effectiveness in monitoring the target area. The mobility provided by the rotating head allows for adaptability to various orientations. Coupled with the zoom lens, it ensures that the camera can capture clear, close-up images of a specific area, thereby enhancing the detail and clarity of the captured video signal.

[0087] To optimize camera positioning, its optimal location is defined. This is achieved using a 3D model of the entire piping system 5, along with the precise locations of the monitored connections. Utilizing the 3D model allows users or system designers to determine the camera's vantage point, ensuring the area of ​​interest is perfectly framed. This precise placement ensures that camera 4 effectively captures the entire chemically chromatic band 3 and its associated portions.

[0088] Artificial intelligence (AI) algorithm 100 performs several tasks to enhance the system's capabilities. Initially, the AI ​​is responsible for precisely locating the exact position of the color stripe within the captured frame. This automated detection ensures the stripe is always monitored, eliminating any need for manual recalibration. Once the stripe is in a 2D frame, the AI ​​uses a 3D model to determine its exact position in the 3D environment, providing a spatial perspective. Furthermore, the AI ​​estimates the stripe's lighting conditions. This assessment ensures that any variations in the stripe's color due to lighting fluctuations are taken into account and adjusted for.

[0089] Over time, chemically chromogenic pigments can undergo natural aging. Therefore, the AI ​​determines the expected pigment color attributable to this aging process. After establishing this baseline, the AI ​​then determines the color difference between the expected pigment color and the observed pigment color (see, for example...). Figure 5 It shows bands made of metal oxide (palladium) after exposure to sunlight and atmospheric factors under different light conditions for at least 2 years.

[0090] Color detection can be performed using relatively simple programs, many of which are based on Python. In fact, similar algorithms have been envisioned for detecting color changes in thermal coatings. The system examines the ratio of RGB values, rather than relying on absolute RGB values ​​that can be affected by various external factors. This technique provides a more reliable measure of color change because it is unaffected by changes in absolute brightness or contrast.

[0091] To further enhance color reading capabilities, the system uses focused light. Guiding a concentrated beam onto the strips improves the camera's ability to detect minute color changes, making the system more sensitive and accurate. Additionally, to improve the accuracy and consistency of tracking color changes, "control strips" are employed. These strips help identify color shifts not attributable to hydrogen (H2) exposure, such as those caused by environmental factors or pigment aging.

[0092] Importantly, it's important to note that the applications of chemichromic pigments are not limited to hydrogen detection. In fact, these pigments have diverse applications; a familiar example is their use in pH banding across various industries. Furthermore, chemichromism is not confined to the visible light spectrum. Many pigments react and change color in response to chemical exposure, and even if those color changes are imperceptible to the human eye, they can be detected in other spectra, such as ultraviolet or infrared.

[0093] at last, Figure 4 An implementation scheme with a flowchart illustrating the main operations of the generator neural network 110 is shown. The instructions shown are to generate a realistic image of pigment 86 under specific environmental conditions and in the absence of gas leakage, and to train the discriminator neural network 120 87 by generating the image generated in step 86.

[0094] The system outlined here leverages the fusion of advanced camera technology, comprehensive 3D modeling, and AI-driven analysis to set a new paradigm in the field of gas leak detection and monitoring. By controlling and utilizing the properties of chemically chromogenic pigments and integrating advanced analytical measures, this invention promises unprecedented efficiency and accuracy in gas leak detection.

[0095] advantage

[0096] The advantage of gas leak detection systems is the ability of cameras to monitor remote, inaccessible, or completely inaccessible locations. By employing cameras, the system ensures continuous monitoring of these critical points without compromising human safety.

[0097] Another advantage of gas leak detection systems is the ease of sensor placement. Sensors can be precisely located at the most likely site of a leak, facilitating immediate identification of the actual leak location. This proactive approach not only enables rapid detection but also expedites necessary response measures, ensuring minimal gas waste and enhanced safety.

[0098] Another advantage of gas leak detection systems is the integration of artificial intelligence (AI), which further amplifies their effectiveness. One of the long-standing challenges in gas leak detection is the occurrence of false positives, often triggered by changes in environmental conditions. However, by leveraging AI's ability to distinguish between genuine leaks and mere changes in environmental conditions (such as variations in lighting), the system significantly reduces false alarms. This ensures that generated alarms are genuine, allowing operators to respond with confidence every time.

[0099] Another advantage of the gas leak detection system disclosed herein is its versatility. The system is designed to operate seamlessly in both indoor and outdoor settings, unaffected by the challenges present in each environment. One such challenge, particularly relevant in open atmospheres, is the dilution of leaking hydrogen. Thanks to the system's innovative design and technology, its detection capability is not hindered even when hydrogen diffuses in the atmosphere. This resilience ensures rapid leak detection regardless of the surrounding environment. Another AI-related advantage is the ability to detect leaks of specific gases, both within the fuel mixture and during specific combustion stages. This feature is crucial for dual-fuel machines, where fuels, such as natural gas and other hydrocarbons, can be delivered to the turbine purely or as a mixture with hydrogen, ammonia, or others. Hydrogen can also be used as a pure fuel during certain combustion stages.

[0100] Other implementations are within the scope and substance of the disclosed subject matter. For example, due to the nature of software, the functions described above can be implemented using software, hardware, firmware, hardwired, or any combination thereof. Features implementing the functions can also be physically located in various locations, including parts distributed such that the functions are implemented in different physical locations.

[0101] While various aspects of the invention have been described with reference to specific embodiments, it will be apparent to those skilled in the art that numerous modifications, variations, and omissions are possible without departing from the spirit and scope of the claims. Furthermore, unless otherwise specified herein, the sequence or order of any process or method steps may be altered or rearranged according to alternative embodiments.

[0102] Reference has been made in detail to embodiments of this disclosure, one or more examples of which are illustrated in the accompanying drawings. Each example is provided by way of interpretation and not limitation of this disclosure. Indeed, it will be apparent to those skilled in the art that various modifications and variations can be made to this disclosure without departing from its scope or substance. Throughout this specification, references to “one embodiment” or “some embodiments” mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the disclosed subject matter. Therefore, the phrases “in one embodiment” or “in some embodiments” appearing in various places throughout this specification do not necessarily refer to the same embodiment. Furthermore, in one or more embodiments, a particular feature, structure, or characteristic may be combined in any suitable manner.

[0103] When describing the elements of various embodiments, the articles “a,” “the,” and “the” are intended to mean that one or more of the elements are present. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that additional elements may be present in addition to those listed.

[0104] The subject matter described herein may be implemented in digital electronic circuits, or in computer software, firmware, or hardware (including the structural components disclosed herein and their structural equivalents), or a combination thereof. The subject matter described herein may be implemented as one or more computer program products, such as those tangibly embodied in an information carrier (e.g., in a machine-readable storage device) or embodied in a propagated signal, for execution by or control of a data processing device (e.g., a programmable processor, a computer, or multiple computers) of one or more computer programs. A computer program (also referred to as a program, software, software application, or code) may be written in any form of programming language (including compiled or interpreted languages) and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for a computing environment. A computer program does not necessarily correspond to a file. A program may be stored as a portion of a file containing other programs or data, in a single file dedicated to the program under consideration, or in multiple co-located files (e.g., a file storing portions of one or more modules, subroutines, or code). Computer programs can be deployed to execute on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communication network.

[0105] The processes and logical flows described in this specification (including the method steps of the subject matter described herein) can be executed by one or more programmable processors that execute one or more computer programs to perform the functions of the subject matter described herein by manipulating input data and generating output. These processes and logical flows can also be executed by special-purpose logic circuitry (e.g., FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits)), and the devices of the subject matter described herein can be implemented as special-purpose logic circuitry (e.g., FPGAs or ASICs).

[0106] By way of example, processors suitable for executing computer programs include both general-purpose microprocessors and special-purpose microprocessors, as well as any one or more processors in any kind of digital computer. Generally, a processor receives instructions and data from read-only memory or random access memory, or both. The basic components 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 one or more mass storage devices (e.g., magnetic disks, magneto-optical disks, or optical disks) for storing data, or operatively coupled to receive data from or transfer data to one or more mass storage devices (e.g., magnetic disks, magneto-optical disks, or optical disks) for storing data. Information carriers suitable for embodying computer program instructions and data include all forms of non-volatile memory, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices); magnetic disks (e.g., internal hard disks or removable magnetic disks); magneto-optical disks; and optical disks (e.g., CDs and DVDs). The processor and memory may be supplemented by or incorporated into special-purpose logic circuitry.

[0107] To provide interaction with the user, the subjects 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 pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user. 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 sound, speech, or tactile input.

[0108] 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. However, at a minimum, a module should not be construed as software not implemented on hardware, firmware, or recorded on a non-transitory processor-readable and recordable storage medium (i.e., a module itself is not software). In practice, a "module" will be interpreted as always including at least some physical non-transitory hardware, such as a processor or part of a computer. Two different modules may share the same physical hardware (e.g., two different modules may use the same processor and network interface). The modules described herein can be combined, integrated, separated, and / or replicated to support a variety of applications. Additionally, instead of functions performed at a particular module, or functions described herein as performing at a particular module, functions may be performed at one or more other modules and / or by one or more other devices. Furthermore, modules may be implemented locally or remotely across multiple devices and / or other components relative to each other. Additionally, modules may be moved from one device and added to another device, and / or may be included in two devices.

[0109] The subject matter described herein can be implemented in a computing system that includes back-end components (e.g., a data server), middleware components (e.g., an application server), or front-end components (e.g., a client computer with a graphical user interface or web browser through which a user can interact with a specific implementation of the subject matter described herein), or any combination of such back-end, middleware, and front-end components. Components of the system can be interconnected via any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks (“LANs”) and wide area networks (“WANs”), such as the Internet.

Claims

1. A leak detection system (1) for detecting gas leaks from a factory or machine to be monitored, the leak detection system comprising: One or more gas sensors (3), the one or more gas sensors being adaptable to one or more corresponding parts (P) of the plant or machine to be monitored for detecting possible gas leaks; Its features Each gas sensor (3) contains at least one chemichromic pigment, which, in the event of a leak, changes color upon contact with the gas to be detected; and The system (1) is characterized in that it further includes: One or more image acquisition detectors (4). Each image acquisition detector (4) is arranged to acquire an image of a region of interest, in which a part (P) of the factory or machine on which the sensor (3) is applied exists, and Each of the image acquisition detectors (4) is adapted to generate a signal corresponding to the region of interest; and A control logic unit (2) is operatively connected to the one or more image acquisition detectors (4) and includes a processor (21). The control logic unit (2) is configured to receive the signal from the one or more image acquisition detectors (4). The control logic unit (2) is configured to process the signals from the one or more image acquisition detectors (4) using an artificial intelligence-based algorithm (100) residing in the processor (21), the artificial intelligence-based algorithm being trained to: - In the event of a gas leak from one or more parts (P) of the factory or machine to be tested, identify the color change of the chemically color-changing pigment of the sensor (3); and - A safety signal is generated in the event that the chemical color-changing pigment in the gas sensor (3) changes color due to a gas leak from one of the parts (P) of the plant or machine to be monitored, wherein the safety signal indicates the area of ​​the plant or machine to be monitored where the gas leak occurs.

2. The system (1) according to claim 1, wherein the at least one sensor is a gas detection strip (3) made of chemically color-changing pigment, wherein the strip (3) can be applied to the mechanical parts or surface of the factory or machine to be monitored.

3. The system (1) according to any one of the preceding claims. The image acquisition detector mentioned above includes one or more cameras (4), and Each camera (4) is adapted to detect multiple gas sensors (3).

4. The system (1) according to any one of the preceding claims, wherein the one or more gas sensors (3) applicable to one or more corresponding parts (P) of the plant or machine are located on or within the mechanical parts or surfaces due to noise or environmental constraints.

5. The system (1) according to any one of the preceding claims. The control logic unit (2) and the one or more image acquisition detectors (4) are connected via a method such as Bluetooth. ® Wireless system connections such as Wi-Fi, infrared, and radio frequency.

6. The system (1) according to any one of the preceding claims, wherein the gas to be detected is hydrogen (H2).

7. The system (1) according to any one of the preceding claims, wherein the artificial intelligence-based algorithm (100) is implemented by at least one neural network, preferably a convolutional neural network (CNN).

8. The system (1) according to any one of the preceding claims, wherein the artificial intelligence algorithm (100) includes a discriminator neural network (120) for predicting whether an input image acquired by the image acquisition detector (4) was captured under specific aging and environmental conditions without leakage, and for displaying signs of gas leakage.

9. The system (1) according to the preceding claims. The artificial intelligence algorithm (100) includes a generator neural network (110) for generating realistic images of the pigment under specific environmental conditions and in the absence of gas leakage. The image generated by the generator neural network (110) is designed to train the discriminator neural network (120).

10. The system (1) according to any one of the preceding claims. The control logic unit (2) is operatively connected to the safety module (5) of the plant or machine to be monitored, which is arranged on the gas supply pipe (6). The safety signal is used to control the safety module (5) of the factory or machine to be monitored to interrupt the supply of gas through the gas supply pipe (6).

11. The system (1) according to any one of the preceding claims, wherein the control logic unit (2) includes a database (23) storing the signal generated by the optical detector (4).

12. The system (1) according to any one of the preceding claims, wherein the control logic unit (2) includes a display (22) for signaling possible gas leaks and displaying the location of such gas leaks in the plant or machine to be monitored.

13. The system (1) according to any one of the preceding claims, the system comprising another optical sensor operatively connected to the control logic unit (2), the control logic unit being configured to generate another optical signal for the region of interest. The control logic unit (2) is configured to detect different refractions and adjust the image of the region of interest acquired by the image acquisition detector (4).

14. A computer-implemented method (8) for detecting gas leaks, the computer-implemented method comprising a control logic unit (2) having a processor (21) and a memory (23), the memory storing a computer program implementing an artificial intelligence-based algorithm (100) for execution by the processor (21), the computer program including instructions for the following operations: The correct acquisition of the signal generated by the position of the image acquisition detector (4) is determined by the control logic unit (2) (81); The signal is rectified (82) by the control logic unit (2); The artificial intelligence-based algorithm (100) includes a discriminator neural network (120), which includes instructions for the following operations: The signal is processed (83) by the control logic unit (2) to detect the color difference of the chemically color-changing pigment of at least one gas sensor (3), wherein the gas sensor (3) is adaptable to one or more corresponding parts (P) of the plant to be monitored for detecting possible gas leaks, and wherein each gas sensor (3) contains at least one chemically color-changing pigment that, in the event of a leak, is capable of changing its color upon contact with the gas to be detected; and In the event that the chemical color-changing pigment in the gas sensor (3) changes color due to a gas leak from one or more parts (P) of the factory or machine to be monitored, a safety signal (84) is generated, wherein the safety signal indicates the area of ​​the factory or machine to be monitored where the gas leak occurs.

15. The method (8) according to the preceding claim, wherein the gas to be detected is hydrogen (H2).

16. The method (8) according to any one of claims 14 or 15, wherein the correct acquisition signal determination (81) step comprises the following sub-steps: Obtain a 3D model of the (811) piping system and connection locations; Determine the stripe positions in the (812) 3D environment; Determine the (813) stripe illumination conditions; and Determine (814) the position of the optimal image acquisition detector (4), wherein the control logic unit (2) determines whether the region of interest is correctly framed, and then determines whether the focus of the lens and the position of the region of interest within the frame are optimal or can be improved.

17. The method (8) according to any one of claims 14 to 16, wherein the signal processing (83) step comprises the following sub-steps: Detect (821) noise or environmental constraints; The baseline of (822) is determined by eliminating the noise described in (823); Determine the working constraints of the region of interest (823); Determine the operational threshold (825) for defining the color difference threshold between the expected and actual values ​​for various alarm levels.

18. The method (8) according to the preceding claim, wherein the working constraints determined in the step of determining working constraints (823) include pressure and / or temperature and / or percentage of the gas to be detected based on the plant or machine to be monitored, for the purpose of detecting possible gas leaks.

19. The method (8) according to the preceding claim, wherein the exposure time of the chemical color-changing pigment is defined according to the determined partial pressure.

20. The method (8) according to any one of claims 17 to 19, wherein The determining step (823) or the signal processing (83) includes: Determine the spectrum of the pixels in the image acquired by camera (4), and Analyze the spectral data of each pixel and compare it with the spectrum of the same pixel in the same image or from a previous image to identify spectral differences to determine changes in the physical or chemical properties of the observed part (P) in order to determine gas leaks.

21. The method (8) according to any one of claims 14 to 20, wherein the signal processing (83) comprises the following sub-steps: Definition (831) The expected pigment color due to aging; Determine (832) the pigment color; and The color difference is associated with a security signal.

22. The method (8) according to any one of claims 14 to 21, the method comprising the following steps: The safety signal is adjusted (85) by the possible refraction of the image of the part (P) on which the sensor (3) is applied to the factory or machine.

23. The method (8) according to the preceding claim, wherein the safety signal conditioning (85) comprises the following steps: Another optical signal is received by another optical sensor for detecting different refractions, wherein the other optical sensor is connected to the control logic unit (2); and The refractive index is correlated with the gas leak.

24. The method (8) according to any one of claims 14 to 23, wherein the artificial intelligence-based algorithm (100) comprises a generator neural network (110), the generator neural network comprising instructions for the following operations: Generate (86) a realistic image of the pigment under specific environmental conditions and in the absence of gas leakage; and The discriminator neural network (120) is trained (87) by the image generated in the generation step (86).

Images