New energy equipment fault detection system and method fused with double-end infrared spectrum monitoring

Abstract

The invention provides a new energy equipment fault detection system and method fused with double-end infrared spectrum monitoring, and relates to the technical field of new energy equipment fault detection. The deviation recognition model construction module is used for constructing an output deviation recognition model; an equipment fault judgment module; the downstream equipment open-circuit isolation module is used for performing downstream equipment open-circuit isolation if the equipment is in the fault state; an input real-time spectrogram acquisition module; an equipment fault verification module; and the fault component positioning module is used for performing fault component positioning on the new energy equipment to obtain a target fault component if the new energy equipment is in the fault state. The technical problems that equipment fault detection in the prior art generally depends on simple signal monitoring or regular manual detection in operation of the new energy equipment, response to sudden faults is slow, the accuracy is low, the fault position cannot be rapidly positioned, and the operation efficiency and stability of the new energy equipment are limited are solved.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology for new energy equipment, and specifically to a fault detection system and method for new energy equipment that integrates dual-end infrared spectral monitoring. Background Technology

[0002] In the production and operation of new energy equipment, the operating status of the equipment directly affects the overall energy efficiency and environmental performance. However, in traditional new energy equipment, physical sensors such as pressure, temperature, and gas flow are typically used for basic signal monitoring of equipment operation. While these sensors can monitor changes in the external state of the equipment, their ability to monitor internal energy conversion efficiency or system operating efficiency in real time is limited. When new energy equipment malfunctions, especially in critical energy conversion and storage stages, traditional sensor monitoring methods may suffer from fault detection delays, failing to identify performance degradation in a timely and accurate manner. More complexly, when equipment malfunctions, traditional detection methods often struggle to quickly locate the fault, typically requiring comprehensive inspection or gradual disassembly, leading to excessive downtime and impacting production efficiency and energy output. Summary of the Invention

[0003] This application provides a fault detection system and method for new energy equipment that integrates dual-end infrared spectral monitoring. It aims to solve the technical problems of existing technologies that usually rely on simple signal monitoring or periodic manual inspection during equipment operation, which result in slow response to sudden faults, low accuracy, and inability to quickly locate faults, thus limiting the operating efficiency and stability of new energy equipment.

[0004] The first aspect disclosed in this application provides a fault detection system for new energy equipment integrating dual-end infrared spectroscopy monitoring. The system includes: a spectrometer configuration module for configuring a dual-end infrared spectrometer on the new energy equipment to be tested, wherein the dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment, the new energy equipment being a hydrogen purification device; a deviation identification model construction module for predefining an output reference impurity component and constructing an output deviation identification model based on the output reference impurity component; an equipment fault judgment module for real-time monitoring of the gas composition after processing by the new energy equipment using the second infrared spectrometer to generate a first output spectrum and synchronizing the first output spectrum to the output deviation identification model for equipment fault judgment; and downstream equipment circuit breaker isolation. The system comprises the following modules: a downstream equipment circuit breaker module, used to isolate the downstream equipment of the new energy equipment and activate the first infrared spectrometer if the output result of the output deviation identification model indicates a fault state; an input real-time spectrum acquisition module, used to continue operating the new energy equipment and upstream equipment after completing the downstream equipment circuit breaker isolation, and continuously monitor the gas composition entering the new energy equipment through the first infrared spectrometer to obtain an input real-time spectrum; an equipment fault verification module, used to interactively obtain the control parameter settings of the new energy equipment, and perform equipment fault verification based on the input real-time spectrum and control parameter settings, and output the equipment verification result; and a fault component location module, used to locate the fault component of the new energy equipment based on the first output spectrum if the equipment verification result indicates a fault state, and obtain the target fault component.

[0005] The second aspect of this application discloses a method for fault detection of new energy equipment integrating dual-end infrared spectroscopy monitoring. The method is implemented using the aforementioned fault detection system for new energy equipment integrating dual-end infrared spectroscopy monitoring. The method includes: configuring a dual-end infrared spectrometer on the new energy equipment to be tested, wherein the dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment, and the new energy equipment is a hydrogen purification device; predefining an output reference impurity component and constructing an output deviation identification model based on the output reference impurity component; the second infrared spectrometer real-time monitoring the gas composition after processing by the new energy equipment to generate a first output spectrum, and displaying the first output spectrum... The output deviation identification model is synchronized to determine equipment faults. If the output deviation identification model indicates a fault, the downstream equipment of the new energy equipment is disconnected and isolated, and the first infrared spectrometer is activated. After the downstream equipment is disconnected, the new energy equipment and upstream equipment continue to operate, and the gas composition entering the new energy equipment is continuously monitored by the first infrared spectrometer to obtain the input real-time spectrum. The control parameter settings of the new energy equipment are obtained interactively, and the equipment fault is verified based on the input real-time spectrum and control parameter settings, and the equipment verification result is output. If the equipment verification result indicates a fault, the faulty component of the new energy equipment is located based on the first output spectrum to obtain the target faulty component.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: By configuring first and second infrared spectrometers at the inlet and outlet of the hydrogen purification equipment, respectively, the composition of the input and output gases can be monitored simultaneously, enabling dual-end monitoring of the equipment's operation. A predefined output baseline impurity component ensures a clear standard for measuring the output gas concentration during equipment operation. This allows for the construction of an output deviation identification model, automatically comparing the actual output with the baseline standard to ensure the impurity concentration remains within acceptable limits. The second infrared spectrometer monitors the treated gas in real time, generating a first output spectrum, which is then analyzed using the deviation identification model. Automatic spectral comparison reduces manual intervention and improves the accuracy of fault diagnosis, enabling rapid and accurate determination of equipment malfunction. When a malfunction is detected, downstream equipment is automatically activated. The circuit breaker isolation ensures that the fault does not spread to other equipment, reducing the occurrence of cascading failures, and activates the first infrared spectrometer at the inlet to provide data support for further fault diagnosis. After completing the circuit breaker isolation of downstream equipment, the new energy equipment and upstream equipment continue to operate, and the input gas composition is monitored by the first infrared spectrometer to obtain the input real-time spectrum. Combined with the control parameter settings of the new energy equipment, equipment fault verification is performed. By interactively verifying the difference between the input and output spectra, it is further determined whether the fault is caused by impurities at the input end, making the judgment result more accurate. After determining that the equipment does have a fault, the faulty component is located by the first output spectrum. By accurately locating the faulty component, it is possible to avoid a complete disassembly of the equipment or a lengthy inspection, thereby shortening the fault repair time and improving the equipment operation and maintenance efficiency.

[0007] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the structure of a new energy equipment fault detection system that integrates dual-end infrared spectral monitoring, provided in an embodiment of this application. Figure 2 This is a schematic flowchart of a new energy equipment fault detection method that integrates dual-end infrared spectral monitoring, provided in an embodiment of this application.

[0009] Explanation of reference numerals in the attached figures: Spectrometer configuration module 10, Deviation identification model construction module 20, Equipment fault judgment module 30, Downstream equipment circuit breaker isolation module 40, Input real-time spectrum acquisition module 50, Equipment fault verification module 60, Fault component location module 70. Detailed Implementation

[0010] This application provides a fault detection system and method for new energy equipment that integrates dual-end infrared spectral monitoring. This solves the technical problems of existing technologies that typically rely on simple signal monitoring or periodic manual inspection during equipment operation, resulting in slow response to sudden faults, low accuracy, and inability to quickly locate faults, thus limiting the operating efficiency and stability of new energy equipment.

[0011] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0012] Example 1, as Figure 1 As shown in the figure, this application embodiment provides a fault detection system for new energy equipment that integrates dual-end infrared spectral monitoring. The system includes: The system includes: a spectrometer configuration module 10, used to configure a dual-end infrared spectrometer for the new energy equipment under test, wherein the dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment, the new energy equipment being a hydrogen purification device; a deviation identification model construction module 20, used to predefine the output reference impurity component and construct an output deviation identification model based on the output reference impurity component; a device fault judgment module 30, used to monitor the gas composition after processing by the new energy equipment in real time with the second infrared spectrometer to generate a first output spectrum, and synchronize the first output spectrum to the output deviation identification model for device fault judgment; and a downstream device disconnection isolation module 40, used to isolate the downstream device if the... If the output of the deviation identification model indicates a fault state, then the downstream equipment of the new energy equipment is disconnected and isolated, and the first infrared spectrometer is activated; the input real-time spectrum acquisition module 50 is used to continue operating the new energy equipment and upstream equipment after the downstream equipment disconnection and isolation is completed, and continuously monitor the gas composition entering the new energy equipment through the first infrared spectrometer to obtain the input real-time spectrum; the equipment fault verification module 60 is used to interactively obtain the control parameter settings of the new energy equipment, and perform equipment fault verification according to the input real-time spectrum and control parameter settings, and output the equipment verification result; the fault component location module 70 is used to locate the fault component of the new energy equipment according to the first output spectrum if the equipment verification result indicates a fault state, and obtain the target fault component.

[0013] Furthermore, the deviation identification model construction module 20 also includes the following operation steps: The output reference impurity components are disassembled based on impurity type to obtain K reference output concentrations for K types of impurity components; a pre-constructed gas spectral characteristic database is built, wherein multiple sample molar absorptivity coefficients and multiple sample optical path lengths of various sample impurity components are associated and stored in the gas spectral characteristic database; the K types of impurity components are used to traverse the gas spectral characteristic database to obtain K real-time molar absorptivity coefficients and K real-time optical path lengths; the K real-time molar absorptivity coefficients, K real-time optical path lengths, and K reference output concentrations are used as conversion conditions, and absorbance conversion is performed using Lambert-Beer's law to obtain K reference absorbances; K reference spectra are generated based on the K reference absorbances, and the K reference spectra are used as comparison benchmarks to construct the output deviation identification model using a Siamese neural network.

[0014] Furthermore, the deviation identification model construction module 20 also includes the following operation steps: The process involves interactively obtaining K sets of first sample spectra of the K impurity components, wherein the K sets of first sample spectra are identified by the K sets of first sample impurity concentrations; calculating the concentration deviation of the K sets of first sample impurity concentrations based on the K benchmark output concentrations to obtain K sets of sample deviation coefficients, wherein the K sets of first sample impurity concentration mappings are greater than the K benchmark output concentrations; constructing K output deviation recognition channels using a Siamese neural network, and then using the K sets of first sample spectra, the K benchmark spectra, and the K sets of sample deviation coefficients as model training data to update the model parameters of the K sets of output deviation recognition channels; connecting the K sets of output deviation recognition channels in parallel to obtain an output deviation recognition layer; calling the spectral separation layer, and cascading the spectral separation layer and the output deviation recognition layer to obtain the output deviation recognition model.

[0015] Furthermore, the deviation identification model construction module 20 also includes the following operation steps: Multiple composite spectra of samples and multiple sets of separated impurity spectra of samples are obtained interactively; based on the multiple composite spectra of samples and the multiple sets of separated impurity spectra of samples, impurity concentration weights are calculated to obtain multiple sets of sample concentration weights; based on the K impurity components, the multiple sets of sample concentration weights are recombined, and the mean of the recombination results is calculated to obtain K standard impurity concentration weights; the spectral separation layer is constructed based on the K standard impurity concentration weights.

[0016] Furthermore, the equipment fault diagnosis module 30 also includes the following operation steps: The first output spectrum is spectrally demixed by the spectral separation layer of the output deviation identification model to obtain K output impurity spectra; the K output impurity spectra are synchronized to the K output deviation identification channels of the output deviation identification layer for impurity deviation analysis to obtain K deviation analysis results; when the K deviation analysis results are empty sets, the output result of the output deviation identification model is in a non-fault state; when any of the K deviation analysis results is an impurity deviation coefficient, the output result of the output deviation identification model is in a fault state.

[0017] Furthermore, the device fault verification module 60 also includes the following operation steps: Interactively obtain K sets of second sample spectra of the K impurity components, wherein the K sets of second sample spectra are identified by the K sets of second sample impurity concentrations, and the K sets of second sample impurity concentrations are mapped to concentrations less than the K reference output concentrations; construct K impurity concentration identification channels using the K sets of first sample spectra, K sets of second sample spectra, K sets of first sample impurity concentrations, and K sets of second sample impurity concentrations; after obtaining an impurity concentration identification layer by connecting the K sets of K impurity concentration identification channels in parallel, configure the spectrum separation layer in front of the impurity concentration identification layer to complete the construction of the impurity concentration identification model; through the... The impurity concentration identification model analyzes the input real-time spectrum to obtain K input impurity concentrations; based on the K input impurity concentrations and control parameter settings, it predicts the output impurity concentration to obtain K predicted output concentrations; the K output impurity spectra are directly synchronized to the K impurity concentration identification channels of the impurity concentration identification model to perform impurity concentration identification, obtaining K output impurity concentrations; the internal deviation of the K predicted output concentrations and the K output impurity concentrations is evaluated based on Euclidean distance to obtain a purification deviation coefficient; if the purification deviation coefficient does not meet the preset purification deviation threshold, the equipment verification result is a fault state.

[0018] Furthermore, the fault component location module 70 also includes the following operation steps: When the equipment verification result is a fault state, H types of impurity components are located based on the K deviation analysis results, wherein the H deviation analysis results corresponding to the H types of impurity components are impurity deviation coefficients, and H is a positive integer less than or equal to K; an impurity-component fault tree is pre-constructed, wherein the impurity-component fault tree includes K groups of purification fault-related components of the K types of impurity components; the impurity-component fault tree is traversed using the H types of impurity components to obtain H groups of purification fault-related components; the H groups of purification fault-related components are aggregated to obtain the target fault component.

[0019] Furthermore, if the equipment verification result is a non-faulty state, then the control parameters of the new energy equipment are optimized and adjusted according to the K output impurity concentrations.

[0020] Through the detailed description of the new energy equipment fault detection method integrating dual-end infrared spectroscopy monitoring in the following specification, those skilled in the art can clearly understand the new energy equipment fault detection system integrating dual-end infrared spectroscopy monitoring in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0021] Example 2, based on the same inventive concept as the new energy equipment fault detection system integrating dual-end infrared spectral monitoring in the aforementioned examples, such as... Figure 2 As shown in the embodiment of this application, a fault detection method for new energy equipment integrating dual-end infrared spectral monitoring is provided. The method includes: A dual-end infrared spectrometer is configured for the new energy equipment to be tested, wherein the dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment, and the new energy equipment is a hydrogen purification device.

[0022] A dual-end infrared spectrometer was configured for the new energy equipment under test. The equipment was a hydrogen purification device, used to remove impurities from raw hydrogen gas, typically including water vapor, carbon dioxide, carbon monoxide, nitrogen, and oxygen, ideally outputting high-purity hydrogen. An infrared spectrometer is an analytical instrument that uses the absorption characteristics of infrared light to analyze the composition of substances. Different molecules produce specific absorption peaks when absorbing infrared light; these peaks reflect the specific components and concentrations in the gas.

[0023] A dual-end infrared spectrometer suitable for detecting hydrogen impurities is selected. This includes a first infrared spectrometer and a second infrared spectrometer, installed at the inlet and outlet of the new energy equipment, respectively. Specifically, the first infrared spectrometer is installed at the inlet to monitor the composition of hydrogen before purification, for subsequent comparison of the gas composition before and after purification. The second infrared spectrometer is installed at the outlet to detect hydrogen after purification. The monitoring results at this point show the actual output composition and purification efficiency of the purification equipment. This dual-end monitoring allows for monitoring whether the hydrogen purification equipment is operating normally and whether it has achieved the expected purification effect.

[0024] A predefined output reference impurity component is used, and an output deviation identification model is constructed based on the output reference impurity component.

[0025] Under normal operating conditions, the output composition of hydrogen purified by new energy equipment should meet the standard as the baseline impurity component, including impurity type and impurity concentration within the normal range. Based on the predefined baseline impurity component, an output deviation identification model is constructed. This model is used to determine whether the output gas composition deviates significantly from the baseline value, thereby determining whether the equipment is faulty. The specific model construction process will be detailed in subsequent steps and will not be elaborated here.

[0026] The second infrared spectrometer monitors the gas composition after processing by the new energy equipment in real time to generate a first output spectrum, and synchronizes the first output spectrum to the output deviation identification model for equipment fault judgment.

[0027] When hydrogen passes through the hydrogen purification equipment, a second infrared spectrometer installed at the equipment outlet continuously monitors the composition of the purified hydrogen. Based on the monitored hydrogen output composition data, a first output spectrum is generated in real time. This spectrum displays the spectral absorption characteristics of different impurities in the hydrogen and their corresponding concentration information. The generated first output spectrum is automatically synchronized to the constructed output deviation identification model. This model compares the actual output spectrum with the reference spectrum to determine if there is a deviation exceeding the normal range. If there is no significant deviation, the equipment is operating normally; if there is a significant deviation, it indicates that the equipment is in a faulty state.

[0028] If the output result of the output deviation identification model is a fault state, then the downstream equipment of the new energy equipment is disconnected and isolated, and the first infrared spectrometer is activated.

[0029] When the output deviation identification model indicates a fault state, to prevent the fault from further affecting downstream equipment in the hydrogen supply chain, it is necessary to disconnect and isolate the downstream equipment. This can be achieved by stopping the flow of hydrogen to downstream equipment through the control system. The purpose is to temporarily isolate the downstream equipment from the purification equipment, ensuring that the fault does not affect other hydrogen processing systems, such as storage devices, compression equipment, or fuel cells. After disconnecting and isolating the downstream equipment, the hydrogen purification equipment continues to operate, and the first infrared spectrometer installed at the inlet of the purification equipment is activated. This spectrometer begins continuous monitoring of the gas composition entering the purification equipment, thereby obtaining more data for fault type identification.

[0030] After the downstream equipment is disconnected and isolated, the new energy equipment and upstream equipment continue to operate, and the gas composition entering the new energy equipment is continuously monitored by the first infrared spectrometer to obtain the input real-time spectrum.

[0031] After the downstream equipment is isolated, the hydrogen purification equipment and its upstream gas supply equipment, such as the gas generator or gas supply system, continue to operate normally. This is to obtain more fault-related data in the isolated state to more accurately determine the source of the problem. A first infrared spectrometer installed at the inlet of the new energy equipment continuously monitors the gas composition entering the equipment, including various impurities and their concentrations, generating a real-time input spectrum.

[0032] The system interactively obtains the control parameter settings of the new energy equipment, performs equipment fault verification based on the input real-time spectrum and control parameter settings, and outputs the equipment verification results.

[0033] The control parameter settings for the new energy equipment are obtained through interaction with the equipment's control system. These parameters include the working pressure of hydrogen during the purification process, the temperature of hydrogen and the equipment during purification, the flow rate of hydrogen within the purification equipment, and the current operating mode of the equipment, such as full-speed operation or partial-load operation. Obtaining these control parameters helps determine the actual operating status of the equipment, and combined with spectral analysis, enables more accurate fault diagnosis. Equipment fault verification is performed based on the input real-time spectral data and control parameter settings, and the verification results are output. The specific equipment fault verification process will be detailed in subsequent steps.

[0034] If the device verification result is a fault state, the faulty component of the new energy device is located based on the first output spectrum to obtain the target faulty component.

[0035] When equipment calibration results indicate that the equipment is in a faulty state, it is necessary to analyze the specific location of the fault. Specifically, the first output spectrum shows the impurity components and their concentrations in the purified gas. Each impurity has a specific processing path in different components of the hydrogen purification equipment. Therefore, by analyzing the concentration and type of different impurities in the spectrum, it is possible to preliminarily determine which part of the purification effect has a problem. Based on the pre-constructed impurity-component fault tree, the impurity information in the spectrum is mapped to the specific components of the equipment. For example, if the concentration of a certain impurity (such as carbon monoxide) is abnormally high, it may indicate that the purification component (such as adsorption device or catalyst) related to that impurity has a problem. Through this mapping, the specific component in the equipment that may be faulty can be quickly located. In fault tree analysis, one impurity may usually involve multiple components. By analyzing multiple impurities, the fault range can be further narrowed down, and finally the target faulty component most likely to fail can be obtained.

[0036] Furthermore, the method includes predefining an output reference impurity component and constructing an output deviation identification model based on the output reference impurity component. The output reference impurity components are disassembled based on impurity type to obtain K reference output concentrations for K types of impurity components; a pre-constructed gas spectral characteristic database is built, wherein multiple sample molar absorptivity coefficients and multiple sample optical path lengths of various sample impurity components are associated and stored in the gas spectral characteristic database; the K types of impurity components are used to traverse the gas spectral characteristic database to obtain K real-time molar absorptivity coefficients and K real-time optical path lengths; the K real-time molar absorptivity coefficients, K real-time optical path lengths, and K reference output concentrations are used as conversion conditions, and absorbance conversion is performed using Lambert-Beer's law to obtain K reference absorbances; K reference spectra are generated based on the K reference absorbances, and the K reference spectra are used as comparison benchmarks to construct the output deviation identification model using a Siamese neural network.

[0037] The standard that the output components of hydrogen after purification should meet under normal working conditions of new energy equipment is determined as the output benchmark impurity components, including impurity type and impurity concentration within the normal range. The output benchmark impurity components are decomposed into several different impurity types, each of which represents a specific component that may remain in hydrogen.

[0038] After disassembly, K baseline output concentrations of K impurity components are obtained, where K is the number of impurity types and K≥1. These impurities need to be removed from hydrogen as much as possible, but in reality, a certain amount of residual impurities is allowed. The baseline output concentration represents the maximum allowable concentration of each impurity. This is the highest impurity concentration allowed by the hydrogen purification equipment under normal operating conditions. This concentration value is usually set according to industry standards, equipment specifications, or safety requirements. These baseline concentrations serve as reference standards in subsequent spectral analysis to determine whether the equipment output gas meets the requirements.

[0039] The gas spectral properties database stores spectral characteristic data for various sample impurity components, including molar absorptivity and optical path length. This data was obtained through experiments or existing scientific data and accurately reflects the spectral absorption characteristics of different sample impurity components. The molar absorptivity describes the light absorption capacity of a specific gas at a specific wavelength. It is a constant related to the properties of gas molecules, representing the light intensity absorbed by the gas at a unit concentration and unit path length. Each impurity gas has a specific molar absorptivity, reflecting its absorption intensity of infrared light at a specific wavelength. The optical path length is the distance infrared light travels in the gas sample. Different optical path lengths affect the degree of infrared light absorption; the longer the optical path length, the greater the amount of light absorbed by the gas.

[0040] Using K identified impurity components, the pre-constructed gas spectral characteristic database is iterated one by one. For each impurity component, the corresponding molar absorption coefficient and optical path length are retrieved from the database to obtain K real-time molar absorption coefficients and K real-time optical path lengths.

[0041] The Lambert-Beer law is used to convert molar absorptivity, optical path length, and reference output concentration into absorbance, representing the degree of absorption of infrared light by a gas. The expression for the Lambert-Beer law is: A = ε·c·l, where A is absorbance, ε is the molar absorptivity, c is the reference output concentration, and l is the optical path length. For K impurities in hydrogen gas, this formula can be used to calculate K reference absorbance values, corresponding to the absorbance of the K different impurities.

[0042] K reference absorbance values ​​are used to generate K reference spectra. Typically, the spectra will have different peaks, reflecting the absorption peaks of different impurity molecules. These spectra are the spectral signals that the hydrogen purification equipment should produce under normal operating conditions with the allowable impurity content. They will serve as normal references for the composition of the output gas during the hydrogen purification process.

[0043] A Siamese neural network (SNN) is used to construct an output deviation identification model. A Siamese neural network is a special deep learning architecture that can determine the existence of deviation by comparing the similarity of two input spectra. Specifically, the Siamese neural network receives two spectra as input: the actual measured output spectrum and a reference spectrum. The network extracts features from each spectrum, generating embedding vectors. By calculating the distance between the two embedding vectors, such as Euclidean distance, their similarity is determined. If the distance is small, it indicates that the difference between the spectra is small, and the equipment is working normally; if the distance is large, it indicates that the equipment output deviates from the reference state, and the equipment may be malfunctioning. By training the Siamese neural network on a large number of reference and sample output spectra, the network can learn how to identify output deviations, ultimately obtaining an output deviation identification model for automatically determining the operating status of hydrogen purification equipment.

[0044] Furthermore, using the K reference spectra as comparison benchmarks, the output deviation identification model is constructed using a Siamese neural network. The method includes: The process involves interactively obtaining K sets of first sample spectra of the K impurity components, wherein the K sets of first sample spectra are identified by the K sets of first sample impurity concentrations; calculating the concentration deviation of the K sets of first sample impurity concentrations based on the K benchmark output concentrations to obtain K sets of sample deviation coefficients, wherein the K sets of first sample impurity concentration mappings are greater than the K benchmark output concentrations; constructing K output deviation recognition channels using a Siamese neural network, and then using the K sets of first sample spectra, the K benchmark spectra, and the K sets of sample deviation coefficients as model training data to update the model parameters of the K sets of output deviation recognition channels; connecting the K sets of output deviation recognition channels in parallel to obtain an output deviation recognition layer; calling the spectral separation layer, and cascading the spectral separation layer and the output deviation recognition layer to obtain the output deviation recognition model.

[0045] The system interactively obtains K sets of first-sample spectra of K impurity components. The acquisition methods include generating impurity sample spectra with specific concentrations through simulation experiments under specific conditions, or extracting impurity spectral data generated during previous operation from the device's historical records. Each sample spectrum contains the infrared absorption characteristics of the impurity, reflecting its spectral absorption behavior at a specific concentration. For each impurity, there is a corresponding sample spectrum. Each data point in the sample spectrum is identified by its corresponding impurity concentration, reflecting the impurity concentration in the current sample spectrum. For example, one set of sample spectra corresponds to a water vapor concentration of 10 ppm, and another set corresponds to 20 ppm.

[0046] For each impurity component, the impurity concentration in the sample spectrum is compared with its corresponding reference concentration. For example, the difference between the two is calculated as a sample bias coefficient, representing the degree to which the sample impurity concentration deviates from the reference concentration. This process is repeated for K groups of first-sample impurity concentrations to obtain K groups of sample bias coefficients. The K groups of first-sample impurity concentrations are mapped to values ​​greater than K reference output concentrations, providing necessary data support for subsequent model training and fault identification. This data can help identify faults present in the equipment.

[0047] Siamese neural networks (SiNs) are a special deep learning architecture used to compare the similarity between two input data sets. Based on SiNs, a SiN output deviation identification channel is constructed for each impurity component, resulting in a total of K channels. For each impurity component's output deviation identification channel, the first sample spectrum, the reference spectrum, and the sample deviation coefficient are received as training data. The Siamese neural network extracts features from the two sets of input spectra—the first sample spectrum and the reference spectrum—and generates feature embedding vectors. The network calculates the distance between these two feature embedding vectors to determine their difference; a larger distance indicates a greater deviation between the sample spectrum and the reference spectrum. During training, the model also incorporates the sample deviation coefficient to adjust its parameters, enabling it to better identify the difference between the actual output spectrum and the reference spectrum.

[0048] These K independent output deviation identification channels are connected in parallel to construct a unified output deviation identification layer. This layer can simultaneously process the spectra of all impurity components in the device output. If any channel is found to have a significant deviation, the layer will output a fault signal to indicate that the device is abnormal, thereby improving the overall efficiency of fault identification.

[0049] A spectral separation layer is constructed to isolate impurity signals from the actual monitored composite spectra and generate independent spectra for each impurity. The spectral separation layer and the output deviation identification layer are cascaded; that is, the output of the spectral separation layer is connected to the input of the output deviation identification layer. Specifically, after the spectral separation layer, the spectra of each separated impurity are input to the output deviation identification layer. Through this cascaded structure, the separation layer and the deviation identification layer jointly complete the spectral analysis and fault identification. The resulting deviation identification model can efficiently and accurately analyze the output of the hydrogen purification equipment and identify equipment faults.

[0050] Furthermore, prior to invoking the spectral separation layer, the method includes: Multiple composite spectra of samples and multiple sets of separated impurity spectra of samples are obtained interactively; based on the multiple composite spectra of samples and the multiple sets of separated impurity spectra of samples, impurity concentration weights are calculated to obtain multiple sets of sample concentration weights; based on the K impurity components, the multiple sets of sample concentration weights are recombined, and the mean of the recombination results is calculated to obtain K standard impurity concentration weights; the spectral separation layer is constructed based on the K standard impurity concentration weights.

[0051] In the operation of hydrogen purification equipment, the actual monitored spectra are usually the result of superposition of multiple impurity components, called composite spectra. These spectra reflect the simultaneous presence of multiple impurities in hydrogen, with the infrared absorption peaks of the impurities overlapping each other. Sample-separated impurity spectra are generated under conditions where each impurity exists independently. Through experiments or data processing, individual spectra of each impurity can be obtained, reflecting the characteristic of that impurity absorbing infrared light independently. By using data sources such as experimental measurements, simulation calculations, or historical data, composite and separated spectra of multiple impurities can be extracted, resulting in multiple sample composite spectra and multiple sets of sample-separated impurity spectra.

[0052] The composite spectrum can be viewed as a linear superposition of the spectra of each impurity, meaning that each impurity appears in the overall spectrum with a certain weight. To accurately calculate the weight of each impurity, the least squares method is used for fitting. The goal is to find a set of weights that makes the composite spectrum closest to the linear superposition of the spectra of each impurity. By minimizing the error function, the concentration weight of each impurity can be obtained, representing the concentration ratio of the impurity in the composite spectrum. Finally, multiple sets of sample concentration weights are obtained, with each set of weights corresponding to K weights for K impurities. The larger the weight value, the greater the contribution of the impurity to the composite spectrum.

[0053] The weights of each impurity in different composite spectra are extracted separately. For example, for water vapor weights in multiple sample groups, the water vapor weights in all samples are extracted separately and summarized. For each impurity, a corresponding weight set is formed, containing the weight values ​​of that impurity extracted from the composite spectra of multiple samples. The average of the weight sets of each impurity is calculated to obtain the standard impurity concentration weight of that impurity. K impurity components are traversed to obtain the corresponding K standard impurity concentration weights. These weight values ​​serve as reference standards in the model for separating composite spectra, helping to accurately separate the spectral signals of each impurity.

[0054] Based on the calculated K standard impurity concentration weights, a spectral separation layer is constructed. The task of this layer is to separate the impurity signals in the actual monitored composite spectral image according to the standard concentration weights and generate an independent spectral image for each impurity.

[0055] Furthermore, the method of synchronizing the first output spectrum to the output deviation identification model for equipment fault diagnosis includes: The first output spectrum is spectrally demixed by the spectral separation layer of the output deviation identification model to obtain K output impurity spectra; the K output impurity spectra are synchronized to the K output deviation identification channels of the output deviation identification layer for impurity deviation analysis to obtain K deviation analysis results; when the K deviation analysis results are empty sets, the output result of the output deviation identification model is in a non-fault state; when any of the K deviation analysis results is an impurity deviation coefficient, the output result of the output deviation identification model is in a fault state.

[0056] The first output spectrum is input to the spectrum separation layer. The spectrum contains composite spectral signals of K impurities. The spectrum separation layer separates the composite spectrum of the first output spectrum according to the calculated K standard impurity concentration weights, and demixes the composite signal into K independent impurity spectra. After demixing, K different output impurity spectra are obtained, and each spectra represents the spectral characteristics of one impurity.

[0057] After demixing, the resulting K output impurity spectra are synchronously transmitted to the corresponding channels in the output deviation identification layer. Each deviation identification channel receives a corresponding output impurity spectrum and compares it with the reference spectrum. The channel compares the difference between the output spectrum and the reference spectrum through similarity calculation. If the impurity concentration exceeds the reference value, the channel outputs an impurity deviation coefficient, indicating that the concentration of the impurity exceeds the allowable range; if there is no deviation, the analysis result is empty.

[0058] If the deviation analysis results of K impurities are empty sets, it means that the hydrogen purified by the equipment meets the expected purity standard and no significant impurity deviation is detected. In this case, the output deviation identification model determines that the equipment is in a non-fault state, that is, the purification function is normal and the impurity concentration in the hydrogen is within the allowable range.

[0059] If the deviation analysis result of any one of the K impurities is not empty, that is, if it contains an impurity deviation coefficient, it indicates that the purification effect of that impurity is not up to standard. In this case, the output deviation identification model judges that the equipment is in a faulty state. At this time, it indicates that there is a problem with the purification function and the equipment needs to be further inspected.

[0060] Furthermore, the method involves interactively obtaining the control parameter settings of the new energy equipment, performing equipment fault verification based on the input real-time spectrum and control parameter settings, and outputting the equipment verification result. The method includes: Interactively obtain K sets of second sample spectra of the K impurity components, wherein the K sets of second sample spectra are identified by the K sets of second sample impurity concentrations, and the K sets of second sample impurity concentrations are mapped to concentrations less than the K reference output concentrations; construct K impurity concentration identification channels using the K sets of first sample spectra, K sets of second sample spectra, K sets of first sample impurity concentrations, and K sets of second sample impurity concentrations; after obtaining an impurity concentration identification layer by connecting the K sets of K impurity concentration identification channels in parallel, configure the spectrum separation layer in front of the impurity concentration identification layer to complete the construction of the impurity concentration identification model; through the... The impurity concentration identification model analyzes the input real-time spectrum to obtain K input impurity concentrations; based on the K input impurity concentrations and control parameter settings, it predicts the output impurity concentration to obtain K predicted output concentrations; the K output impurity spectra are directly synchronized to the K impurity concentration identification channels of the impurity concentration identification model to perform impurity concentration identification, obtaining K output impurity concentrations; the internal deviation of the K predicted output concentrations and the K output impurity concentrations is evaluated based on Euclidean distance to obtain a purification deviation coefficient; if the purification deviation coefficient does not meet the preset purification deviation threshold, the equipment verification result is a fault state.

[0061] The construction process of the impurity concentration identification model is similar to that of the aforementioned output deviation identification model, and will be briefly summarized here. Specifically, K sets of second sample spectra of K impurity components are obtained interactively. These spectra can be obtained through experiments, simulations, or historical data acquisition. The second sample spectra represent the impurity spectral signals at lower concentrations. Each set of spectra has its corresponding impurity concentration label. These concentrations are lower than the reference output concentration. The reference output concentration is the highest impurity concentration allowed by the equipment during normal operation, while the concentration label of the second sample spectra reflects the situation where the impurities are below this reference value.

[0062] The first sample spectrum corresponds to the spectrum of impurities at higher concentrations, while the second sample spectrum reflects the spectrum of impurities at lower concentrations. These spectra correspond to the baseline output concentration. A corresponding impurity concentration identification channel is established for each impurity component. Its input data includes the corresponding first and second sample spectra, as well as the impurity concentrations of the first and second samples. By comparing the spectral characteristics at different concentrations, the concentration identification channel can learn the impact of impurity concentration on the spectrum. The model learns how to infer the actual impurity concentration from the spectral data using these samples. The goal is for each channel to independently identify changes in the concentration of a specific impurity, regardless of whether the impurity concentration is higher or lower than the baseline output concentration; the channel should be able to accurately predict the actual concentration. The same training is performed on K impurity components, ultimately obtaining K impurity concentration identification channels.

[0063] The obtained K impurity concentration identification channels are connected in parallel to form an overall impurity concentration identification layer. This layer can simultaneously process the concentration prediction of K types of impurities, where each channel works independently to identify the concentration of a certain impurity, and finally the concentration prediction results of each impurity are summarized.

[0064] Before the concentration identification layer, a spectral separation layer is configured. This separation layer extracts the independent spectral signal of each impurity from the composite spectrum, ensuring that each channel of the concentration identification layer receives the spectrum of its corresponding impurity, rather than a composite spectrum containing interference from other impurities. By cascading the spectral separation layer and the concentration identification layer, a complete impurity concentration identification model is formed. This model can separate the spectrum of each impurity from the composite spectrum and predict its actual concentration.

[0065] The input real-time spectrum is output as an impurity concentration identification model. The model's spectrum separation layer decomposes the composite input real-time spectrum into K independent impurity spectra. Then, the concentration identification layer performs corresponding concentration predictions based on these impurity spectra. Each impurity identification channel specifically predicts the concentration of one type of impurity, ultimately obtaining K input impurity concentrations for K impurities. These concentration values ​​reflect the impurity concentration in the hydrogen before entering the purification equipment.

[0066] The output impurity concentration is predicted based on K input impurity concentrations and control parameter settings. For example, a BP neural network is used to predict the output impurity concentration. The BP neural network is a common artificial neural network that is trained using input and output data and adjusts the weights through backpropagation of errors to learn nonlinear relationships. Before prediction, the BP model is trained using historical data to ensure that the model can accurately capture the relationship between the input gas components, control parameters, and output gas components. After training, the output concentration of each impurity in the purified hydrogen is directly predicted based on the K input impurity concentrations and control parameter settings, thus obtaining K predicted output concentrations.

[0067] K output impurity spectra are simultaneously input into K independent concentration recognition channels in the impurity concentration recognition model. Each channel analyzes the spectrum of the corresponding impurity to obtain the actual output concentration of each impurity, thus obtaining K output impurity concentrations. These concentrations reflect the treatment effect of the hydrogen purification equipment on each impurity and indicate the residual concentration of each impurity in the hydrogen.

[0068] Euclidean distance is used to measure the difference between two vectors. Here, the two vectors are K predicted output concentrations and K actual output concentrations. By calculating the Euclidean distance between the predicted and actual output concentrations, the deviation between the predicted effect and the actual effect of the purification equipment is measured. The larger the distance, the greater the difference between the purification effect and the predicted value. The purification deviation coefficient is obtained by calculating the Euclidean distance. This coefficient is used to measure the degree of matching between the predicted purification effect and the actual purification effect.

[0069] A pre-set purification deviation threshold represents the allowable error range between the predicted and actual output concentrations. As long as the deviation coefficient is within this threshold, the equipment is considered to be operating normally. The calculated purification deviation coefficient is compared with the preset threshold. If the deviation coefficient is within the threshold, it indicates that the actual purification effect is close to the predicted effect, and the purification function is normal. Conversely, if the deviation coefficient exceeds the threshold, it indicates that the actual purification effect differs significantly from the predicted value, suggesting a equipment malfunction. When the purification deviation coefficient exceeds the preset threshold, the equipment's verification result is determined to be in a fault state, indicating the need for further analysis.

[0070] Furthermore, if the equipment verification result indicates a fault state, then the faulty component of the new energy equipment is located based on the first output spectral map to obtain the target faulty component. The method includes: When the equipment verification result is a fault state, H types of impurity components are located based on the K deviation analysis results, wherein the H deviation analysis results corresponding to the H types of impurity components are impurity deviation coefficients, and H is a positive integer less than or equal to K; an impurity-component fault tree is pre-constructed, wherein the impurity-component fault tree includes K groups of purification fault-related components of the K types of impurity components; the impurity-component fault tree is traversed using the H types of impurity components to obtain H groups of purification fault-related components; the H groups of purification fault-related components are aggregated to obtain the target fault component.

[0071] Through the deviation analysis in the previous steps, a deviation result will be obtained for the purification effect of each impurity. If the deviation result of a certain impurity is a deviation coefficient, it means that the concentration of the impurity exceeds the benchmark and the purification effect is not ideal; otherwise, the deviation result is empty, which means that the concentration of the impurity meets the benchmark and the purification effect meets the standard.

[0072] When the equipment calibration result is in a fault state, the deviation analysis results with empty deviation results are filtered out, and H deviation analysis results with impurity deviation coefficients are retained. These deviation impurity components are located and denoted as H types of impurity components, where H is a positive integer less than or equal to K, representing the number of non-compliant impurities. Each impurity corresponds to a deviation coefficient, indicating that the equipment has failed to effectively purify these impurities.

[0073] During fault diagnosis, an impurity-component fault tree is constructed to analyze internal equipment problems, enabling rapid fault location. Specifically, in hydrogen purification equipment, each impurity is handled by a specific component; for example, the adsorption bed removes moisture, and the catalyst treats carbon monoxide. Therefore, the impurity-component fault tree associates each impurity component with its corresponding purification component. For K impurity components, a fault tree structure containing K sets of purification fault-associated components is pre-constructed, with each node representing an impurity associated with one or more purification components. When the deviation coefficient of a certain impurity is abnormal, the fault tree can indicate the component associated with that impurity, thereby narrowing down the scope of fault investigation.

[0074] H types of impurity components are substituted into the impurity-component fault tree and traversed to identify which components are related to the purification failure of these impurities. During the traversal of the fault tree, the purification component corresponding to each impurity is found based on the deviation analysis results of each impurity. Each impurity corresponds to a set of possible purification failure associated components, and finally, H sets of purification failure associated components are obtained. These components are related to the faulty impurities and may be the root cause of the failure to meet the purification standards.

[0075] Some components in the H-group purification fault-related components may be used interchangeably in the purification processes of different impurities. Therefore, by summarizing and cross-analyzing these components, if a component appears in the list of related components for multiple impurities, it is likely the primary source of the fault. Through aggregation analysis, the target faulty component is ultimately identified, which is the main cause of the equipment's malfunctioning purification function. This process can quickly and accurately identify the root cause of equipment failure, aiding in equipment maintenance and repair.

[0076] Furthermore, if the equipment verification result is a non-faulty state, then the control parameters of the new energy equipment are optimized and adjusted according to the K output impurity concentrations.

[0077] When the equipment calibration shows that the hydrogen purification equipment is operating normally and the concentrations of each impurity are within the allowable range, the equipment is considered to be in a non-faulty state. Even though the equipment is operating normally, its purification efficiency and operating effect can still be further improved by optimizing the equipment's control parameters. Specifically, the performance of the hydrogen purification equipment can be optimized by adjusting some key control parameters, such as temperature, pressure, gas flow rate, hydrogen flow rate, and catalyst reaction time. Based on the actual output concentrations of K impurities, the purification effect of the current equipment is analyzed. For example, if the output concentration of a certain impurity is lower than the standard but close to the upper limit, it indicates that the purification effect of the equipment can be further optimized. Through optimization algorithms, such as gradient descent or genetic algorithms, suitable combinations of control parameters are automatically found so that the equipment can operate under more efficient conditions, ensuring that the impurity concentration is as close as possible to the minimum allowable value, thus achieving optimal adjustment of control parameters.

[0078] In summary, the new energy equipment fault detection method integrating dual-end infrared spectral monitoring provided in this application has the following technical effects: By configuring first and second infrared spectrometers at the inlet and outlet of the hydrogen purification equipment, respectively, the composition of the input and output gases can be monitored simultaneously, enabling dual-end monitoring of the equipment's operation. A predefined output baseline impurity component ensures a clear standard for measuring the output gas concentration during equipment operation. This allows for the construction of an output deviation identification model, automatically comparing the actual output with the baseline standard to ensure the impurity concentration remains within acceptable limits. The second infrared spectrometer monitors the treated gas in real time, generating a first output spectrum, which is then analyzed using the deviation identification model. Automatic spectral comparison reduces manual intervention and improves the accuracy of fault diagnosis, enabling rapid and accurate determination of equipment malfunction. When a malfunction is detected, downstream equipment is automatically activated. The circuit breaker isolation ensures that the fault does not spread to other equipment, reducing the occurrence of cascading failures, and activates the first infrared spectrometer at the inlet to provide data support for further fault diagnosis. After completing the circuit breaker isolation of downstream equipment, the new energy equipment and upstream equipment continue to operate, and the input gas composition is monitored by the first infrared spectrometer to obtain the input real-time spectrum. Combined with the control parameter settings of the new energy equipment, equipment fault verification is performed. By interactively verifying the difference between the input and output spectra, it is further determined whether the fault is caused by impurities at the input end, making the judgment result more accurate. After determining that the equipment does have a fault, the faulty component is located by the first output spectrum. By accurately locating the faulty component, it is possible to avoid a complete disassembly of the equipment or a lengthy inspection, thereby shortening the fault repair time and improving the equipment operation and maintenance efficiency.

[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault detection system for new energy equipment integrating dual-end infrared spectral monitoring, characterized in that, The system includes: A spectrometer configuration module is used to configure a dual-end infrared spectrometer for the new energy equipment to be tested. The dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment. The new energy equipment is a hydrogen purification device. A deviation identification model construction module is used to predefine the output reference impurity component and construct an output deviation identification model based on the output reference impurity component; The equipment fault judgment module is used to monitor the gas composition after the new energy equipment is processed by the second infrared spectrometer in real time to generate a first output spectrum, and synchronize the first output spectrum to the output deviation recognition model to judge the equipment fault. The downstream equipment circuit breaker module is used to disconnect the downstream equipment of the new energy equipment and activate the first infrared spectrometer if the output result of the output deviation identification model is a fault state. The input real-time spectrum acquisition module is used to continue operating the new energy equipment and upstream equipment after the downstream equipment is disconnected and isolated, and to continuously monitor the gas composition entering the new energy equipment through the first infrared spectrometer to obtain the input real-time spectrum. The equipment fault verification module is used to interactively obtain the control parameter settings of the new energy equipment, and perform equipment fault verification based on the input real-time spectrum and control parameter settings, and output the equipment verification results. The fault component location module is used to locate the fault component of the new energy equipment based on the first output spectral map if the equipment verification result is a fault state, and obtain the target fault component.

2. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 1, characterized in that, The deviation identification model construction module also includes the following operation steps: Based on the impurity type, the output reference impurity components are disassembled to obtain K reference output concentrations of K impurity components; A pre-constructed gas spectral property database is provided, wherein the gas spectral property database stores multiple sample molar absorption coefficients and multiple sample optical path lengths associated with various sample impurity components; The K impurity components are used to traverse the gas spectral characteristic database to obtain K real-time molar absorption coefficients and K real-time optical path lengths; Using the K real-time molar absorption coefficients, K real-time optical path lengths, and K reference output concentrations as conversion conditions, absorbance conversion is performed using the Lambert-Beer law to obtain K reference absorbance. K reference spectra are generated based on the K reference absorbances, and the K reference spectra are used as comparison benchmarks to construct the output deviation recognition model using a Siamese neural network.

3. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 2, characterized in that, The deviation identification model construction module also includes the following operation steps: Interactively obtain K sets of first sample spectra of the K impurity components, wherein the K sets of first sample spectra are identified by the K sets of first sample impurity concentrations; Based on the K benchmark output concentrations, the concentration deviation of the impurity concentrations of the K groups of first samples is calculated to obtain the K group sample deviation coefficients, wherein the impurity concentration mapping of the K groups of first samples is greater than the K benchmark output concentrations; After constructing K output deviation recognition channels using a Siamese neural network, the K sets of first sample spectra, K reference spectra, and K sets of sample deviation coefficients are used as model training data to update the model parameters of the K output deviation recognition channels. The K output deviation recognition channels are connected in parallel to obtain the output deviation recognition layer; The spectral separation layer is invoked, and the output deviation recognition model is obtained by cascading the spectral separation layer and the output deviation recognition layer.

4. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 3, characterized in that, The deviation identification model construction module also includes the following operation steps: Interactively obtain composite spectra of multiple samples and spectra of separated impurities from multiple samples; Based on the composite spectrum of multiple samples and the spectra of multiple sets of separated impurities in the samples, the impurity concentration weights are calculated to obtain the concentration weights of multiple sets of samples. Based on the K impurity components, the concentration weights of the multiple groups of samples are recombined, and the mean of the recombination results is calculated to obtain the K standard impurity concentration weights. The spectral separation layer is constructed based on the K standard impurity concentration weights.

5. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 4, characterized in that, The equipment fault diagnosis module also includes the following operation steps: The first output spectrum is spectrally demixed by the spectral separation layer of the output deviation identification model to obtain K output impurity spectra; The K output impurity spectra are synchronized to the K output deviation identification channels of the output deviation identification layer for impurity deviation analysis to obtain K deviation analysis results; When the K deviation analysis results are an empty set, the output result of the output deviation identification model is a non-fault state; When any one of the K deviation analysis results is an impurity deviation coefficient, the output result of the output deviation identification model is a fault state.

6. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 5, characterized in that, The device fault verification module also includes the following operation steps: Interactively obtain K sets of second sample spectra of the K impurity components, wherein the K sets of second sample spectra are identified by the K sets of second sample impurity concentrations, and the K sets of second sample impurity concentrations are mapped to less than the K reference output concentrations; K impurity concentration identification channels are constructed using the K groups of first sample spectra, K groups of second sample spectra, K groups of first sample impurity concentrations, and K groups of second sample impurity concentrations. After obtaining the impurity concentration identification layer by connecting the K impurity concentration identification channels in parallel, the spectral separation layer is configured in front of the impurity concentration identification layer to complete the construction of the impurity concentration identification model. The input real-time spectrum is analyzed using the impurity concentration identification model to obtain K input impurity concentrations. Based on the K input impurity concentrations and control parameter settings, the output impurity concentration is predicted to obtain K predicted output concentrations. The K output impurity spectra are directly synchronized to the K impurity concentration recognition channels of the impurity concentration recognition model to perform impurity concentration recognition and obtain the K output impurity concentrations; The internal deviation of the K predicted output concentrations and K output impurity concentrations is evaluated based on Euclidean distance to obtain the purification deviation coefficient; If the purification deviation coefficient does not meet the preset purification deviation threshold, the equipment verification result is a fault state.

7. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 5, characterized in that, The fault component location module further includes the following operation steps: When the equipment verification result is a fault state, H types of impurity components are located based on the K deviation analysis results, wherein the H types of impurity components are the H deviation analysis results corresponding to the H types of impurity components, and H is a positive integer less than or equal to K; A pre-constructed impurity-component fault tree is provided, wherein the impurity-component fault tree includes K groups of purification fault-associated components for the K types of impurity components; The impurity-component fault tree is traversed using the H types of impurity components to obtain H groups of purification fault-related components. The target fault component is obtained by aggregating the H group of purification fault-related components.

8. The new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in claim 6, characterized in that, If the equipment verification result is a non-fault state, then the control parameters of the new energy equipment are optimized and adjusted according to the K output impurity concentrations.

9. A fault detection method for new energy equipment integrating dual-end infrared spectroscopy monitoring, characterized in that, Based on the new energy equipment fault detection system integrating dual-end infrared spectral monitoring as described in any one of claims 1-8, the method includes: A dual-end infrared spectrometer is configured for the new energy equipment to be tested, wherein the dual-end infrared spectrometer includes a first infrared spectrometer installed at the inlet of the new energy equipment and a second infrared spectrometer installed at the outlet of the new energy equipment, and the new energy equipment is a hydrogen purification device; A predefined output reference impurity component is used, and an output deviation identification model is constructed based on the output reference impurity component. The second infrared spectrometer monitors the gas composition after processing by the new energy equipment in real time to generate a first output spectrum, and synchronizes the first output spectrum to the output deviation identification model for equipment fault judgment. If the output result of the output deviation identification model is a fault state, then the downstream equipment of the new energy equipment is disconnected and isolated, and the first infrared spectrometer is activated. After the downstream equipment is disconnected and isolated, the new energy equipment and the upstream equipment continue to operate, and the gas composition entering the new energy equipment is continuously monitored by the first infrared spectrometer to obtain the input real-time spectrum. The system interactively obtains the control parameter settings of the new energy equipment, performs equipment fault verification based on the input real-time spectrum and control parameter settings, and outputs the equipment verification results. If the device verification result is a fault state, the faulty component of the new energy device is located based on the first output spectrum to obtain the target faulty component.

Images