Hydraulic system oil foreign matter traceability analysis method and system

Abstract

This invention relates to the field of hydraulic system reliability technology, specifically to a method and system for tracing and analyzing foreign matter in hydraulic system fluids. The method includes: establishing a source database; establishing a particle database; constructing a foreign matter tracing model and training it using the source database and particle database; performing electron microscopy-energy dispersive spectroscopy (EDS) on the fluid to be tested to obtain image data and elemental proportion data of particles in the fluid; inputting the image data and elemental proportion data into the trained foreign matter tracing model to obtain the source of particle detachment; and performing ferrographic analysis on the fluid to be tested to obtain the particle detachment mode. This invention, by constructing a foreign matter tracing model based on the fusion of multiple physical quantities, achieves accurate tracing of the source and determination of the detachment mode of foreign particles in the hydraulic system. This provides precise data support for material optimization and structural design of key wear parts, thereby significantly improving the operational reliability and service life of the hydraulic system.

Description

Technical Field

[0001] This invention relates to the field of hydraulic system reliability technology, and specifically to a method and system for tracing and analyzing foreign matter in hydraulic system oil. Background Technology

[0002] Hydraulic systems are the core drive and control systems for high-end equipment and key components in modern industry, playing an irreplaceable role in aerospace electromechanical systems, engineering machinery, dam gates, high-voltage circuit breakers, and other fields. As industrial equipment develops towards high-pressure, high-flow, and long-term continuous operation, the requirements for the operational stability of hydraulic systems are also increasing. In actual operation, metal particles inevitably accumulate in the friction pairs within the hydraulic system under prolonged continuous operation. These tiny particles, once inside the hydraulic fluid, not only exacerbate wear on the friction pairs and shorten the service life of key components, but may also lead to valve core jamming, seal failure, and other malfunctions. More seriously, internal leakage caused by wear directly reduces the system's efficiency and reliability. Under high-pressure, high-flow conditions, intensified wear also significantly increases the types and quantities of metal particles in the hydraulic fluid, further complicating fault diagnosis and system maintenance.

[0003] Tracing the source of solid particles in hydraulic fluid and optimizing key friction pairs accordingly has become an important way to improve the stability of hydraulic equipment. Specific methods include establishing a foreign particle source tracing model, analyzing the morphology, material, and size of the particles to determine their wear sources and detachment mechanisms, and using the analysis results to guide material improvements, surface treatments, or structural optimizations of key components, thereby achieving a rapid improvement in equipment reliability.

[0004] However, existing methods for tracing and analyzing foreign matter in hydraulic systems still have significant limitations: on the one hand, traditional methods mainly rely on offline sampling and laboratory analysis, which are not timely and make it difficult to achieve real-time monitoring and early warning of operating status; on the other hand, most tracing methods are based on a single characteristic of the particles (such as composition or morphology), which can easily lead to misjudgment in complex operating conditions and systems with multiple friction pairs, especially when the particles undergo secondary crushing or oxidation, the ambiguity of the characteristics increases, and the accuracy of tracing decreases further; in addition, existing methods often lack in-depth modeling of the relationship between particle generation mechanism and operating conditions, making it difficult for tracing results to directly guide accurate optimization design and maintenance decisions.

[0005] Therefore, there is an urgent need for a method for tracing the source of foreign matter in hydraulic system fluid and monitoring wear status that can perform real-time, accurate, and multi-dimensional integrated analysis. This method would address the problems of insufficient timeliness, single feature utilization, weak correlation of working conditions, and limited accuracy of tracing in existing technologies, thereby providing reliable technical support for targeted optimization and stability improvement of key components in hydraulic systems. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for tracing and analyzing foreign matter in hydraulic systems, in order to solve the technical problems of insufficient timeliness, single feature utilization, weak correlation with working conditions, and limited accuracy of tracing in existing hydraulic system foreign matter tracing and analysis methods.

[0007] To achieve the above objectives, embodiments of the present invention provide a method for tracing and analyzing foreign matter in hydraulic system fluid, including: Establish a source database; Establish a particle database; A foreign object source tracing model was constructed and trained using the aforementioned source database and particle database; Electron microscopy-energy dispersive spectroscopy (EDS) was performed on the oil to be tested in order to obtain image data and elemental composition data of particles in the oil. The image data and element proportion data are input into the trained foreign object tracing model to obtain the source of particle detachment. Ferrographic analysis was performed on the oil to be tested to determine the particle detachment pattern.

[0008] Optionally, establishing a source database includes: Components that may generate particles are selected as key components; Solid particulate powder of the key component was extracted and subjected to electron microscopy-energy dispersive spectroscopy to obtain particle morphology images and elemental composition.

[0009] Optionally, establishing a particle database includes: Foreign matter particles were obtained and prepared into oil particle samples; The oil particle sample was subjected to electron microscopy-energy dispersive spectroscopy to obtain the morphology image and elemental composition of the foreign particles in the oil. Based on the morphological images and elemental proportions of the foreign particles in the oil, the particle source tags in the source database are associated and labeled.

[0010] Optionally, constructing a foreign object source tracing model and training it using the aforementioned source database and particle database includes: Image data is input into an image recognition model to extract features at different scales; A feature pyramid network is used to fuse the features at different scales to obtain a multi-scale fused feature map; A multilayer perceptron is used to project the element proportion data into a dimensional space consistent with the number of channels in the multi-scale fused feature map to obtain the channel saliency weight vector. The modulated multi-scale feature map is obtained based on the channel saliency weight vector; The multi-scale feature map is used as input and trained using a convolutional neural network.

[0011] Optionally, a feature pyramid network is used to fuse the features at different scales to obtain a multi-scale fused feature map, including: Starting from the top layer, perform a convolution operation on the i-th feature map to obtain the i-th fused feature map; The i-th fused feature map is upsampled, and the upsampled feature map is added element-wise to the (i-1)-th feature map; Perform a convolution operation on the feature map after element-wise addition to obtain the (i-1)th fused feature map; Determine whether the fusion of all feature maps has been completed; If it is determined that the fusion of all feature maps has not been completed, proceed to step i-1 and return to perform the step of upsampling the i-th fused feature map and adding the upsampled feature map to the i-1-th feature map element by element.

[0012] Optionally, a multilayer perceptron is used to project the element proportion data into a dimensional space consistent with the number of channels in the multi-scale fused feature map to obtain the channel saliency weight vector, including: The element proportion data is standardized according to formula (1). (1) in, For the standardized vector, This is a vector representing the proportion of elements. Let be the mean vector of all elements in the training set. This is the standard deviation vector of each element in the training set; The original channel weight scores are obtained according to formulas (2) and (3). (2) (3) in, The activation value of the hidden layer. This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. The original channel weight score; Obtain the channel significance weight vector according to formula (4). (4) in, This is the channel significance weight vector. This is the Sigmoid activation function.

[0013] Optionally, obtaining the modulated multi-scale feature map based on the channel saliency weight vector includes: Broadcast the channel saliency weight vector along the spatial dimension; Perform element-wise multiplication according to formula (5). (5) in, This is the modulated multi-scale feature map. For multi-scale fused feature maps, This is the expanded channel saliency weight vector.

[0014] Optionally, training a convolutional neural network using the multi-scale feature map as input includes: The multi-scale feature map is passed through a convolutional layer, an activation layer, and a pooling layer to obtain a one-dimensional feature vector. The one-dimensional feature vector is mapped to a category space through a fully connected layer, the category space including the source of foreign objects.

[0015] Optionally, the analysis method further includes: The oil detection data is added to the particle database, and the foreign matter tracing model is retrained based on the expanded particle database.

[0016] On the other hand, the present invention provides a hydraulic system oil foreign matter tracing and analysis system, the system including a processor configured to perform any of the methods described above.

[0017] The beneficial effects of this invention are: This invention, through the construction of a foreign object tracing model based on the fusion of multiple physical quantities, enables precise tracing of the source and determination of the detachment mechanism of foreign particles in hydraulic systems. This technology provides accurate data support for material optimization and structural design of critical wear parts, thereby significantly improving the operational reliability and service life of hydraulic systems. Especially in high-end hydraulic equipment, malfunctions caused by foreign objects can lead to serious consequences. This invention can accurately identify the source and detachment mechanism of foreign particles in hydraulic fluid, providing a reliable basis for targeted performance improvement and preventative maintenance of key components.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1A flowchart of a hydraulic system oil foreign matter tracing and analysis method according to an embodiment of the present invention; Figure 2 This is a schematic image of the particle morphology of a key component in the source database according to an embodiment of the present invention; Figure 3 A schematic diagram of the proportion of key component particulate elements in the source database according to an embodiment of the present invention; Figure 4 This is a schematic image of the morphology of foreign particles in a particle database according to an embodiment of the present invention; Figure 5 This is a schematic diagram showing the proportion of foreign particle elements in a particle database according to an embodiment of the present invention; Figure 6 A flowchart illustrating a method for constructing and training a foreign object tracing model according to an embodiment of the present invention; Figure 7 A flowchart illustrating a method for fusing features at different scales using a feature pyramid network according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the detection results of a foreign particle ferrography instrument according to one embodiment of the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0021] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0022] like Figure 1 The diagram shows a flowchart of a hydraulic system oil foreign matter tracing and analysis method according to an embodiment of the present invention. Figure 1 In this analysis, the method may include the following steps: In step S10, a source database is established; In step S11, a particle database is established; In step S12, a foreign object tracing model is constructed and trained using a source database and a particle database; In step S13, electron microscopy-energy dispersive spectroscopy is performed on the oil to be tested to obtain image data and elemental composition data of the particles in the oil to be tested. In step S14, the image data and element proportion data are input into the trained foreign object tracing model to obtain the source of particle detachment; In step S15, ferrographic analysis is performed on the oil to be tested to obtain the particle shedding pattern.

[0023] In such Figure 1 In the hydraulic system foreign matter tracing and analysis method shown, step S10 is used to establish a source database. In this embodiment, the specific method for establishing the source database can be to first extract solid particle powder from key components that may generate particles, then observe its morphology using an electron microscope to obtain particle morphology images, and analyze it using an energy dispersive spectroscopy (EDS) instrument to obtain elemental proportions. The source database is then established using the particle morphology images and elemental proportions. The content of the source database includes the name of the source of foreign matter in the key friction pairs of the corresponding hydraulic equipment, the particle morphology of the relevant components, and the elemental proportions. For example... Figure 2 The image shown is a schematic diagram of the particle morphology of key components in the source database. For example... Figure 3 The figure shows the percentage of key component particulate elements in the source database.

[0024] In step S11, a particle database is established. In this embodiment, the specific method for establishing the particle database can be to collect a certain amount of oil from the hydraulic equipment under test and prepare it as a sample. The sample preparation method for oil particles includes: separating suspended particles from the oil using a centrifuge; washing the separated solid particles with anhydrous ethanol; dispersing the washed particles in an ethanol solution to prepare a sample suitable for analysis using scanning electron microscopy-energy dispersive spectroscopy (SEM-EDS) and ferrography. The detection results are stored in the particle database. Subsequently, the morphological images and elemental composition of the particles are obtained through SEM-EDS analysis, and a preliminary particle database is established. Figure 4 The image shown is a schematic diagram of the morphology of foreign particles in a particle database. For example... Figure 5 The figure shows the percentage of foreign particles in the particle database.

[0025] Step S12 is used to construct a foreign object tracing model and train it using a source database and a particle database. In this embodiment, the specific method for constructing and training the foreign object tracing model in step S12 can be of various forms known to those skilled in the art. In one example of the present invention, step S12 may include, for example... Figure 6 The steps shown are described. Figure 6 In this context, step S12 may include: In step S20, the image data is input into the image recognition model to extract features at different scales; In step S21, a feature pyramid network is used to fuse features at different scales to obtain a multi-scale fused feature map. In step S22, a multilayer perceptron is used to project the element proportion data onto a dimensional space consistent with the number of channels in the multi-scale fusion feature map to obtain the channel saliency weight vector. In step S23, the modulated multi-scale feature map is obtained based on the channel saliency weight vector; In step S24, the multi-scale feature map is used as input and trained using a convolutional neural network.

[0026] In such Figure 6 In the method shown, step S20 is used to extract features at different scales. In this example, a specific method for extracting features at different scales can be to input image data into ResNet101 to extract multi-layer features at different scales and semantic levels. These feature maps respectively cover shallow high-resolution texture edge details and deep low-resolution high semantic target features. The image to be processed (single-channel grayscale image, dimension 1×H×W, where H is height and W is width) is transformed into a 3-channel image (dimensional 3×H×W, where H is height and W is width) by channel duplication (i.e., the values ​​of the three channels are exactly the same) to meet the input channel requirements of ResNet101 (the first convolutional layer of ResNet101 requires the number of input channels to be 3). Traverse all grayscale images in the dataset, calculate the global mean μ and variance σ of the pixel values, and standardize them according to the following formula: (6) Where I[h,w] represents the pixel value (0~255 integers) at position (h,w) of the original single-channel grayscale image. The standardized image is input into ResNet101. The first convolutional layer (conv1) of ResNet101 uses a 7×7 kernel and a stride of 2, resulting in 64 output channels and generating a feature map of size H / 2×W / 2×64. After ReLU activation and 3×3 max pooling, the size is further reduced to a feature map of size H / 4×W / 4×64. This feature map represents the shallowest layer, preserving basic edge, texture, and color details of the image, laying the groundwork for subsequent processing.

[0027] The feature maps then undergo four residual stages: Conv2_x, Conv3_x, Conv4_x, and Conv5_x, generating feature maps with different semantic and detail characteristics: After Conv2_x (3 residual blocks), the number of channels increases from 64 to 256, generating a feature map with a scale of H / 4×W / 4. This feature map has high resolution, retains rich texture and edge details, but has weaker semantic information; after Conv3_x (4 residual blocks), the number of channels increases from 256 to 512, generating a feature map with a scale of H / 8×W / 8. The feature map resolution is reduced, and texture details are slightly compressed, beginning to include local target component features. After Conv4_x (23 residual blocks), the number of channels increases from 512 to 1024, generating a feature map with a scale of H / 16×W / 16. This feature map has enhanced semantic information, corresponding to the overall outline and component combination of the target. After Conv5_x (3 residual blocks), the number of channels increases from 1024 to 2048, generating a feature map with a scale of H / 32×W / 32. This feature map has the lowest resolution and the strongest semantic information, with features abstracted as target category and global semantics. Feature maps are extracted from the four residual stages of ResNet101 to form a feature pyramid: the top layer corresponds to Conv5_x, which has low resolution and high semantic features, used to identify the target category; the middle layer corresponds to Conv3_x and Conv4_x, which balance resolution and semantics, used to locate the target position; the bottom layer corresponds to Conv2_x, which has high resolution and low semantic features, used to refine the target edges and texture details.

[0028] Step S21 is used to fuse features at different scales using a feature pyramid network to obtain a multi-scale fused feature map. In this embodiment, the specific method for feature fusion in step S21 can be of various forms known to those skilled in the art. In one example of the present invention, step S21 may include, for example... Figure 7 The steps shown are described. Figure 7 In this context, step S21 may include: In step S30, starting from the top layer, a convolution operation is performed on the i-th feature map to obtain the i-th fused feature map; In step S31, the i-th fused feature map is upsampled, and the upsampled feature map is added element-wise to the (i-1)-th feature map; In step S32, a convolution operation is performed on the feature map after element-wise addition to obtain the (i-1)th fused feature map; In step S33, it is determined whether the fusion of all feature maps has been completed; In step S34, if it is determined that the fusion of all feature maps has not been completed, i-1 is returned to perform the step of upsampling the i-th fused feature map and adding the upsampled feature map to the i-1-th feature map element by element.

[0029] In such Figure 7 In the method shown, steps S30 to S34 are used to obtain multi-scale fused feature maps layer by layer from top to bottom. Specifically, in this example, the original feature layers can be extracted from the four residual stages of ResNet101 and defined as downsampling layers of the basic feature pyramid: C2 corresponds to Conv2_x, C3 corresponds to Conv3_x, C4 corresponds to Conv4_x, and C5 corresponds to Conv5_x. Using a top-down process, the small feature maps of the upper layers are magnified to the adjacent large feature maps of the lower layers through upsampling. Figure 1 The sample size is determined using nearest neighbor interpolation for upsampling. Starting from the top layer C5, a 1×1 convolution operation is performed on C5 to change the number of channels in the feature map (compressing it from 2048 to 256; subsequent 1×1 convolution operations also compress the number of channels to 256), resulting in M5. M5 is upsampled and added to C4 (each element at the same position in the feature map is directly added, which is the feature fusion process), and then subjected to another 1×1 convolution to obtain M4. Similarly, M4 is upsampled and added to C3, and then convolved to obtain M3, and so on to obtain M2. Then, a 3×3 convolution is applied to the M-layer feature map to mitigate the aliasing effect caused by nearest neighbor interpolation, resulting in the final multi-scale fused feature map containing features from layers P2, P3, P4, and P5.

[0030] Step S22 involves using a multilayer perceptron to project the element proportion data into a dimensional space consistent with the number of channels in the multi-scale fused feature map, thereby obtaining the channel saliency weight vector. In this example, the specific method for obtaining the channel saliency weight vector may include: In step S40, the element proportion data is standardized; In step S41, the original channel weight scores are obtained; In step S42, the channel saliency weight vector is obtained.

[0031] Step S40 is used to standardize the element proportion data. Specifically, it can be done, for example, by using formula (1) to standardize the element proportion data: (1) in, For the standardized vector, This is a vector representing the proportion of elements. Let be the mean vector of all elements in the training set. Let be the standard deviation vector of each element in the training set.

[0032] Step S41 is used to obtain the original channel weight score based on the standardized element proportion data. Specifically, in this example, the original channel weight score can be obtained by formula (2) and formula (3): (2) (3) in, The activation value of the hidden layer. This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. This represents the original channel weight score.

[0033] Step S42 is used to obtain the channel saliency weight vector. Specifically, in this example, the channel saliency weight vector can be obtained, for example, using formula (4): (4) in, This is the channel significance weight vector. This is the Sigmoid activation function.

[0034] Step S23 is used to obtain the modulated multi-scale feature map based on the channel saliency weight vector. Specifically, in this example, the channel saliency weight vector can be broadcast and extended along the spatial dimension, and then multiplied element-wise according to formula (5). (5) in, This is the modulated multi-scale feature map. For multi-scale fused feature maps, This is the expanded channel saliency weight vector.

[0035] Step S24 is used to train a convolutional neural network by taking the multi-scale feature map as input. In this example, the specific method for training the convolutional neural network may include: In step S50, the multi-scale feature map is passed through a convolutional layer, an activation layer, and a pooling layer to obtain a one-dimensional feature vector; In step S51, the one-dimensional feature vector is mapped to the category space through a fully connected layer, and the category space includes the source of foreign objects.

[0036] Step S50 is used to obtain a one-dimensional feature vector, which can be achieved by average pooling according to the following formula: (7) in, This indicates the rectangular region related to the k-th feature map. The average pooling output value, Represents a rectangular area Central The element at that location, Represents a rectangular area The number of elements in the middle.

[0037] Step S51 is used to map the pooled one-dimensional feature vector to the category space through a fully connected layer. The category space consists of the foreign object source names in the source database.

[0038] Step S13 involves performing electron microscopy-energy dispersive spectroscopy (EDS) on the oil to be tested to obtain image data and elemental composition data of the particles in the oil. Specifically, in this example, the oil in the hydraulic system under test can be sampled first, and the oil sample can be divided into two equal parts (50 mL each) for electron microscopy-EDS analysis and ferrography analysis, respectively, and samples can be prepared. First, the sample is analyzed by electron microscopy-EDS. The obtained particle morphology image and elemental composition are input into the foreign matter source tracing model in step S14 for judgment. The model will output the corresponding detachment source. The detachment source is related to the specific hydraulic equipment, which is the detachment location of the particles in the specific hydraulic equipment, such as the piston, cylinder wall guide ring, etc. Then, the judgment result can be used to further expand the sample in the particle database, thereby improving the accuracy of the model. At this point, the accurate judgment of the source of foreign matter detachment in the oil is completed.

[0039] Then, in step S15, the sampled oil was subjected to ferrographic analysis, such as... Figure 8 The diagram shows the results of a ferrography analysis of foreign matter particles. Standard SH / T0573-1993 specifies a qualitative analysis method for the ferrographic shedding patterns, used to qualitatively determine the shedding mechanism of foreign matter particles. The shedding pattern is determined by combining ferrographic analysis with the particle shedding analysis. Specific shedding patterns include normal abrasive particles, severely abrasive particles, cutting abrasive particles, spalling, layered abrasive particles, spherical particles, dark metal oxides, red oxides of iron, corrosive abrasive particles, non-ferrous metal abrasive particles (copper), non-metallic particles with refractive effects, friction polymers, fibers, etc. Based on this foreign matter source tracing model, combined with the source and shedding pattern of foreign matter in the oil, technicians can quickly analyze the easily worn components of the tested hydraulic system and perform targeted optimization.

[0040] On the other hand, the present invention provides a hydraulic system oil foreign matter tracing and analysis system, the system including a processor configured to perform any of the methods described in the hydraulic system oil foreign matter tracing and analysis method.

[0041] The beneficial effects of this invention are: This invention, through the construction of a foreign object tracing model based on the fusion of multiple physical quantities, enables precise tracing of the source and determination of the detachment mechanism of foreign particles in hydraulic systems. This technology provides accurate data support for material optimization and structural design of critical wear parts, thereby significantly improving the operational reliability and service life of hydraulic systems. Especially in high-end hydraulic equipment, malfunctions caused by foreign objects can lead to serious consequences. This invention can accurately identify the source and detachment mechanism of foreign particles in hydraulic fluid, providing a reliable basis for targeted performance improvement and preventative maintenance of key components.

[0042] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0043] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0046] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0047] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0048] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0049] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0050] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for tracing and analyzing foreign matter in hydraulic system fluid, characterized in that, The analytical method includes: Establish a source database; Establish a particle database; Constructing a foreign object source tracing model and training it using the aforementioned source database and particle database includes: inputting image data into an image recognition model to extract features at different scales; A feature pyramid network is used to fuse the features at different scales to obtain a multi-scale fused feature map; A multilayer perceptron is used to project the element proportion data into a dimensional space consistent with the number of channels in the multi-scale fused feature map to obtain the channel saliency weight vector. The modulated multi-scale feature map is obtained based on the channel saliency weight vector; The multi-scale feature map is used as input and trained using a convolutional neural network; Specifically, a multilayer perceptron is used to project the element proportion data onto a dimensional space consistent with the number of channels in the multi-scale fused feature map to obtain the channel saliency weight vector, including: The element proportion data is standardized according to formula (1). ，（1） in, For the standardized vector, This is a vector representing the proportion of elements. Let be the mean vector of all elements in the training set. This is the standard deviation vector of each element in the training set; The original channel weight scores are obtained according to formulas (2) and (3). ，（2） ，（3） in, The activation value of the hidden layer. This is the first layer weight matrix. This is the first layer bias vector. This is the weight matrix for the second layer. This is the second layer bias vector. The original channel weight score; Obtain the channel significance weight vector according to formula (4). ，（4） in, This is the channel significance weight vector. Use the Sigmoid activation function; Electron microscopy-energy dispersive spectroscopy (EDS) was performed on the oil to be tested in order to obtain image data and elemental composition data of particles in the oil. The image data and element proportion data are input into the trained foreign object tracing model to obtain the source of particle detachment. Ferrographic analysis was performed on the oil to be tested to determine the particle detachment pattern.

2. The analytical method according to claim 1, characterized in that, Establishing the source database includes: Components that may generate particles are selected as key components; Solid particulate powder of the key component was extracted and subjected to electron microscopy-energy dispersive spectroscopy to obtain particle morphology images and elemental composition.

3. The analytical method according to claim 1, characterized in that, Establishing a particle database includes: Foreign matter particles were obtained and prepared into oil particle samples; The oil particle sample was subjected to electron microscopy-energy dispersive spectroscopy to obtain the morphology image and elemental composition of the foreign particles in the oil. Based on the morphological images and elemental proportions of the foreign particles in the oil, the particle source tags in the source database are associated and labeled.

4. The analytical method according to claim 1, characterized in that, The feature pyramid network is used to fuse features at different scales to obtain a multi-scale fused feature map, including: Starting from the top layer, perform a convolution operation on the i-th feature map to obtain the i-th fused feature map; The i-th fused feature map is upsampled, and the upsampled feature map is added element-wise to the (i-1)-th feature map; Perform a convolution operation on the feature map after element-wise addition to obtain the (i-1)th fused feature map; Determine whether the fusion of all feature maps has been completed; If it is determined that the fusion of all feature maps has not been completed, proceed to step i-1 and return to perform the step of upsampling the i-th fused feature map and adding the upsampled feature map to the i-1-th feature map element by element.

5. The analytical method according to claim 1, characterized in that, Obtaining the modulated multi-scale feature map based on the channel saliency weight vector includes: Broadcast the channel saliency weight vector along the spatial dimension; Perform element-wise multiplication according to formula (5). ，（5） in, This is the modulated multi-scale feature map. For multi-scale fused feature maps, This is the expanded channel saliency weight vector.

6. The analytical method according to claim 1, characterized in that, Using the multi-scale feature map as input, training a convolutional neural network includes: The multi-scale feature map is passed through a convolutional layer, an activation layer, and a pooling layer to obtain a one-dimensional feature vector. The one-dimensional feature vector is mapped to a category space through a fully connected layer, the category space including the source of foreign objects.

7. The analytical method according to claim 1, characterized in that, The analytical method further includes: The oil detection data is added to the particle database, and the foreign matter tracing model is retrained based on the expanded particle database.

8. A hydraulic system oil foreign matter tracing and analysis system, characterized in that, The system includes a processor configured to perform the method as described in any one of claims 1 to 7.

Images