Gearbox lubrication performance evaluation method based on national base oil endoscopy detection

Abstract

This invention relates to the field of gearbox lubrication assessment and localization technology, and particularly to a method for assessing gearbox lubrication performance using endoscopic inspection based on domestically produced base oil. The method involves acquiring multi-dimensional data by obtaining polarized light images, transient torque, and high-frequency vibration data streams of the microscopic region of the gear surface during operation. Polarization reflection information is used to decouple the microscopic oil film thickness topology matrix, and backscattering characteristics are combined to calculate the proportion of local shading area. A transient elastohydrodynamic shear strain field is generated through time-domain alignment, and initial degradation parameters are output from a physicochemical feature mapping library. Correction coefficients are generated by combining high-frequency vibration characteristics, and a composite dynamic decay state tensor is output through tensor compensation. Finally, a prediction network is used to fit the degradation curve, calculate the remaining safe operating span, and convert it into control commands. This invention solves the problems of in-situ quantification of oil film state and lack of dedicated models, supporting precise operation and maintenance using domestically produced base oil.

Description

Technical Field

[0001] This invention relates to the field of gearbox lubrication assessment and localization technology, and in particular to a method for assessing gearbox lubrication performance using an endoscope based on domestically produced base oil. Background Technology

[0002] In the wind power industry, existing heavy-duty gearbox lubrication performance testing and evaluation technologies primarily rely on the monitoring and analysis of multiple physicochemical indicators of lubricating oil. The core assessment of lubrication performance depends on viscosity testing, which reflects the lubricating oil's fluidity and film-forming ability. Contamination monitoring quantifies the concentration of solid particulate impurities, directly correlated with the wear state of friction pairs. Wear particle analysis uses ferrography or spectroscopy to distinguish the morphology and elemental composition of wear debris, thereby tracing the type and location of wear. Furthermore, indicators such as water content and total acid value can determine the degree of oil oxidation and deterioration. By continuously tracking and trend analysis of these key parameters, it is possible to perform final assessments of the actual lubrication performance and remaining service life of the lubricating oil under extreme loads and complex operating conditions, providing a basis for preventative maintenance of the gearbox.

[0003] Existing technologies for testing and evaluating the lubrication performance of heavy-duty gearboxes suffer from the following technical challenges: Firstly, existing offline disassembly and oil sampling physicochemical analysis methods inevitably require equipment shutdown, and the sampling process is highly susceptible to external contamination, making it impossible to quantitatively extract core characteristics such as the microscopic oil film state on the gear surface while the equipment is running. Secondly, conventional industrial endoscopes only possess macroscopic defect observation capabilities and do not fully integrate the differences in physicochemical properties of various domestically produced base oils, such as coal-based PAO and hydrotreated PAO, to construct a dedicated quantitative evaluation model. Furthermore, they do not introduce a dynamic correction mechanism for oil film strength under complex operating conditions such as high load and wide temperature variations. These deficiencies in monitoring methods and analytical dimensions combine to directly render maintenance personnel completely unable to... Accurately tracking the dynamic degradation of lubrication performance of domestically produced base oils under complex and variable operating conditions without shutting down the system leads to a situation where oil change intervals are set entirely based on human experience and blindly executed, resulting in a serious disconnect between maintenance decisions and the actual operational needs of the equipment. For example, heavy-duty wind turbine gearboxes are in a high-altitude variable wind load environment for a long time. If a fixed oil change interval is used for daily maintenance, not only will coal-based PAO base oils with excellent oxidation stability and that have not yet failed be discarded and replaced prematurely, but hydrogenated base oils that have already shown microscopic oil film rupture and lubrication failure under extreme high temperature and high load conditions will not receive timely warnings and replacements. Ultimately, this will inevitably induce severe tooth surface scuffing or even cause the entire gearbox transmission to fail. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for evaluating the lubrication performance of gearboxes using endoscopic testing with domestically produced base oils. This invention solves the problem that existing offline disassembly and conventional physicochemical analysis cannot quantify the microscopic oil film state on the gear surface in situ. Furthermore, conventional endoscopic techniques lack specific evaluation models and dynamic operating condition adaptation mechanisms for the physicochemical differences of different domestically produced base oils, such as coal-based PAO and hydrotreated base oils. This results in the inability to accurately track the dynamic decay process of lubrication performance of domestically produced base oils under complex operating conditions without shutting down the system, leading to blind oil change cycles and a serious disconnect between maintenance decisions and actual equipment requirements. The domestically produced base oil refers to a lubricating medium with pre-defined consistent physicochemical properties.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: The present invention provides a method for evaluating the lubrication performance of gearboxes using endoscopic examination based on domestically produced base oils, comprising: Step 1: Acquire the polarized light sequence image data stream of the micro-region of the gear tooth surface inside the gearbox under operation, collect the transient torque data stream and high-frequency vibration data stream during the operation of the gearbox spindle, and retrieve the preset molecular-level type parameters; Step 2: Extract the orthogonal polarization reflection information of the grating from the polarized light sequence image data stream, and generate a microscopic oil film thickness topology distribution matrix based on the orthogonal polarization reflection information of the grating; extract the backlight scattering feature vector from the polarized light sequence image data stream, and calculate the local occlusion area ratio based on the backlight scattering feature vector. Step 3: Align the microscopic oil film thickness topology distribution matrix with the transient torque data stream in the time domain, extract the topological deformation rate of the microscopic oil film thickness topology distribution matrix at the moment corresponding to the peak change in the transient torque data stream, and generate the transient elastohydrodynamic shear strain field based on the topological deformation rate; compare the molecular-level type parameters, microscopic oil film thickness topology distribution matrix, local shading area ratio, and transient elastohydrodynamic shear strain field with the preset optical rheological mapping library, and output the initial degradation benchmark parameters based on the comparison results; Step 4: Extract the frequency domain features of vibration energy and the time domain fluctuation features of load from the high-frequency vibration data stream, and fuse the frequency domain features of vibration energy and the time domain fluctuation features of load to generate dynamic compensation correction coefficients; use the dynamic compensation correction coefficients to perform high-dimensional tensor compensation calculation on the initial deterioration reference parameters, and output the composite dynamic decay state tensor. Step 5: Input the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor into a pre-established time-varying fluid dynamics prediction network with physical monotonicity constraints, and output the microscopic lubrication degradation evolution curve; extract the remaining time span corresponding to the point when the microscopic lubrication degradation evolution curve breaks through the preset critical friction failure threshold; and convert the remaining time span into an oil change cycle control command for external output.

[0006] Furthermore, the method for evaluating the lubrication performance of a gearbox using endoscopic inspection with domestically produced base oils provided by this invention acquires a polarized light sequence image data stream of the microscopic bearing area on the tooth surface inside the gearbox during operation, including: A dual-band polarized light beam is emitted toward the micro-region bearing the tooth surface. The dual-band polarized light beam penetrates the lubricating medium on the surface of the micro-region bearing the tooth surface and generates an optical echo through reflection and backscattering by the micro-region bearing the tooth surface. It receives optical echoes and converts them into continuous analog electrical signals; Analog-to-digital conversion is performed on the analog electrical signal to generate a polarized light sequence image data stream.

[0007] Furthermore, the method for evaluating the lubrication performance of a gearbox using endoscopic inspection with domestically produced base oil provided by this invention extracts orthogonal polarization reflection information from a polarized light sequence image data stream, and generates a microscopic oil film thickness topology distribution matrix based on this orthogonal polarization reflection information, including: Two-dimensional phase unwrapping and background grayscale suppression processing are performed on the polarized light sequence image data stream to remove ambient light noise data and separate the orthogonal polarization reflection information of the grating; Extract the amplitude variation characteristics of phase delay of the corresponding band in the orthogonal polarization reflection information of the grating; The phase delay amplitude variation characteristics are mapped to the microscopic sight path difference value; The microscopic path difference value is converted into a microscopic oil film thickness topological distribution matrix based on the refractive index of the lubricating medium.

[0008] Furthermore, the method for evaluating the lubrication performance of a gearbox using endoscopic inspection with domestically produced base oils provided by this invention extracts the backscattering feature vector from the polarized light sequence image data stream and calculates the proportion of local occlusion area based on the backscattering feature vector, including: Spatial frequency filtering is performed on the polarized light sequence image data stream to extract the backlight scattering feature vector; Based on the backlight scattering feature vector, discrete pixel clusters with scattered light intensity deviating from the reference smooth light intensity are identified; Calculate the area ratio of discrete pixel clusters in the total number of pixels in the overall field of view, and output the proportion of local occlusion area.

[0009] Furthermore, the method for evaluating the lubrication performance of a gearbox using endoscopic examination with domestically produced base oil provided by this invention aligns the microscopic oil film thickness topological distribution matrix with the transient torque data stream in the time domain, extracts the topological deformation rate of the microscopic oil film thickness topological distribution matrix at the moment corresponding to the peak abrupt change in the transient torque data stream, and generates a transient elastohydrodynamic shear strain field based on the topological deformation rate, including: The microscopic oil film thickness topology distribution matrix and the transient torque data stream are placed in the same global timestamp coordinate system; The moment when the peak change occurs in the transient torque data stream is taken as the corresponding moment of the peak change. Extract the topological deformation rate of the microscopic oil film thickness topological distribution matrix at the time corresponding to the peak abrupt change; The transient elastohydrodynamic shear strain field of the lubricating medium in the bearing micro-region of the tooth surface is derived based on the topological deformation rate.

[0010] Furthermore, the method for evaluating the lubrication performance of gearboxes using endoscopic testing of domestically produced base oils provided by this invention compares molecular-level type parameters, microscopic oil film thickness topological distribution matrix, local shading area ratio, and transient elastohydrodynamic shear strain field input into a pre-set optical rheological mapping library. Based on the comparison results, it outputs initial degradation benchmark parameters, including: When the molecular-level type parameter characterizes the first base oil with the preset physicochemical properties, the corresponding molecular chain conformation uncoiling optical birefringence critical curve in the optical rheological mapping library is retrieved. The coordinates of the transient elastohydrodynamic shear strain field are projected onto the critical curve of optical birefringence of molecular chain conformation decoupling. Based on the preset normalization rules, the spatial Euclidean distance between the projection point and the microrheological yield limit is calculated. The spatial Euclidean distance is specified as the shear-resistant degradation irreversible depolymerization dispersion, and the shear-resistant degradation irreversible depolymerization dispersion is output as the initial degradation baseline parameter.

[0011] Furthermore, the method for evaluating the lubrication performance of gearboxes using endoscopic testing with domestically produced base oils provided by this invention outputs initial degradation baseline parameters based on comparison results, including: When the molecular-level type parameter characterizes the second base oil with preset physicochemical properties, the corresponding detergent dispersant micelle aggregation optical dispersion threshold in the optical rheological mapping library is retrieved. The proportion of the local shielding area is multiplied by the transient elastohydro-shear strain field to obtain the result of the nonlinear cross-product operation. Based on the preset normalization rule, the difference between the result of the nonlinear cross-product operation and the optical dispersion threshold of the micelle agglomeration of the cleaning and dispersing agent is calculated. The difference is specified as a nonlinear oxidative carbon deposition accumulation gradient, and the nonlinear oxidative carbon deposition accumulation gradient is output as the initial degradation baseline parameter.

[0012] Furthermore, the method for evaluating the lubrication performance of gearboxes using endoscopic inspection with domestically produced base oils provided by this invention extracts the frequency domain characteristics of vibration energy and the time domain fluctuation characteristics of load from a high-frequency vibration data stream, and fuses the frequency domain characteristics of vibration energy and the time domain fluctuation characteristics of load to generate a dynamic compensation correction coefficient, including: The peak value of the main frequency resonance energy is extracted from the high-frequency vibration data stream as the frequency domain feature of the vibration energy, and the impact load envelope fluctuation rate of the micro-region of the tooth surface bearing is extracted as the time domain fluctuation feature of the load. The peak energy of the dominant frequency resonance is converted into a thermodynamic catalytic degradation correction coefficient. The impact load envelope fluctuation rate of the micro-region bearing the tooth surface is converted into a physical-mechanical fatigue spalling correction coefficient. The thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient are combined to generate a dynamic compensation correction coefficient.

[0013] Furthermore, the method for evaluating the lubrication performance of a gearbox using endoscopic inspection with domestically produced base oils provided by this invention utilizes the dynamic compensation correction coefficients generated in step 4 to perform high-dimensional tensor compensation calculations on the initial degradation baseline parameters, outputting a composite dynamic decay state tensor, including: High-dimensional tensor multiplication compensation operation was performed on the initial deterioration baseline parameter using the thermodynamic catalytic degradation correction coefficient and the physical mechanical fatigue spalling correction coefficient to obtain the high-dimensional tensor multiplication compensation operation value. The numerical values ​​of the dot product compensation operation of the high-dimensional tensor are concatenated to output a composite dynamic decay state tensor.

[0014] Furthermore, the method for evaluating the lubrication performance of gearboxes using endoscopic inspection with domestically produced base oils provided by this invention inputs the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor into a pre-established time-varying fluid dynamics prediction network with physical monotonicity constraints, outputting a microscopic lubrication degradation evolution curve; extracting the remaining time span corresponding to when the microscopic lubrication degradation evolution curve breaks through a preset critical friction failure threshold; and converting the remaining time span into an oil change cycle control command for external output, including: The microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are received within the time-varying fluid dynamics prediction network. The long short-term memory convolutional layer of the time-varying hydrodynamic prediction network is used to perform nonlinear forward inference fitting on the micro oil film thickness topology distribution matrix and the composite dynamic decay state tensor, and outputs the micro lubrication degradation evolution curve in future continuous time nodes. The detection algorithm is executed to scan the micro-lubrication degradation evolution curve downward along the time axis, and to locate the absolute time coordinate corresponding to the point when the micro-lubrication degradation evolution curve breaks through the preset critical friction failure threshold. The remaining time span is obtained by subtracting the absolute time coordinate from the current running time. The remaining time span is converted into an oil change cycle control command, and the oil change cycle control command is output.

[0015] Beneficial effects of this invention: This invention achieves in-situ, non-contact, precise decoupling of the microscopic oil film thickness on the tooth surface through polarized light sequence image data streams. Combined with the microscopic oil film thickness topological distribution matrix, it can intuitively map the three-dimensional distribution deformation of the lubricating medium on the actual meshing surface, overcoming the technical limitation of offline sampling physicochemical analysis in quantifying the microscopic lubrication state of the tooth surface. By introducing molecular-level type parameters and combining them with a pre-built optical rheological mapping library, the evaluation process possesses adaptive analytical capabilities for the physicochemical differences of different base oils, such as coal-based polyalphaolefins and hydrogenated oils, thereby improving the objectivity and adaptability of the evaluation conclusions of domestically produced lubrication systems. The time-domain dynamic alignment mechanism established using transient torque data streams and transient elastohydrodynamic shear strain fields can capture lubrication stability fluctuations under extreme extrusion loads. Combined with dynamic compensation correction coefficients calculated from high-frequency vibration data streams, it effectively offsets monitoring interference caused by complex wind power operating conditions, significantly enhancing the characterization accuracy of the composite dynamic decay state tensor for the true degree of oil degradation. By fitting the microscopic lubrication degradation evolution curve generated by the nonlinear time-varying elastohydrodynamic prediction network, a logical leap from transient sampling data to long-term performance trends is achieved. The extracted remaining safe operating time span provides the operation and maintenance end with oil change cycle control instructions based on real-time performance decay laws, effectively avoiding resource waste or lubrication failure risks caused by experience-based oil changes, and significantly improving the long-term intelligent operation and maintenance reliability of heavy-duty gearboxes. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of the method for evaluating the lubrication performance of gearboxes using endoscopic testing based on domestically produced base oil, as described in this invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] Please see Figure 1 The present invention provides a method for evaluating the lubrication performance of gearboxes using endoscopic examination based on domestically produced base oils, comprising: Step 1: Acquire the polarized light sequence image data stream of the micro-region of the gear tooth surface inside the gearbox under operation, collect the transient torque data stream and high-frequency vibration data stream during the operation of the gearbox spindle, and retrieve the preset molecular-level type parameters; Step 2: Extract the orthogonal polarization reflection information of the grating from the polarized light sequence image data stream, and generate a microscopic oil film thickness topology distribution matrix based on the orthogonal polarization reflection information of the grating; extract the backlight scattering feature vector from the polarized light sequence image data stream, and calculate the local occlusion area ratio based on the backlight scattering feature vector. Step 3: Align the microscopic oil film thickness topology distribution matrix with the transient torque data stream in the time domain, extract the topological deformation rate of the microscopic oil film thickness topology distribution matrix at the moment corresponding to the peak change in the transient torque data stream, and generate the transient elastohydrodynamic shear strain field based on the topological deformation rate; compare the molecular-level type parameters, microscopic oil film thickness topology distribution matrix, local shading area ratio, and transient elastohydrodynamic shear strain field with the preset optical rheological mapping library, and output the initial degradation benchmark parameters based on the comparison results; Step 4: Extract the frequency domain features of vibration energy and the time domain fluctuation features of load from the high-frequency vibration data stream, and fuse the frequency domain features of vibration energy and the time domain fluctuation features of load to generate dynamic compensation correction coefficients; use the dynamic compensation correction coefficients to perform high-dimensional tensor compensation calculation on the initial deterioration reference parameters, and output the composite dynamic decay state tensor. Step 5: Input the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor into a pre-established time-varying fluid dynamics prediction network with physical monotonicity constraints, and output the microscopic lubrication degradation evolution curve; extract the remaining time span corresponding to the point when the microscopic lubrication degradation evolution curve breaks through the preset critical friction failure threshold; and convert the remaining time span into an oil change cycle control command for external output.

[0020] In the application scenarios of domestically produced base oil replacement and long-term intelligent operation and maintenance of heavy-duty gearboxes in wind power and industrial transmission fields, domestically produced base oil (also known as a lubricating medium with consistent physicochemical properties across multiple categories), specifically oil types (such as "coal-based polyalphaolefin"), refers to a lubricating medium with a first preset physicochemical property, preferably a coal-based polyalphaolefin (PAO) lubricating medium in this embodiment; the second base oil refers to a lubricating medium with a second preset physicochemical property, preferably a hydrogenated lubricating medium in this embodiment. The first and second base oils are collectively referred to as domestically produced base oils, and their molecular-level type parameters are calibrated through a specific synthesis or hydrogenation process. Obtaining a polarized light sequence image data stream of the microscopic region of the gearbox's internal tooth surface under operating conditions requires a rigorous photoelectric signal conversion process. During testing, a dual-band polarized light beam is emitted towards the micro-region of the gear tooth surface. This beam penetrates the lubricating medium on the surface of this region and generates optical echoes through reflection and backscattering. These echoes are received and converted into continuous analog electrical signals. Analog-to-digital conversion is then performed on these signals to generate a polarized light sequence image data stream. Simultaneously, transient torque and high-frequency vibration data streams from the gearbox's main shaft are acquired. The physical signals of the main shaft's rotational deformation and the gearbox's high-frequency oscillations are continuously captured. These signals are amplified and filtered for noise reduction, resulting in digitized transient torque and high-frequency vibration data streams. After acquiring the underlying data, preset molecular-level type parameters are retrieved, and the molecular-level type parameters of the domestically produced base oil being used are parsed from the data stored in the execution terminal.

[0021] Extracting orthogonal polarization reflection information from the polarized light sequence image data stream is a fundamental step in constructing oil film features. Two-dimensional phase unwrapping and background grayscale suppression are performed on the polarized light sequence image data stream to remove ambient light noise and separate the orthogonal polarization reflection information. A microscopic oil film thickness topology distribution matrix is ​​then generated based on this information. During matrix construction, the phase delay amplitude variation characteristics of the corresponding bands of the dual-band polarized beams in the orthogonal polarization reflection information are extracted. These phase delay amplitude variation characteristics are mapped to microscopic path difference values, which are then converted into a microscopic oil film thickness topology distribution matrix based on the refractive index of the lubricating medium. In the parallel data processing branch, backlight scattering feature vectors are extracted from the polarized light sequence image data stream. Spatial frequency filtering is performed on the polarized light sequence image data stream to extract the backlight scattering feature vectors, and the local occlusion area ratio is calculated based on these backlight scattering feature vectors. In the specific proportion calculation stage, discrete pixel clusters that deviate from the reference smooth scattered light intensity are identified based on the back optical scattering feature vector, and the area ratio of discrete pixel clusters in the total number of pixels in the overall field of view is calculated to output the proportion of local occlusion area.

[0022] Aligning the microscopic oil film thickness topology distribution matrix with the transient torque data stream in the time domain requires a unified time reference. The microscopic oil film thickness topology distribution matrix and the transient torque data stream are placed in the same global timestamp coordinate system, and the moment when the transient torque data stream experiences a peak abrupt change is taken as the corresponding moment of the peak abrupt change. For the key time nodes after alignment, the topological deformation rate of the microscopic oil film thickness topology distribution matrix at the moment corresponding to the peak abrupt change in the transient torque data stream is extracted. Based on the topological deformation rate, a transient elastohydrodynamic shear strain field is generated, and the transient elastohydrodynamic shear strain field of the lubricating medium in the bearing micro-region of the tooth surface is derived. After completing the physical field decoupling, the molecular-level type parameters, the microscopic oil film thickness topology distribution matrix, the proportion of local shading area, and the transient elastohydrodynamic shear strain field are compared with a pre-set optical rheological mapping library. Based on the comparison results, the initial degradation baseline parameters are output. When the molecular-level type parameter characterizes the first base oil with preset physicochemical properties, the corresponding molecular chain conformational uncoiling optical birefringence critical curve is retrieved from the optical rheological mapping library. The coordinate points of the transient elastohydrodynamic shear strain field are projected onto the molecular chain conformational uncoiling optical birefringence critical curve. Based on the preset normalization rule, the spatial Euclidean distance between the projection point and the micro-rheological yield limit is calculated. The spatial Euclidean distance is designated as the irreversible depolymerization dispersion against shear degradation, and the irreversible depolymerization dispersion against shear degradation is output as the initial degradation baseline parameter. ... When the type parameter characterizes the second base oil with preset physicochemical properties, the corresponding detergent dispersant micelle aggregation optical dispersion threshold is retrieved from the optical rheological mapping library. The proportion of local shading area and the transient elastohydrodynamic shear strain field are subjected to nonlinear cross-product operation to obtain the nonlinear cross-product operation result. Based on the preset normalization rule, the difference between the nonlinear cross-product operation result and the detergent dispersant micelle aggregation optical dispersion threshold is calculated. The difference is specified as the nonlinear oxidation carbon deposition accumulation gradient and the nonlinear oxidation carbon deposition accumulation gradient is output as the initial degradation benchmark parameter.

[0023] Extracting the frequency domain features of vibration energy and the time domain fluctuation features of load from a high-frequency vibration data stream requires distinguishing physical quantities of different dimensions. The peak value of the dominant frequency resonance energy is extracted as the frequency domain feature of vibration energy, and the envelope fluctuation rate of the impact load in the micro-region of the tooth surface is extracted as the time domain fluctuation feature of load. To fuse the frequency domain features of vibration energy and the time domain fluctuation features of load to generate dynamic compensation correction coefficients, the peak value of the dominant frequency resonance energy is converted into a thermodynamic catalytic degradation correction coefficient, and the envelope fluctuation rate of the impact load in the micro-region of the tooth surface is converted into a physical-mechanical fatigue spalling correction coefficient. The thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient are then combined to generate dynamic compensation correction coefficients. Based on the generated compensation coefficients, high-dimensional tensor compensation calculations are performed on the initial deterioration reference parameters using the dynamic compensation correction coefficients generated in step 4 to output a composite dynamic decay state tensor. High-dimensional tensor dot product compensation operations are then performed on the initial deterioration reference parameters using the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient to obtain high-dimensional tensor dot product compensation operation values. These high-dimensional tensor dot product compensation operation values ​​are then concatenated to output a composite dynamic decay state tensor.

[0024] The microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are input into a pre-established time-varying hydrodynamic prediction network with physical monotonicity constraints, outputting a microscopic lubrication degradation evolution curve. In the network's forward inference stage, the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are received within the time-varying hydrodynamic prediction network. The network's long short-term memory convolutional layers are used to perform nonlinear forward inference fitting on these matrixes, outputting the microscopic lubrication degradation evolution curve for future consecutive time nodes. After generating the evolution curve, the remaining time span corresponding to when the microscopic lubrication degradation evolution curve breaks through a preset critical friction failure threshold is extracted. A detection algorithm is then executed to scan the microscopic lubrication degradation evolution curve downwards along the time axis, locating the absolute time coordinate corresponding to when the curve breaks through the preset critical friction failure threshold. The remaining time span is obtained by subtracting the absolute time coordinate from the current operating time. After completing the time span calculation, the remaining time span is converted into an oil change cycle control command and output.

[0025] Emitting a dual-band polarized light beam towards the micro-region of the tooth surface triggers a series of photophysical reactions. When the dual-band polarized beam penetrates the lubricating medium on the surface of the micro-region and generates an optical echo through reflection and backscattering, the receiving end simultaneously captures the optical echo and converts it into a continuous analog electrical signal. Since the analog electrical signal contains full-dimensional state information of the micro-interface, the execution terminal then performs analog-to-digital conversion on the analog electrical signal, thereby generating a polarized light sequence image data stream for subsequent analysis.

[0026] To deconstruct the hidden physical state within the generated polarized light sequence image data stream, the execution terminal performs two-dimensional phase unwrapping and background grayscale suppression processing on the polarized light sequence image data stream. After signal filtering, ambient light noise data can be removed, thereby separating the pure grating orthogonal polarization reflection information. After obtaining the pure reflection signal, the execution terminal extracts the phase delay amplitude variation characteristics of the corresponding bands of the dual-band polarized beam in the grating orthogonal polarization reflection information. Since the phase delay amplitude variation characteristics directly reflect the path difference when the light wave penetrates the medium, the phase delay amplitude variation characteristics are mapped to microscopic path difference values. Based on the inherent correlation between the microscopic path difference values ​​and the physical properties of the lubricating material, the microscopic path difference values ​​are transformed into a microscopic oil film thickness topological distribution matrix according to the refractive index of the lubricating medium, following optical mapping logic.

[0027] In parallel with the oil film thickness decoupling process, a specific algorithmic space is used to perform spatial frequency filtering on the polarized light sequence image data stream. This extracts the backscattering feature vector, which carries the geometric distribution of suspended particles within the medium. Subsequently, the execution terminal identifies discrete pixel clusters that deviate from the baseline smoothed scattered light intensity based on the backscattering feature vector. Considering that discrete pixel clusters represent free wear debris or carbon deposits attached to the tooth surface, the area ratio of discrete pixel clusters in the total number of pixels in the overall field of view is calculated using a geometric area statistical algorithm, thus outputting the proportion of local occlusion area.

[0028] Constructing a dynamic mapping model requires placing the microscopic oil film thickness topology distribution matrix and the transient torque data stream into the same global timestamp coordinate system. The moment when the transient torque data stream experiences a peak abrupt change is located within this global timestamp coordinate system as the corresponding moment of the peak abrupt change. Since the moment corresponding to the peak abrupt change represents the gear experiencing the most severe mechanical impact, the topological deformation rate of the microscopic oil film thickness topology distribution matrix at the moment of the peak abrupt change is extracted. Then, based on the topological deformation rate reflecting the passive compression state of lubricating droplets under extreme stress, the transient elastohydrodynamic shear strain field of the lubricating medium in the bearing micro-region of the tooth surface is derived.

[0029] When the molecular-level type parameter characterizes the first base oil with preset physicochemical properties, the execution terminal retrieves the corresponding molecular chain conformational uncoiling optical birefringence critical curve from the optical rheological mapping library. Since the molecular chain conformational uncoiling optical birefringence critical curve reflects the theoretical boundary of the polymer's resistance to shear failure, the coordinates of the transient elastohydrodynamic shear strain field are projected onto the molecular chain conformational uncoiling optical birefringence critical curve according to the spatial mapping rules. The projection position shows the degree of deviation between the actual force and the theoretical limit. Based on the preset normalization rules, the spatial Euclidean distance of the projection point from the micro-rheological yield limit is calculated to quantify the fracture risk of the base oil molecular skeleton. Finally, the spatial Euclidean distance is designated as the shear degradation irreversible depolymerization dispersion, and the shear degradation irreversible depolymerization dispersion is output as the initial degradation baseline parameter.

[0030] For base oils in different formulation systems, when the molecular-level type parameter characterizes the second base oil with preset physicochemical properties, the evaluation logic will switch accordingly, retrieving the corresponding detergent-dispersant micelle agglomeration optical dispersion threshold from the optical rheological mapping library. Based on the physical law that failure of hydrogenated lubricating media often stems from sludge deposition caused by detergent depletion, a nonlinear cross-product operation is performed on the proportion of localized shielding area and the transient elastohydrodynamic shear strain field to obtain the nonlinear cross-product operation result. Since the nonlinear cross-product operation result comprehensively reflects the coupling effect of mechanical shearing and particle agglomeration, then based on preset normalization rules, the difference between the nonlinear cross-product operation result and the detergent-dispersant micelle agglomeration optical dispersion threshold is calculated. This difference, representing the accelerated trend of carbon deposit formation, is designated as the nonlinear oxidative carbon deposit accumulation gradient, and the nonlinear oxidative carbon deposit accumulation gradient is output as the initial degradation baseline parameter.

[0031] To introduce external interference from real-world operating conditions on the lubricating material, the peak value of the dominant frequency resonance energy is extracted from the high-frequency vibration data stream as the frequency domain feature of the vibration energy. Simultaneously, the impact load envelope fluctuation rate of the micro-region bearing the tooth surface is extracted as the time domain fluctuation feature of the load. Combining the dominant frequency resonance energy peak value with the physical mechanism of abnormal heat accumulation within the actuator, the peak value is converted into a thermodynamic catalytic degradation correction coefficient. Furthermore, based on the impact load envelope fluctuation rate reflecting the characteristics of metal fatigue spalling on the tooth surface and its mechanical cutting effect on the oil film, the impact load envelope fluctuation rate of the micro-region bearing the tooth surface is converted into a physical-mechanical fatigue spalling correction coefficient. Since thermodynamic and mechanical damage factors need to interact synergistically, the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient are combined to generate a dynamic compensation correction coefficient.

[0032] After generating the external operating condition compensation term, the execution terminal uses the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient to perform a high-dimensional tensor multiplication compensation operation on the initial degradation baseline parameter, obtaining the high-dimensional tensor multiplication compensation operation value. Since the high-dimensional tensor multiplication compensation operation value represents the true microscopic degradation state after both thermodynamic and mechanical corrections, the high-dimensional tensor multiplication compensation operation value is finally spliced ​​to output a composite dynamic decay state tensor.

[0033] At the input of the prediction network, the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are received within the time-varying hydrodynamic prediction network. As the underlying data injection is completed, the long short-term memory convolutional layer within the time-varying hydrodynamic prediction network performs nonlinear forward inference fitting on the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor, outputting the microscopic lubrication degradation evolution curve within future continuous time nodes. The microscopic lubrication degradation evolution curve depicts the complete trajectory of base oil performance degradation over time. Therefore, a detection algorithm is executed to scan the microscopic lubrication degradation evolution curve downwards along the time axis, locating the absolute time coordinate corresponding to the point where the microscopic lubrication degradation evolution curve breaks through the preset critical friction failure threshold. Utilizing the principle of calibrating the physical time of severe dry friction in the gearbox using the absolute time coordinate, the remaining time span is obtained by subtracting the absolute time coordinate from the current operating time. Leveraging the data advantage of providing maintenance personnel with an intuitive time window for the remaining time span, the remaining time span is converted into an oil change cycle control command, which is ultimately output.

[0034] At the input of the prediction network, the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are received within the time-varying hydrodynamics prediction network. The low-level data injection triggers the nonlinear forward inference fitting mechanism of the long short-term memory (LSM) convolutional layer within the time-varying hydrodynamics prediction network. The front-end two-dimensional convolutional kernel within the LSM convolutional layer performs a sliding feature extraction operation along the spatial dimension of the microscopic oil film thickness topology distribution matrix. This sliding feature extraction operation transforms the originally isolated microscopic oil film thickness pixels into a high-dimensional spatial feature map representing the local oil film rupture trend. Along with the generation of the high-dimensional spatial feature map, a tensor splicing mechanism fuses the composite dynamic decay state tensor with the high-dimensional spatial feature map. This dimensional fusion process constructs a multidimensional temporal feature sequence that includes spatial morphological distortion features and physicochemical degradation features. This multidimensional temporal feature sequence is then directionally fed into the core gating unit within the LSM structure.

[0035] Within the data flow of the core gating unit, the forget gate network reads the current input feature value and the hidden state value of the previous time node from the multidimensional temporal feature sequence. Based on the read values, the forget gate network generates a historical information retention weight matrix through adaptive computation. The retention weight matrix performs element-wise multiplication filtering on the long-term memory cell states stored internally by the network, eliminating non-destructive interference noise data caused by short-term drastic fluctuations in operating conditions. Next, the input gate network decouples the actual mechanical wear increment and chemical oxidation accumulation increment based on the multidimensional temporal feature sequence. The actual mechanical wear increment and chemical oxidation accumulation increment are superimposed on the filtered long-term memory cell states, completing the update of the physical decay facts of the long-term memory cell states. The output gate network of the long short-term memory structure then reads the updated long-term memory cell states. The output gate network, combined with the mapping effect of the nonlinear activation function, outputs a predicted hidden state vector representing the future lubrication deterioration trend.

[0036] The predicted hidden state vector is mapped to continuous data coordinates extending with future physical time through regression operations in the terminal fully connected layer. These continuous data coordinates are smoothly connected along the time dimension to form the micro-lubrication degradation evolution curve within consecutive future time nodes. The micro-lubrication degradation evolution curve visually depicts the complete evolution trajectory of the base oil's performance gradually declining over time. The execution detection algorithm dynamically scans the micro-lubrication degradation evolution curve downwards along the time axis, locating the absolute time coordinate corresponding to the point where the curve breaks through the preset critical friction failure threshold. The absolute time coordinate defines the ultimate physical time boundary at which irreversible dry friction damage will occur in the gearbox. The processor subtracts the absolute time coordinate from the current execution terminal's running time to obtain the remaining time span representing the safe operating margin. This remaining time span is then converted into oil change cycle control commands adapted to the maintenance monitoring execution terminal through a data encoding conversion process via the communication interface and output externally.

[0037] In the long-term intelligent operation and maintenance application scenario of heavy-duty gearboxes in wind power and industrial transmission fields, the first base oil with preset physicochemical properties in this embodiment of the invention is preferably a coal-based polyalphaolefin (PAO) lubricating medium; the second base oil with preset physicochemical properties is preferably a hydrogenated lubricating medium. The first base oil and the second base oil can be collectively referred to as domestically produced base oils (also known as lubricating media with consistent physicochemical properties across multiple categories). For the above base oils, acquiring a polarized light sequence image data stream of the gearbox's internal tooth surface bearing micro-region under operating conditions requires a rigorous photoelectric signal conversion process. During the detection phase, emitting a dual-band polarized beam towards the tooth surface bearing micro-region triggers a series of photophysical reactions. The dual-band polarized beam penetrates the lubricating medium on the surface of the tooth surface bearing micro-region and generates optical echoes through reflection and backscattering by the tooth surface bearing micro-region. The receiving end simultaneously captures the optical echoes and then converts them into continuous analog electrical signals. The analog electrical signals contain full-dimensional state information of the micro-interface. The execution terminal then performs analog-to-digital conversion processing on the analog electrical signals, thereby generating a polarized light sequence image data stream for subsequent analysis. While acquiring image data streams, the sensor network simultaneously collects transient torque data streams and high-frequency vibration data streams from the gearbox's main shaft, continuously capturing the physical signals of main shaft rotational deformation and gearbox high-frequency oscillation. The signal processing module amplifies and filters these signals, converting them into digitized transient torque and high-frequency vibration data streams. After acquiring the underlying data, the background execution terminal retrieves preset molecular-level type parameters and reads the data stored in the execution terminal to parse out the corresponding molecular-level type parameters of the domestically produced base oil currently in use.

[0038] To deconstruct the hidden physical state within the generated polarized light sequence image data stream, the execution terminal performs two-dimensional phase unwrapping and background grayscale suppression processing on the polarized light sequence image data stream. Signal filtering removes ambient light noise, thus separating the pure orthogonal polarized reflection information of the grating. After obtaining the pure reflection signal, the analysis model extracts the phase delay amplitude variation characteristics of the corresponding bands of the dual-band polarized beams in the orthogonal polarized reflection information of the grating. Since the phase delay amplitude variation characteristics directly reflect the path difference when light waves penetrate the medium, the phase delay amplitude variation characteristics are mapped to microscopic path difference values. Based on the inherent correlation between the microscopic path difference values ​​and the physical properties of the lubricating material, the microscopic path difference values ​​are transformed into a microscopic oil film thickness topological distribution matrix according to the refractive index of the lubricating medium, following optical mapping logic. Parallel to the oil film thickness decoupling process, a specific algorithm space is used to perform spatial frequency filtering on the polarized light sequence image data stream to extract the backscattering feature vector carrying the geometric distribution state of suspended particles within the medium. The execution terminal identifies discrete pixel clusters that deviate from the reference smooth scattered light intensity based on the backscattering feature vector. Considering that discrete pixel clusters represent free wear debris or carbon deposits attached to the tooth surface, the area ratio of discrete pixel clusters in the total number of pixels in the overall field of view can be calculated using a geometric area statistical algorithm, and the local occlusion area ratio can be output.

[0039] Constructing a dynamic mapping model requires placing the microscopic oil film thickness topological distribution matrix and the transient torque data stream into the same global timestamp coordinate system. The moment when the transient torque data stream experiences a peak abrupt change is located within this global timestamp coordinate system as the corresponding moment of the peak abrupt change. Since the moment corresponding to the peak abrupt change represents the gear experiencing severe mechanical impact, the extraction module extracts the topological deformation rate of the microscopic oil film thickness topological distribution matrix at the moment corresponding to the peak abrupt change. Based on the topological deformation rate reflecting the passive compression state of lubricating droplets under extreme stress, the transient elastohydrodynamic shear strain field of the lubricating medium in the bearing micro-region of the gear surface is derived. When the molecular-level type parameter characterizes the first base oil with preset physicochemical properties, the underlying logic retrieves the corresponding molecular chain conformational uncoiling optical birefringence critical curve from the optical rheological mapping library. The molecular chain conformational uncoiling optical birefringence critical curve reflects the theoretical boundary of the polymer's resistance to shear failure. According to spatial mapping rules, the coordinate points of the transient elastohydrodynamic shear strain field are projected onto the molecular chain conformational uncoiling optical birefringence critical curve. The projection position demonstrates the deviation between the actual stress and the theoretical limit. Based on a preset normalization rule, the spatial Euclidean distance between the projection point and the micro-rheological yield limit is calculated to quantify the risk of fracture in the base oil molecular skeleton. The execution terminal specifies the spatial Euclidean distance as the shear-resistant degradation irreversible depolymerization dispersion and outputs it as the initial degradation baseline parameter. For base oils in different formulation systems, when the molecular-level type parameter characterizes the second base oil with preset physicochemical properties, the evaluation logic switches accordingly, retrieving the corresponding detergent-dispersant micelle agglomeration optical dispersion threshold from the optical rheological mapping library. Based on the physical law that failure of hydrogenated lubricating media often stems from detergent depletion leading to sludge deposition, the calculator performs a nonlinear cross-product operation on the proportion of local shielding area and the transient elastohydrodynamic shear strain field to obtain the nonlinear cross-product operation result. Since the nonlinear cross-product operation result comprehensively reflects the coupling effect of mechanical shearing and particle agglomeration, the difference between the nonlinear cross-product operation result and the detergent-dispersant micelle agglomeration optical dispersion threshold is then calculated based on a preset normalization rule. The difference representing the accelerating trend of carbon deposit formation is specified as the nonlinear oxidative carbon deposit accumulation gradient, and the nonlinear oxidative carbon deposit accumulation gradient is output as the initial degradation baseline parameter.

[0040] To incorporate external interference from real-world operating conditions on the lubricating material, the algorithm module extracts the peak resonant energy of the dominant frequency from the high-frequency vibration data stream as the frequency domain feature of the vibration energy, and simultaneously extracts the impact load envelope fluctuation rate of the micro-region bearing the tooth surface as the time domain fluctuation feature of the load. Combining the dominant frequency resonant energy peak with the physical mechanism representing the abnormal accumulation of internal thermal energy in the execution terminal, the peak resonant energy is converted into a thermodynamic catalytic degradation correction coefficient. Based on the characteristic that the impact load envelope fluctuation rate reflects the mechanical cutting effect of tooth surface metal fatigue spalling on the oil film, the impact load envelope fluctuation rate of the micro-region bearing the tooth surface is converted into a physical-mechanical fatigue spalling correction coefficient. Since thermodynamic and mechanical damage factors need to synergistically interact, the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient are combined to generate a dynamic compensation correction coefficient. After generating the external operating condition compensation term, the compensator uses the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient to perform a high-dimensional tensor multiplication compensation operation on the initial deterioration baseline parameter, obtaining the high-dimensional tensor multiplication compensation operation value. The high-dimensional tensor dot product compensation operation values ​​represent the true microscopic degradation state after both thermodynamic and mechanical corrections. Finally, the high-dimensional tensor dot product compensation operation values ​​are spliced ​​together to output a composite dynamic decay state tensor.

[0041] At the input of the prediction network, the time-varying hydrodynamics prediction network receives the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor. The low-level data injection triggers the nonlinear forward inference fitting mechanism of the long short-term memory (LSM) convolutional layer within the time-varying hydrodynamics prediction network. The front-end two-dimensional convolutional kernel within the LSM convolutional layer performs a sliding feature extraction operation along the spatial dimension of the microscopic oil film thickness topology distribution matrix. This sliding feature extraction transforms the originally isolated microscopic oil film thickness pixels into a high-dimensional spatial feature map representing the local oil film rupture trend. Along with the generation of the high-dimensional spatial feature map, a tensor splicing mechanism fuses the composite dynamic decay state tensor with the high-dimensional spatial feature map. This dimensional fusion process constructs a multidimensional temporal feature sequence that includes spatial morphological distortion features and physicochemical degradation features. This multidimensional temporal feature sequence is then directionally fed into the core gating unit within the LSM structure. Within the data flow of the core gating unit, the forget gate network reads the current input feature value and the hidden state value of the previous time node from the multidimensional temporal feature sequence. The forgetting gate network generates a historical information retention weight matrix based on the read values ​​through adaptive computation. This retention weight matrix performs element-wise multiplication filtering on the long-term memory cell states stored within the network, eliminating non-destructive noise data caused by drastic fluctuations in short-term operating conditions. Next, the input gate network decouples the actual mechanical wear increment and chemical oxidation accumulation increment based on multi-dimensional temporal feature sequences. These actual mechanical wear increments and chemical oxidation accumulation increments are superimposed onto the filtered long-term memory cell states, completing the update of the physical decay facts of the long-term memory cell states. The output gate network of the long short-term memory structure then reads the updated long-term memory cell states. The output gate network, combined with the mapping effect of a nonlinear activation function, outputs a predicted hidden state vector representing the future lubrication degradation trend. This predicted hidden state vector is mapped to continuous data coordinate points extending with future physical time through regression operations in the terminal fully connected layer. These continuous data coordinate points are smoothly connected in the time dimension to form a micro-lubrication degradation evolution curve within consecutive future time nodes. This micro-lubrication degradation evolution curve visually depicts the complete evolution trajectory of the base oil's performance gradually declining over time. The detection algorithm dynamically scans the micro-lubrication degradation evolution curve downwards along the time axis, and locates the absolute time coordinate corresponding to the point when the micro-lubrication degradation evolution curve breaks through the preset critical friction failure threshold during the scanning process. The absolute time coordinate marks the ultimate physical time boundary at which the gearbox is about to experience irreversible dry friction damage. The processor calculates the difference between the absolute time coordinate and the current execution terminal's running time to obtain the remaining time span representing the safe operating margin. The remaining time span is then converted into oil change cycle control commands adapted to the operation and maintenance monitoring execution terminal through a data encoding and conversion process via the communication interface and output externally.

[0042] A polarized light sequence image data stream is essentially a collection of optical mapping information generated by modulating a dual-band polarized light beam of a specific wavelength through a lubricating medium. In the low-light environment inside a wind turbine gearbox, conventional visible light is easily lost due to refraction from oil and gas caused by mechanical agitation. The dual-band polarized light beam includes a near-infrared polarized band with strong penetrating power and a short-wave ultraviolet polarized band highly sensitive to surface morphology. When the dual-band polarized light beam illuminates the high-speed rotating gear surface bearing micro-region, it undergoes depolarization and backscattering effects upon encountering the metal substrate and attached free particles. The image sensor continuously captures reflected light waves carrying phase modulation information and converts the optical phase distribution into a continuous two-dimensional pixel array using an analog-to-digital conversion mechanism. This two-dimensional pixel array is arranged along the time axis at an extremely high frame rate, forming a polarized light sequence image data stream that records the microscopic flow and breakage state of the lubricating oil film.

[0043] Transient torque data streams and high-frequency vibration data streams constitute the core mechanical boundary conditions characterizing the underlying physical stress state of the wind turbine drivetrain. When subjected to gusts of wind or grid disturbances, the main shaft of a heavy-duty gearbox experiences highly destructive alternating torsional stress. A strain gauge bridge deployed on the main shaft surface converts mechanical deformation into voltage fluctuation signals. After high-frequency sampling, a transient torque data stream is extracted, which is used to invert the ultimate compressive load borne by the micro-meshing area of ​​the gear teeth at the microsecond level. The high-frequency vibration data stream, acquired in parallel with the torque signal, originates from a piezoelectric accelerometer attached to the outside of the bearing housing of the gearbox. The piezoelectric accelerometer captures the broadband excitation energy generated by the collision of microscopic metal protrusions on the gear meshing surface due to oil film rupture. These two sets of physical signal streams are integrated into the control terminal in real time via an industrial bus, providing a realistic operating load reference for subsequent multidimensional tensor compensation.

[0044] Molecular-level type parameters serve as the underlying chemical identifier, bridging the gap between microscopic mechanical response and macroscopic physicochemical degradation. Domestically produced base oils exhibit significant differences in their internal molecular skeletons due to variations in refining processes. Specifically, molecular-level type parameters encompass the degree of polymerization of isomerized branches within coal-based polyalphaolefins, the molar ratio of saturated cycloalkanes in hydrotreated base oils, and the concentration ratios of detergents, dispersants, and anti-wear agents added to the blending formulation. During the terminal initialization phase, maintenance personnel input the corresponding chemical composition data into the terminal's memory. Based on the input chemical composition data, the terminal automatically matches the corresponding molecular chain conformational uncoiling threshold and micelle aggregation dispersion limit from a pre-set optical rheological mapping library. The evaluation model utilizes differentiated physical degradation assessment benchmarks for base oils from different chemical synthesis routes to perform evaluation calculations.

[0045] The microscopic oil film thickness topology distribution matrix is ​​a digital reconstruction of the three-dimensional spatial distribution of the lubricating medium under dynamic compression on the tooth surface. When light waves penetrate an oil film medium with a specific refractive index, the change in optical path difference has a strict physical mapping relationship with the absolute thickness of the oil film. The execution terminal reduces ambient background noise through two-dimensional phase unwrapping processing and extracts the phase delay amplitude variation characteristics inherent in the orthogonal polarization reflection information of the grating. The phase delay amplitude variation characteristics are converted into absolute thickness values ​​according to the local refractive index of the medium. Thousands of discrete thickness values ​​distributed within the detection field of view are arranged together according to a spatial coordinate array, forming a microscopic oil film thickness topology distribution matrix that dynamically changes with the rotation of the principal axis. The microscopic oil film thickness topology distribution matrix clearly shows the geometric morphological characteristics of oil film thinning and even rupture in the meshing area, transition area, and even edge area of ​​the tooth surface.

[0046] The percentage of locally obscured area is a direct optical geometric indicator that quantifies the decline in the internal cleaning and dispersing effectiveness of the lubricating medium and the degree of mechanical wear. In long-term, heavy-load operating environments, base oil is highly susceptible to oxidation at high temperatures, forming insoluble carbon deposits and sludge, or to the shedding of micron-sized metal debris due to boundary lubrication failure. These micron-sized metal debris and carbon deposits suspended in the lubricating medium significantly alter the propagation path of light waves. Spatial frequency filtering removes the smoothed background from the polarized light sequence image data stream, extracting the backscattering feature vector representing high-frequency abrupt changes. The execution terminal identifies discrete pixel clusters in the field of view that deviate from the normal scattered light intensity based on the backscattering feature vector. By calculating the percentage of the total pixel area accumulated in these discrete pixel clusters relative to the total effective detection field of view, the percentage of locally obscured area directly reflects the severity of oil and debris adhesion on the tooth surface.

[0047] The transient elastohydrodynamic shear strain field reveals the micro-rheological tearing state of the macromolecular skeleton within the lubricating medium under extremely high contact stress. At the ultimate compression instant of tooth meshing, the lubricating medium is in a complex elastohydrodynamic pressure state between fluid and solid-like states. The execution terminal strictly aligns the micro-oil film thickness topological distribution matrix with the transient torque data stream in the time dimension, extracting the higher-order spatial partial derivatives of the oil film topological distribution corresponding to the instant of peak torque mutation, i.e., the topological deformation rate. The topological deformation rate, combined with the fluid dynamics constitutive equations, derives a three-dimensional tensor space. Each coordinate node within the three-dimensional tensor space is associated with a micro-shear stress vector in a specific direction, thus forming the transient elastohydrodynamic shear strain field.

[0048] The initial degradation baseline parameter is a raw quantitative coordinate representing the degree of physicochemical degradation of the lubricating medium before external mechanical interference. The execution terminal employs parallel degradation calculation logic for different formulation systems. For coal-based polyalphaolefin lubricating media with long-chain polymer structures, the execution terminal calculates the spatial Euclidean distance between the projection point of the transient elastohydrodynamic shear strain field and the pre-set molecular chain conformation decoupling optical birefringence critical curve, deriving the shear-resistant degradation irreversible depolymerization dispersion, representing the degree of irreversible fracture of the molecular skeleton. For hydrogenated lubricating media prone to sludge formation, the execution terminal performs a nonlinear cross-product operation between the local shielding area ratio and the transient elastohydrodynamic shear strain field, calculating the difference exceeding the optical dispersion threshold of detergent dispersant micelle aggregation, i.e., the nonlinear oxidative carbon deposition accumulation gradient. The shear-resistant degradation irreversible depolymerization dispersion or the nonlinear oxidative carbon deposition accumulation gradient together constitutes the initial degradation baseline parameter for measuring the failure initiation state of the base oil.

[0049] The dynamic compensation correction factor is a dynamic adjustment factor that incorporates the accelerated deterioration effect of harsh alternating operating conditions on base oil into the evaluation system. Gusts and grid shedding in wind farm environments often lead to transient high-temperature surges and high-frequency impacts inside the gearbox. The execution terminal performs frequency domain conversion and time domain envelope demodulation on the high-frequency vibration data stream. The separated dominant frequency resonance energy peak is mapped as a thermodynamic catalytic degradation correction factor characterizing abnormal heat accumulation. The synchronously extracted impact load envelope fluctuation rate is mapped as a physical-mechanical fatigue spalling correction factor characterizing the mechanical cutting effect of tooth surface metal fatigue spalling on the oil film. The thermodynamic catalytic degradation correction factor and the physical-mechanical fatigue spalling correction factor are combined according to a weighted superposition mechanism to form the dynamic compensation correction factor used to adjust the calculation deviation of the initial deterioration state.

[0050] The composite dynamic decay state tensor is a multidimensional data carrier that highly concentrates the dual effects of the degradation of the lubricating medium's physicochemical properties and the catalytic damage caused by harsh external operating conditions. The initial degradation baseline parameter reflects the static degradation trend of the lubricating oil film at the theoretical level. The execution terminal retrieves the thermodynamic catalytic degradation correction coefficient and the physical mechanical fatigue spalling correction coefficient, and performs multiplicative compensation operations on the various degradation indices within the initial degradation baseline parameter according to the high-dimensional tensor multiplication rule. The multiplication compensation operation adds the additional mechanical shear force caused by the high-frequency vibration of the spindle and the cumulative effect of local high temperature to the original degradation data. The numerical lattice after feature splicing and fusion processing is combined into a dense composite dynamic decay state tensor.

[0051] The time-varying hydrodynamic prediction network is an artificial intelligence computing model that predicts the gradual collapse and failure trajectory of the macroscopic lubrication performance of base oil over physical time. The prediction network integrates a two-dimensional convolutional kernel, adept at capturing abrupt changes in spatial topology, and a long short-term memory (LSTM) gating unit, focused on handling long-range temporal dependencies. The front-end two-dimensional convolutional kernel slides along the spatial dimension of the microscopic oil film thickness topological distribution matrix to extract local oil film rupture trend features. These local oil film rupture trend features are fused with a composite dynamic decay state tensor to construct a multidimensional temporal feature sequence. This multidimensional temporal feature sequence enters the forget gate and input gate of the LTM structure for filtering and accumulation operations, completing the real-time update of the physical decay of the long-term memory cell state. The output gate network, combined with the mapping effect of a nonlinear activation function, outputs a hidden state vector representing the future evolution trend. It should be noted that, in addition to the preferred use of the LTM-CNN architecture described above, alternative embodiments of the time-varying hydrodynamic prediction network can also employ a Transformer temporal prediction network based on a self-attention mechanism or a gated recurrent unit (GRU) network. By capturing the nonlinear correlation of the composite dynamic decay state tensor over a long period through a multi-head attention mechanism, or by using the simplified gating structure of GRU to perform inference on computationally limited edge devices, it is also possible to achieve accurate fitting and output of the micro-lubrication degradation evolution curve. The above-mentioned equivalent prediction network architectures are all within the scope of the technical solutions fully disclosed and supported by this invention.

[0052] After acquiring the polarized light sequence image data stream of the micro-region of the gearbox's internal tooth surface under operating conditions, the execution terminal decouples the orthogonal polarization reflection information of the grating through two-dimensional phase unwrapping processing. The extraction of the micro-oil film thickness is based on the principle of polarization phase modulation. The relationship between the phase delay amplitude variation characteristics and the optical path difference follows the interference phase equation, specifically expressed as:

[0053] in, Indicating the orthogonal polarization reflection information of the grating in coordinates Characteristics of the phase delay amplitude variation at the location; Indicates the center wavelength of a dual-band polarized beam; Indicates the refractive index of the lubricating medium; This represents the absolute value of the microscopic oil film thickness to be solved; This represents the incident angle of the polarized beam entering the lubricating medium. By performing the inverse operation on this equation, the execution terminal converts the phase delay amplitude variation characteristics into a microscopic oil film thickness topological distribution matrix, which is denoted as matrix [matrix name missing]. Its elements correspond to each sampling point In the parallel processing path, the execution terminal identifies discrete pixel clusters based on the backscattering feature vector. The calculation of the local occlusion area ratio is achieved by integrating the anomalous scattering pixels within the field of view; the calculation equation is as follows:

[0054] in, Indicates the percentage of the area partially obscured; and These represent the total number of pixels in the horizontal and vertical directions of the polarized light sequence image data stream, respectively. Representing coordinates Pixel value at; For the judgment function, when the coordinates When a pixel at a given location is identified as a member of a discrete pixel cluster by the backscattering feature vector, Otherwise Through this path, the execution terminal transforms microscopic geometric information into quantified data on the degree of sludge and wear debris contamination. After aligning the microscopic oil film thickness topological distribution matrix with the transient torque data stream, the execution terminal needs to generate a transient elastohydrodynamic shear strain field. The topological deformation rate reflects the thickness gradient change of the oil film under normal stress, while the transient elastohydrodynamic shear strain field... The intensity distribution follows the rheological momentum transfer equation:

[0055] in, Represents the transient elastohydrodynamic shear strain field in coordinates Strength at the location; This represents the dynamic viscosity of the lubricating medium as defined in the molecular-level type parameters; This represents the tangential velocity component calculated from the spindle operating speed; This represents the thickness value in the microscopic oil film thickness topology distribution matrix; The gradient value represents the topological distribution matrix of the microscopic oil film thickness, i.e., the topological deformation rate. The rheological momentum transfer equation couples macroscopic mechanical stress with microscopic oil film deformation, generating an initial data stream reflecting the rheological properties of the lubricating medium.

[0056] For coal-based polyalphaolefin lubricating media, the shear degradation and irreversible depolymerization dispersion The calculation is based on the Euclidean distance mapping of the multidimensional feature space. Considering the essential difference in physical dimensions between the intensity and phase of the transient elastohydrodynamic shear strain field, the execution terminal eliminates the dimensional influence before calculating the spatial distance by performing maximum and minimum value normalization, uniformly mapping the physical quantities to the dimensionless interval [0,1]. The specific maximum and minimum value normalization calculation formula is as follows: ,in This is the original collected data. and These represent the theoretical minimum and maximum boundary values ​​of the physical quantity in the optical rheological mapping library, respectively. After normalization, the dimensionless spatial Euclidean distance between the actual physical state point and the critical curve for optical birefringence due to molecular chain conformational uncoiling is calculated using the following formula:

[0057] in, Indicates the degree of irreversible depolymerization under shear degradation resistance; and These represent the actual transient elastohydrodynamic shear strain field intensity and the actual phase after normalization, respectively; and The theoretical threshold coordinates, which are also normalized, are represented in the physicochemical characteristic optical rheological mapping library. and These are the characteristic importance weighting coefficients calibrated based on the physicochemical testing experience of the base oil. By introducing normalization and weight allocation, the rationality of the heterogeneous physical quantities in the mathematical Euclidean space calculation is ensured, and the results are output as the initial degradation baseline parameters. In another parallel evaluation branch, when the molecular-level type parameter is characterized as a second base oil with preset physicochemical properties (such as for hydrotreated lubricating media), the nonlinear oxidative carbon deposit accumulation gradient is calculated. Its logical equation is:

[0058] in, This represents the nonlinear gradient of carbon deposition accumulation due to oxidation. Indicates the percentage of the area partially obscured; Indicates the intensity of the transient elastohydrodynamic shear strain field; This represents the optical dispersion threshold of micelle aggregation of the resolving dispersant in the optical rheological mapping library of physicochemical characteristics. During the operating condition compensation phase, the execution terminal outputs the composite dynamic decay state tensor through high-dimensional tensor dot product operations. This processing path nonlinearly couples the initial degradation baseline parameters with the dynamic compensation correction coefficients, and the operational equation is expressed as:

[0059] in, Represents the composite dynamic decay state tensor; This represents the dynamic compensation correction coefficient vector composed of the thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient; This represents the initial degradation baseline parameter tensor, which is composed of the shear-resistant degradation irreversible depolymerization discreteness or the nonlinear oxidative carbon deposition accumulation gradient. This represents the tensor dot product operator. This operation enables data dimensionality enhancement from the degradation of the oil product itself to the evolution of complex actual working conditions. Finally, the execution terminal uses time-varying hydrodynamics to predict the microscopic lubrication degradation evolution curves of the network. The hidden state update inside the long short-term memory convolutional layer follows the following gated loop equation:

[0060] in, It represents the current state of long-term memory cells, used to convey the fact of physical decay; This represents the weight matrix retained by the forget gate output; It indicates the memory state at the previous moment; This represents the incremental update weights of the input gate's output; This is the weight matrix; This is the hidden state from the previous moment; This is the fusion feature of the currently input composite dynamic decay state tensor and the microscopic oil film thickness topological distribution matrix; This is the bias term. Through this temporal evolution logic, the execution terminal calculates the remaining time span, completing the full data loop from microscopic physical feature extraction to macroscopic operation and maintenance command generation. Before generating the transient elastohydrodynamic shear strain field, considering the order-of-magnitude heterogeneous differences between the frame rate sampling frequency of the polarized light sequence image data stream and the sampling frequencies of the transient torque data stream and the high-frequency vibration data stream, this embodiment uses a timestamp-based polynomial spline interpolation algorithm to fuse and align the data. Specifically, using the low-frequency timestamp of the polarized light sequence image as the reference coordinate, the high-frequency transient torque data stream undergoes mean pooling downsampling processing within a local time window to ensure that the data dimension and time resolution of both are absolutely unified under the same global timestamp coordinate system, thereby providing a computable data foundation for subsequent physical field coupling.

[0061] Before the time-varying fluid dynamics prediction network performs nonlinear forward inference fitting, an offline model training process needs to be performed beforehand. When constructing the training set, multi-dimensional time-series feature sequences are extracted from historical operation and maintenance records as input samples, and the corresponding offline physicochemical analysis true degradation labels are used as target outputs. To enable the model to learn the decay law that conforms to the physical characteristics of the gearbox, this invention uses a customized mean squared error (PINN-MSE) based on physical information constraints as the loss function. The formula for backpropagation training is as follows:

[0062] in, This represents the oil film thickness degradation value predicted by the model. Labels for actual degradation values. This is a physical monotonicity penalty term used to strictly constrain the objective physical law that the degradation trend is irreversible (i.e., the physical state will only deteriorate or remain stable over time, and will not automatically improve). The penalty weight coefficient is used. During the training phase, the Adam optimizer is employed, with an initial learning rate of 0.001, a batch size of 64, and a maximum number of epochs of 500, until the customized loss function converges, thereby determining the optimal parameter boundaries of the long short-term memory convolutional layers in the network. The convergence criterion for the customized loss function is set as a loss decrease rate of less than [value missing] over 10 consecutive epochs. The aforementioned preset critical friction failure threshold is an objective value pre-set based on the minimum safe limit of the hydrodynamic boundary film thickness before microscopic corrosion occurs, from the historical sample database of lubrication failures in the wind power industry.

[0063] Embodiment 1 of this invention focuses on the long-term operation and maintenance scenario of wind turbine gearboxes using coal-based polyalphaolefin (PAO) base oil. For gearboxes in dynamic operation, a sequence of image data streams is acquired using a dual-band polarized endoscope deployed in the micro-regions of the gear teeth. During actual wind farm operation, due to the extremely high instantaneous pressure at gear meshing, the lubricating medium is prone to shear degradation. The execution terminal simultaneously acquires transient torque data streams and high-frequency vibration data streams during spindle operation, and retrieves preset molecular-level type parameters of the coal-based PAO base oil. Based on the principle of polarization phase modulation, the microscopic oil film thickness topological distribution matrix is ​​decoupled from the polarized light sequence image data stream, and the proportion of locally obscured area reflecting the state of oil residue is calculated. The microscopic oil film thickness topological distribution matrix and the transient torque data stream are placed in the same global timestamp coordinate system for time-domain alignment. The topological deformation rate is extracted at the moment when the transient torque exhibits a sudden peak change, thereby deriving the transient elastohydrodynamic shear strain field. By calling the physicochemical characteristic optical rheological mapping library, the spatial Euclidean distance between the coordinate points of the transient elastohydrodynamic shear strain field and the critical curve of optical birefringence of molecular chain conformation decoupling is calculated, and the irreversible depolymerization dispersion of shear degradation, which represents the risk of molecular skeleton breakage, is obtained.

[0064] A thermodynamic catalytic degradation correction coefficient is generated by combining the dominant frequency resonance energy peak in the high-frequency vibration data stream. A high-dimensional tensor multiplication compensation calculation is then performed on the initial degradation baseline parameter, outputting a composite dynamic decay state tensor. Finally, a time-varying hydrodynamic prediction network is used to fit the lubrication degradation evolution curve, calculating the remaining time span for the downward breakdown critical friction failure threshold. When a significant trend of molecular chain breakage in the coal-based polyalphaolefin base oil is detected, the execution terminal automatically converts the remaining time span into an adaptive dynamic oil change cycle control command, switching the wind turbine operation and maintenance scheme from a fixed experience-based oil change mode to a precise response mode based on the measured oil film condition.

[0065] Embodiment 2 of this invention addresses the evaluation scenario of heavy-duty gearboxes using hydrotreated base oils in industrial transmission applications. Hydrotreated base oils, compared to coal-based PAOs, are more prone to carbon deposits and sludge buildup during long-term operation, leading to a decline in detergency and dispersancy. In the implementation process, the execution terminal first acquires the molecular-level type parameters corresponding to the hydrotreated base oil and performs spatial frequency filtering on the polarized light sequence image data stream to identify discrete pixel clusters representing carbon deposit particles, thus determining the proportion of local occlusion area. In the data processing path, the execution terminal performs a nonlinear cross-product operation on the proportion of local occlusion area and the transient elastohydrodynamic shear strain field, calculating the difference between the calculated value and the preset optical dispersion threshold for detergent-dispersant micelle aggregation in the physicochemical characteristic optical rheological mapping library, defining this as the nonlinear oxidative carbon deposit accumulation gradient. For this gradient value, the execution terminal introduces a physical-mechanical fatigue spalling correction coefficient generated from the impact load envelope volatility to complete tensor mapping compensation for the initial degradation baseline parameters. Because hydrotreated oils are more sensitive to particulate contamination, the long short-term memory convolutional layer of the prediction network focuses on feature extraction and hidden state updates targeting local oil film rupture trends. By locating the absolute time coordinates of the lubrication degradation evolution curve at the failure interface, the remaining safe operating time span is calculated at the terminal. Compared to traditional offline oil sample analysis, this method can directly observe the dynamic process of oil film thinning caused by the degradation of detergency in hydrotreated base oils.

[0066] Embodiment 3 of this invention: Dynamic adaptation scenario of domestically produced base oil under extreme high and low temperature alternating conditions. In extreme cold or high temperature environments, the viscosity index of base oil fluctuates drastically, directly affecting the absolute thickness distribution of the oil film. While acquiring polarized light sequence image data streams, the execution terminal deconstructs the microscopic thermodynamic excitation index characterized by the peak energy of the dominant frequency resonance through high-frequency vibration data streams. During data flow, the generated microscopic oil film thickness topology distribution matrix will undergo topological distortion with changes in real-time operating conditions. The execution terminal uses a thermodynamic catalytic degradation correction coefficient to perform real-time compensation against the irreversible depolymerization dispersion of shear degradation. The gating unit inside the prediction network will automatically remove non-destructive interference caused by transient noise through a forgetting gate network based on the abrupt changes in the composite dynamic decay state tensor, thereby preserving the memory cell state reflecting long-term degradation facts. Through the fitted high-precision lubrication degradation evolution curve, the operation and maintenance execution terminal can accurately locate the critical point of base oil performance degradation in different temperature ranges. This approach enables the output of control commands based on the actual molecular-level degradation state of both coal-based PAO and hydrotreated base oils under complex environmental stress. This evaluation method, based on the coupling of physical fields and chemical molecular states, completely solves the technical problem of the disconnect between experimental evaluation and practical application scenarios.

[0067] To verify the beneficial effects and feasibility of the evaluation method described in this invention, the following comparative experiment was conducted: Ten heavy-duty gearboxes of the same model (2.5MW) equipped with different domestically produced base oils were selected from a wind farm as monitoring objects. The experimental group adopted the method based on endoscopy and physical constraint prediction network described in this invention, while the control group adopted the traditional fixed-period sampling physicochemical analysis combined with the standard LSTM network method. After 12 months of follow-up verification, the test results are as follows: Comparison of early warning lead time and accuracy: The experimental group provided an average of 14.5 days earlier warning of microscopic oil film rupture than the control group. In terms of the accuracy of predicting deterioration trends (based on the MAPE index), the control group had an accuracy of only 82.4% due to the inability to avoid transient operating noise interference, while the experimental group of this invention achieved an accuracy of 96.7% through physical constraint compensation.

[0068] End-side inference latency comparison: For the requirement of real-time computing without stopping the machine, the present invention controls the single inference latency of processing fused data to within 45ms, which is significantly better than the 120ms of the comparison group, and fully meets the requirements of in-situ real-time monitoring in industrial sites.

Claims

1. A method for evaluating the lubricating performance of a gearbox based on the use of domestically produced base oil for endoscopic detection, characterized by, include: Step 1: Acquire the polarized light sequence image data stream of the micro-region of the gear tooth surface inside the gearbox under operation, collect the transient torque data stream and high-frequency vibration data stream during the operation of the gearbox spindle, and retrieve the preset molecular-level type parameters; Step 2: Extract the orthogonal polarization reflection information of the grating from the polarized light sequence image data stream, and generate a microscopic oil film thickness topology distribution matrix based on the orthogonal polarization reflection information of the grating; extract the backlight scattering feature vector from the polarized light sequence image data stream, and calculate the local occlusion area ratio based on the backlight scattering feature vector. Step 3: Align the microscopic oil film thickness topology distribution matrix with the transient torque data stream in the time domain, extract the topological deformation rate of the microscopic oil film thickness topology distribution matrix at the moment corresponding to the peak change in the transient torque data stream, and generate the transient elastohydrodynamic shear strain field based on the topological deformation rate; compare the molecular-level type parameters, microscopic oil film thickness topology distribution matrix, local shading area ratio, and transient elastohydrodynamic shear strain field with the preset optical rheological mapping library, and output the initial degradation benchmark parameters based on the comparison results; Step 4: Extract the frequency domain features of vibration energy and the time domain fluctuation features of load from the high-frequency vibration data stream, and fuse the frequency domain features of vibration energy and the time domain fluctuation features of load to generate dynamic compensation correction coefficients; use the dynamic compensation correction coefficients to perform high-dimensional tensor compensation calculation on the initial deterioration reference parameters, and output the composite dynamic decay state tensor. Step 5: Input the microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor into a pre-established time-varying fluid dynamics prediction network with physical monotonicity constraints, and output the microscopic lubrication degradation evolution curve; extract the remaining time span corresponding to the point when the microscopic lubrication degradation evolution curve breaks through the preset critical friction failure threshold; and convert the remaining time span into an oil change cycle control command for external output.

2. The nationally produced base oil endoscope inspection gear box lubrication performance evaluation method according to claim 1, characterized by, Acquire a sequence of polarized light image data streams of the microscopic region bearing the tooth surface inside the gearbox during operation, including: A dual-band polarized light beam is emitted toward the micro-region bearing the tooth surface. The dual-band polarized light beam penetrates the lubricating medium on the surface of the micro-region bearing the tooth surface and generates an optical echo through reflection and backscattering by the micro-region bearing the tooth surface. It receives optical echoes and converts them into continuous analog electrical signals; Analog-to-digital conversion is performed on the analog electrical signal to generate a polarized light sequence image data stream.

3. The method for evaluating the lubrication performance of a gearbox using endoscopic testing of domestically produced base oils according to claim 2, characterized in that, Extract the orthogonal polarization reflection information of the grating from the polarized light sequence image data stream, and generate a microscopic oil film thickness topology distribution matrix based on the orthogonal polarization reflection information, including: Two-dimensional phase unwrapping and background grayscale suppression processing are performed on the polarized light sequence image data stream to remove ambient light noise data and separate the orthogonal polarization reflection information of the grating; Extract the amplitude variation characteristics of phase delay of the corresponding band in the orthogonal polarization reflection information of the grating; The phase delay amplitude variation characteristics are mapped to the microscopic sight path difference value; The microscopic path difference value is converted into a microscopic oil film thickness topological distribution matrix based on the refractive index of the lubricating medium.

4. The method for evaluating the lubrication performance of a gearbox using endoscopic examination with domestically produced base oils according to claim 3, characterized in that, Extract the backlight scattering feature vector from the polarized light sequence image data stream, and calculate the local occlusion area percentage based on the backlight scattering feature vector, including: Spatial frequency filtering is performed on the polarized light sequence image data stream to extract the backlight scattering feature vector; Based on the backlight scattering feature vector, discrete pixel clusters with scattered light intensity deviating from the reference smooth light intensity are identified; Calculate the area ratio of discrete pixel clusters in the total number of pixels in the overall field of view, and output the proportion of local occlusion area.

5. The method for evaluating the lubrication performance of a gearbox using endoscopic examination of domestically produced base oils according to claim 4, characterized in that, The microscopic oil film thickness topological distribution matrix is ​​time-domain aligned with the transient torque data stream. The topological deformation rate of the microscopic oil film thickness topological distribution matrix at the moment corresponding to the peak abrupt change in the transient torque data stream is extracted. Based on the topological deformation rate, a transient elastohydrodynamic shear strain field is generated, including: The microscopic oil film thickness topology distribution matrix and the transient torque data stream are placed in the same global timestamp coordinate system; The moment when the peak change occurs in the transient torque data stream is taken as the corresponding moment of the peak change. Extract the topological deformation rate of the microscopic oil film thickness topological distribution matrix at the time corresponding to the peak abrupt change; The transient elastohydrodynamic shear strain field of the lubricating medium in the bearing micro-region of the tooth surface is derived based on the topological deformation rate.

6. The method for evaluating the lubrication performance of a gearbox using endoscopic examination of domestically produced base oils according to claim 5, characterized in that, The molecular-level type parameters, microscopic oil film thickness topology distribution matrix, local shading area ratio, and transient elastohydrodynamic shear strain field are compared with a pre-set optical rheological mapping library. Based on the comparison results, initial degradation baseline parameters are output, including: When the molecular-level type parameter characterizes the first base oil with the preset physicochemical properties, the corresponding molecular chain conformation uncoiling optical birefringence critical curve in the optical rheological mapping library is retrieved. The coordinates of the transient elastohydrodynamic shear strain field are projected onto the critical curve of optical birefringence of molecular chain conformation decoupling. Based on the preset normalization rules, the spatial Euclidean distance between the projection point and the microrheological yield limit is calculated. The spatial Euclidean distance is specified as the shear-resistant degradation irreversible depolymerization dispersion, and the shear-resistant degradation irreversible depolymerization dispersion is output as the initial degradation baseline parameter.

7. The method for evaluating the lubrication performance of a gearbox using endoscopic testing of domestically produced base oils according to claim 6, characterized in that, Based on the comparison results, the initial degradation baseline parameters are output, including: When the molecular-level type parameter characterizes the second base oil with preset physicochemical properties, the corresponding detergent dispersant micelle aggregation optical dispersion threshold in the optical rheological mapping library is retrieved. The proportion of the local shielding area is multiplied by the transient elastohydro-shear strain field to obtain the result of the nonlinear cross-product operation. Based on the preset normalization rule, the difference between the result of the nonlinear cross-product operation and the optical dispersion threshold of the micelle agglomeration of the cleaning and dispersing agent is calculated. The difference is specified as a nonlinear oxidative carbon deposition accumulation gradient, and the nonlinear oxidative carbon deposition accumulation gradient is output as the initial degradation baseline parameter.

8. The method for evaluating the lubrication performance of a gearbox using endoscopic examination of domestically produced base oils according to claim 7, characterized in that, The vibration energy frequency domain features and load time domain fluctuation features are extracted from the high-frequency vibration data stream. These features are then fused to generate dynamic compensation correction coefficients, including: The peak value of the main frequency resonance energy is extracted from the high-frequency vibration data stream as the frequency domain feature of the vibration energy, and the impact load envelope fluctuation rate of the micro-region of the tooth surface bearing is extracted as the time domain fluctuation feature of the load. The peak energy of the dominant frequency resonance is converted into a thermodynamic catalytic degradation correction coefficient. The impact load envelope fluctuation rate of the micro-region bearing the tooth surface is converted into a physical-mechanical fatigue spalling correction coefficient. The thermodynamic catalytic degradation correction coefficient and the physical-mechanical fatigue spalling correction coefficient are combined to generate a dynamic compensation correction coefficient.

9. The method for evaluating the lubrication performance of a gearbox using endoscopic testing of domestically produced base oils according to claim 8, characterized in that, Using the dynamic compensation correction coefficients generated in step 4, high-dimensional tensor compensation calculations are performed on the initial degradation baseline parameters, outputting a composite dynamic decay state tensor, including: High-dimensional tensor multiplication compensation operation was performed on the initial deterioration baseline parameter using the thermodynamic catalytic degradation correction coefficient and the physical mechanical fatigue spalling correction coefficient to obtain the high-dimensional tensor multiplication compensation operation value. The numerical values ​​of the dot product compensation operation of the high-dimensional tensor are concatenated to output a composite dynamic decay state tensor.

10. The method for evaluating the lubrication performance of a gearbox using endoscopic examination of domestically produced base oils according to claim 9, characterized in that, The microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are input into a pre-established time-varying fluid dynamics prediction network with physical monotonicity constraints, and the microscopic lubrication degradation evolution curve is output. The remaining time span corresponding to the point when the microscopic lubrication degradation evolution curve breaks through the preset critical friction failure threshold is extracted. The remaining time span is converted into oil change cycle adjustment commands and output externally, including: The microscopic oil film thickness topology distribution matrix and the composite dynamic decay state tensor are received within the time-varying fluid dynamics prediction network. The long short-term memory convolutional layer of the time-varying hydrodynamic prediction network is used to perform nonlinear forward inference fitting on the micro oil film thickness topology distribution matrix and the composite dynamic decay state tensor, and outputs the micro lubrication degradation evolution curve in future continuous time nodes. The detection algorithm is executed to scan the micro-lubrication degradation evolution curve downward along the time axis, and to locate the absolute time coordinate corresponding to the point when the micro-lubrication degradation evolution curve breaks through the preset critical friction failure threshold. The remaining time span is obtained by subtracting the absolute time coordinate from the current running time. The remaining time span is converted into an oil change cycle control command, and the oil change cycle control command is output.

Images