Multi-source sensing lubricating oil real-time demulsibility prediction method and system

Abstract

The invention belongs to the field of industrial equipment lubrication state monitoring, and discloses a lubricating oil real-time demulsibility prediction method and system based on multi-source sensing, and the method comprises the steps: building and training a demulsibility prediction model comprising GBDT and SVM based on historical data, and in the training stage, taking oil product types and working conditions in the historical data as input features, and carrying out the prediction of the demulsibility of lubricating oil; taking a corresponding laboratory actually-measured lubricating oil demulsibility value as a training target, and enabling a model output value to be continuously close to the actually-measured value, so as to obtain a trained demulsibility prediction model; the oil product type and the working condition of the lubricating oil are collected in real time and input into the demulsibility prediction model, and a demulsibility prediction value tsep is output; the current state of the lubricating oil is judged according to the comparison result of the tsep and a preset threshold value; the system comprises a multi-source sensing module for acquiring oil product parameters of lubricating oil; the data processing and analysis module stores a trained demulsibility prediction model, takes oil product parameters as input and outputs tsep; and the cloud early warning module judges the current state of the lubricating oil according to a comparison result of the tsep and a preset threshold value. Through the multi-parameter collaborative modeling and dynamic compensation technology, the demulsibility index can be detected on line in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment status monitoring, and in particular to a multi-source sensing method and system for predicting the real-time demulsibility of lubricating oil. Background Art

[0002] In modern industrial production, lubricating oil, the "blood" of mechanical equipment such as generators, ship engines, and aircraft hydraulic systems, is a critical element for ensuring proper operation. However, due to the influence of the operating environment and maintenance operations, the oil often mixes with water or other impurities during operation, forming emulsions. This disrupts the oil film between contact surfaces, reduces the protective effect of the lubricant, and exacerbates wear and corrosion between parts. However, there are currently few means and methods for online monitoring of lubricant anti-emulsification properties.

[0003] Lubricant demulsibility refers to the ability of an oil-water mixture to separate automatically under static conditions and is an important indicator for evaluating lubricant performance. The traditional testing method is based on ASTM D1401 and GB / T 8022-2019 standards, and measures the oil-water separation time (t sep ), has the following inherent defects:

[0004] (1) Offline detection hysteresis: it is necessary to stop the machine to take samples for inspection, which cannot reflect the changes in the oil state during operation in real time; (2) Single parameter judgment error: it only considers the static moisture content, ignoring the synergistic degradation effects such as acid value increase and abrasive catalysis; (3) Lack of early warning: when t sep After >30 minutes, the oil has become irreversibly emulsified and the equipment faces the risk of sudden failure.

[0005] However, the current online monitoring technology for operating oil in related mechanical equipment focuses on monitoring the temperature, viscosity, and dielectric constant parameters of the lubricating oil: for example, Chinese patent publication number CN116953038A discloses a method for analyzing and monitoring the lubricating oil quality of a ship's main engine: based on offline detection and online detection, the oil quality sensitive parameters and corresponding dielectric constants of multiple main engine lubricating oils of the ship are collected to form a relationship curve between the oil quality sensitive parameters and the dielectric constant based on the cumulative sailing time of the ship's main engine. The alarm threshold of the dielectric constant can be determined according to the age of the ship. The dielectric constant of the main engine lubricating oil is detected when the ship is sailing. When it reaches the alarm threshold of the above-mentioned dielectric constant, an alarm is issued; and Chinese patent publication number CN117233040A discloses a lubricating oil quality online monitoring system and method. The lubricating oil quality online monitoring system includes a viscosity measuring device for measuring the viscosity of the lubricating oil and an image acquisition and processing device for obtaining abrasive particle information in the lubricating oil. The viscosity information and abrasive particle information of the lubricating oil to be monitored can be monitored online to comprehensively judge whether the lubricating oil needs to be replaced. At present, there is no report on the real-time anti-emulsification monitoring method of oil. Therefore, the development of a real-time anti-emulsification online monitoring system for lubricating oil has certain industrial value and significance. Summary of the Invention

[0006] The purpose of the present invention is to provide a multi-source sensing lubricant oil real-time demulsibility prediction method and system, the present invention can realize online real-time detection of demulsibility index through multi-parameter collaborative modeling and dynamic compensation technology.

[0007] The above objectives are achieved through the following technical solutions:

[0008] A first aspect of the present invention provides a multi-source sensing method for predicting the real-time demulsibility of lubricating oil, comprising the following steps:

[0009] (1) A demulsibility prediction model consisting of a gradient boosted decision tree (GBDT) and a support vector machine (SVM) was constructed and pre-trained using historical data: the oil type and operating conditions in the historical data were used as input features, and the corresponding measured lubricant demulsibility values ​​were used as training targets. The importance weights of the input features were extracted through GBDT, and nonlinear regression was performed on the input features based on the kernel function mapping of the support vector machine (SVM) combined with the importance weights to obtain the trained demulsibility prediction model.

[0010] (2) Real-time collection of lubricant oil type and operating conditions as input to the trained demulsibility prediction model, which then outputs the demulsibility prediction value;

[0011] (3) According to the comparison result of the demulsibility prediction value and the preset threshold value, the current state of the lubricant is judged as a failure state or a normal state.

[0012] Among them, in (1), pre-training is performed by continuously approximating the output value of the anti-emulsification prediction model to the measured anti-emulsification value of the lubricant.

[0013] In (1) and (2), the operating conditions include: dielectric constant, emulsified water content, acid value, temperature, viscosity and abrasive concentration, and the oil type is the result of abrasive material classification.

[0014] In (2), the oil type and operating conditions are filtered, normalized or feature fused to generate a standardized input vector that is input into the trained demulsibility prediction model.

[0015] In (2), the methods for filtering, normalizing or feature fusion processing of oil type and operating conditions include:

[0016] Filtering: Sliding window filtering is performed on the original signals of dielectric constant and viscosity, and abnormal points are replaced by the median value within the window; a second-order Butterworth low-pass filter is used on the wear particle concentration signal to suppress high-frequency noise;

[0017] Normalization adopts a hybrid normalization strategy: the Min-Max normalization method is adopted for the emulsified water content M and the acid value TAN; the Z-Score normalization is adopted for the viscosity change rate Δμ / Δt; and the logarithmic compression is adopted for the dielectric constant change Δε;

[0018] Feature fusion processing, time series analysis of viscosity change rate Δμ / Δt is performed to extract statistical features within the sliding time window, including mean, variance and slope; vector encoding is performed on the abrasive material classification results to generate material feature vector N=[n1,n2,...,n i ], where n i Represents the probability weight of the i-th type of material.

[0019] In (3), the method for determining whether the current lubricating oil is in a failed state or a normal state includes:

[0020] Level 1 warning: Demulsibility prediction value t sep >Preset threshold, judged as anti-emulsification failure, marked as failure state;

[0021] Normal state: If t sep ≤ the preset threshold, marked as normal status.

[0022] Furthermore, according to GB-T 7605-2024 Determination of the Demulsibility of Turbine Oil in Operation, a threshold value is set. If the predicted demulsibility value t sep >30 minutes, it is judged as anti-emulsification failure, normal state: if t sep ≤30 minutes, marked as normal, and the oil status is continuously monitored.

[0023] In the present invention, the method further comprises:

[0024] (4) After determining whether the current state of the lubricant is a failure state or a normal state, an anti-emulsification failure warning is triggered and a maintenance decision is generated:

[0025] The current lubricating oil is in a failed state, triggering an audible and visual alarm and generating a forced oil change instruction;

[0026] The current lubricating oil is in normal condition and the oil status is continuously monitored;

[0027] (5) The current status of the lubricant and maintenance decisions are pushed to the operation and maintenance terminal through the industrial Internet of Things platform and recorded in the cloud database.

[0028] A second aspect of the present invention provides a multi-source sensing lubricant real-time demulsibility prediction system using the above method, the system comprising:

[0029] The multi-source sensing module is used to monitor and collect lubricating oil parameters in real time, including dielectric constant, emulsified water content, acid value, temperature, viscosity, wear particle concentration, and oil type, and transmit the information to the data processing and analysis module;

[0030] The data processing and analysis module stores the trained demulsibility prediction model, uses the multi-source sensor module to monitor and collect lubricant parameters in real time as input, outputs the demulsibility prediction value, and transmits it to the cloud-based early warning module;

[0031] The cloud-based early warning module determines whether the current state of the lubricant is a failure state or a normal state based on the comparison result of the anti-emulsification prediction value with the preset threshold, triggers the anti-emulsification failure warning, and generates maintenance decisions.

[0032] Furthermore, the multi-source sensing module includes:

[0033] The multi-frequency dielectric constant sensor integrates a temperature compensation circuit to monitor the oil dielectric constant ε in real time. By synchronously detecting the difference between the dielectric constant and loss factor at high and low frequencies, the emulsified water content M is calculated in combination with a calibration formula and a temperature dynamic compensation model.

[0034] Acid value sensor, using potentiometric titration to determine the total acid value TAN;

[0035] Viscosity sensor, which measures dynamic viscosity and calculates the viscosity change rate Δμ / Δt;

[0036] The abrasive particle detection unit obtains the abrasive particle concentration C and abrasive particle material classification results in real time.

[0037] Furthermore, the multi-frequency dielectric constant sensor integrated temperature compensation circuit, acid value sensor and viscosity sensor in the multi-source sensing module are integrated into a sensor valve block, and the communication method adopts serial communication; the sensor valve block adopts electromagnetic shielding coating, and signal transmission adopts differential signal lines to reduce crosstalk.

[0038] Furthermore, in the data processing and analysis module, the lubricating oil demulsibility prediction stage includes:

[0039] Encode the preprocessed parameters into a fixed-dimensional input vector;

[0040] Use GBDT to weight the importance of input features and generate high-order interactive features;

[0041] Nonlinear regression is achieved through SVM kernel function mapping.

[0042] Furthermore, the cloud-based early warning module includes a real-time data dashboard that dynamically displays anti-emulsification index, wear risk level and abrasive distribution heat map.

[0043] Furthermore, in the cloud warning module: if the demulsibility prediction value t sep >30 minutes, the "anti-emulsification failure" alarm is triggered, which is set according to ASTM D1401 and GB / T 8022-2019 standards and meets the actual needs of industrial equipment lubricants; maintenance decision push, generating operation instructions according to the warning level, that is, executing steps (4) and (5) in the above method, which will not be repeated here.

[0044] The multi-source sensing lubricant real-time demulsibility prediction method and system provided by the present invention are applicable to the intelligent management of the lubrication status of key equipment such as power equipment, ship power systems, and heavy machinery.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention can collaboratively model water content, dielectric properties, viscosity, acid value, and other abrasive quality parameters. Using a machine learning model trained based on previous experimental data, this model performs real-time processing to output a real-time oil demulsibility index, expanding the parameter range of online oil monitoring systems. Specifically, through multi-parameter collaborative modeling and dynamic compensation technology, the present invention addresses the problem of traditional demulsibility testing relying on offline analysis and poor real-time performance, effectively reducing the risk of unplanned equipment downtime due to oil emulsification.

[0047] 2. It can accurately obtain the anti-emulsification and abrasive-related quality parameters of lubricating oil in real time, accurately analyze the oil quality, guide the addition of anti-emulsifiers and the filtration and replacement of oil, reduce the number of unit shutdowns and maintenance, improve the economy and safety of equipment operation, and save manpower and material resources; that is, the present invention can realize the online synchronous collection and dynamic compensation of multi-dimensional oil parameters, the prediction and graded warning of anti-emulsification degradation trend based on machine learning, and the active push of maintenance decisions, significantly improving the equipment lubrication reliability and reducing the equipment operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a multi-source sensing method for predicting the real-time demulsibility of lubricating oil provided by the present invention;

[0049] Figure 2 This is a structural diagram of a multi-source sensing lubricant real-time demulsibility prediction system provided by the present invention;

[0050] Among them, 1. Cloud warning module, 2. Wear particle detection unit, 3. Data analysis module, 4. Oil physical and chemical property monitoring unit, 5. Multi-source sensing module, 6. Gear pump, 7. Oil storage tank, 8. Ferromagnetic particles, 9. Non-ferromagnetic particles, 10. Shell, 11. Power supply unit.

[0051] Figure 3 This is a prediction relationship diagram provided by the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0054] Example 1

[0055] like Figure 1 As shown, the multi-source sensing method for predicting the real-time demulsibility of lubricating oil provided in this embodiment includes the following steps:

[0056] (1) A demulsibility prediction model consisting of a gradient boosting decision tree (GBDT) and a support vector machine (SVM) was constructed and pre-trained using historical data: the oil type and operating conditions in the historical data were used as input features, and the corresponding measured lubricant demulsibility values ​​were used as training targets. The importance weights of the input features were extracted through GBDT, and nonlinear regression was performed on the input features based on the kernel function mapping of the support vector machine (SVM) combined with the importance weights to obtain the trained demulsibility prediction model:

[0057] S1. Model training phase: A demulsibility prediction model is constructed using the gradient boosted decision tree (GBDT) and support vector machine (SVM) algorithms. The demulsibility prediction model is trained based on historical experimental data. The historical data covers multiple oil products (ISO VG32 / 46 / 68) and operating conditions, as well as the corresponding demulsibility measured values. The model training phase specifically includes:

[0058] S11, performing median filtering on sensor noise in historical experimental data;

[0059] S12, use Min-Max normalization to scale each parameter to the interval [0,1];

[0060] S13. Key parameters were screened by random forest feature importance evaluation, including moisture, acid value, dielectric constant change, viscosity change rate, and ferromagnetic abrasive particle concentration;

[0061] S14, use cross validation to divide the training set and test set;

[0062] S15. Use mean square error as the loss function and optimize the model hyperparameters through grid search.

[0063] In step S15, the model hyperparameter optimization content includes:

[0064] GBDT part: maximum depth of the tree, learning rate and sub-sample;

[0065] SVM part: kernel function, penalty factor and gamma;

[0066] Integration strategy: The feature importance weights output by GBDT are used as weighting coefficients of the SVM input features. Specifically, the input features are nonlinearly mapped through the feature weighting formula:

[0067] x′i=w i ·x i ;

[0068] Among them, w i is the importance weight of the i-th feature, which is calculated during the GBDT model training process; x i is the original value of the i-th feature; x′ iEnter values ​​for the weighted features.

[0069] (2) Real-time collection of lubricant oil type and operating conditions as input to the trained demulsibility prediction model, and output of the demulsibility prediction value; specifically, this is achieved through steps S2-S4:

[0070] S2, real-time data collection stage: The multi-source sensing module collects the operating parameters of the lubricating oil in real time, including dielectric constant, emulsified water content, acid value, temperature, viscosity, abrasive concentration and abrasive material classification results. The real-time data collection stage is specifically as follows:

[0071] The multi-frequency dielectric constant sensor integrates a temperature compensation circuit to monitor the oil dielectric constant (ε) in real time. Based on this, it fully considers factors such as temperature and flow rate to monitor the emulsified water content (M) in the oil.

[0072] Acid value sensor, which uses potentiometric titration to determine the total acid number (TAN);

[0073] Viscosity sensor, which measures dynamic viscosity and calculates the viscosity change rate (Δμ / Δt);

[0074] The abrasive particle detection unit obtains abrasive particle concentration (C) and material classification results in real time.

[0075] In this embodiment, the method for monitoring the dielectric constant and emulsified water content based on the multi-frequency dielectric constant sensor is as follows: the dielectric constant (ε) and the loss factor (Δtanδ) at high and low frequencies are detected simultaneously, and then the difference in the loss factor between the high and low frequencies (Δtanδ = tanδ) is calculated. 高频 -tanδ 低频 )Finally, the emulsified water content is calculated based on the calibration formula combined with the temperature dynamic compensation model.

[0076] In this embodiment, the calibration formula is: M = k·Δtanδ+b, where the coefficients k and b are calibrated by experiments; the temperature dynamic compensation model is corrected by introducing the real-time temperature (T): ε corrected =ε measured ·[1+α(TT ref )], where α is the temperature coefficient, T ref Take 25℃, ε corrected is the corrected dielectric constant, ε measured is the dielectric constant of the monitored oil.

[0077] S3, data preprocessing stage: filter, normalize and fuse the collected real-time data to generate a standardized input vector; the data preprocessing stage is specifically as follows:

[0078] S31, filtering processing includes:

[0079] Perform sliding window filtering on the original signals of dielectric constant and viscosity, with a window width of 5-7 sampling points. The abnormal point determination conditions are:

[0080] |x t -median(window t )|>3σ history ;

[0081] Among them, median is the median value of the data in the sliding window, window t is the window width, σ history is the standard deviation of the historical data of the parameter, and the abnormal points are replaced by the median value in the window;

[0082] A second-order Butterworth low-pass filter is used for the wear particle concentration signal, with the cutoff frequency set to 10 Hz and the transfer function as follows:

[0083]

[0084] Where s is the complex frequency domain variable in the Butterworth filter transfer function, which is used to calculate the attenuation characteristics of the filter for signals of different frequencies.

[0085] S32, normalization adopts hybrid normalization strategy:

[0086] The Min-Max normalization method is adopted for the emulsified water content M and acid value TAN;

[0087] For the viscosity change rate Δμ / Δt, Z-Score normalization was used;

[0088] Logarithmic compression is used for the dielectric constant change Δε.

[0089] S33, feature fusion processing includes:

[0090] Perform time series analysis on the viscosity change rate Δμ / Δt and extract statistical features within the sliding time window, including mean, variance and slope;

[0091] The abrasive material classification results are vector-encoded to generate the material feature vector N=[n1,n2,...,n i ], where n i Represents the probability weight of the i-th type of material.

[0092] S4, demulsibility prediction stage: The standardized input vector is input into the trained demulsibility prediction model, feature importance weights are extracted through gradient boosting decision tree (GBDT), and nonlinear regression is performed based on kernel function mapping of support vector machine (SVM), and the demulsibility prediction value of the current lubricant is output; in this embodiment, specifically:

[0093] Encode the preprocessed parameters into a 7-dimensional vector in a fixed order: V input =[M,Δε,TAN,T,Z,C,M encode ], where Z is the viscosity after Z-Score standardization, M encode is the principal component of the abrasive material feature vector after PCA dimensionality reduction;

[0094] The GBDT sub-model performs feature weighting to generate high-order interaction features;

[0095] The SVM sub-model receives weighted features and maps them to high-dimensional space through the RBF kernel function;

[0096] In this embodiment, the GBDT sub-model sets the maximum tree depth to 7 layers, the learning rate to 0.05, and the feature sampling ratio to 0.8; the gamma parameter of the SVM kernel function is set to 0.01, and the penalty factor C=1.5.

[0097] (3) Based on the comparison result of the demulsibility prediction value and the preset threshold value, the current state of the lubricant is judged to be a failure state or a normal state; after judging the current state of the lubricant to be a failure state or a normal state, the demulsibility failure warning is triggered, and a maintenance decision is generated: the current state of the lubricant and the maintenance decision are pushed to the operation and maintenance terminal through the Industrial Internet of Things (IIoT) platform and recorded in the cloud database; specifically, this is achieved through step S5:

[0098] S5. Based on the comparison result between the demulsibility prediction value and the preset threshold value, determine whether the current state of the lubricant is a failure state or a normal state, trigger the demulsibility failure warning level, and generate a maintenance decision:

[0099] In this embodiment, the preset threshold is set to 30 minutes. The hierarchical warning logic is:

[0100] Level 1 warning (failure state): If t sep If the time exceeds 30 minutes, it is judged as anti-emulsification failure, the current state is marked as failure, and the maintenance decision is: trigger the sound and light alarm and generate a forced oil change instruction;

[0101] Normal state: If t sep ≤30 minutes, mark the current state as normal, and the maintenance decision is: continue to monitor the oil status;

[0102] The current status of the lubricant and maintenance decisions are pushed to the operation and maintenance terminal through the Industrial Internet of Things (IIoT) platform and recorded in the cloud database.

[0103] Example 2

[0104] like Figure 2As shown, the multi-source sensing lubricant real-time anti-emulsification prediction system provided by this embodiment includes: a multi-source sensing module 5, a data analysis module 3 and a cloud-based early warning module 1.

[0105] In this embodiment, wear particles generated by the friction pairs in the equipment are divided into ferromagnetic particles 8 and non-ferromagnetic particles 9, which are stored in an oil tank 7. During system operation, the oil is delivered to the multi-source sensing module 5 via a gear pump 6 for monitoring. To reduce external environmental interference and ensure system stability and prediction accuracy, the above devices and modules are installed in a housing 10, and the entire system is powered by a power supply unit 11.

[0106] In this embodiment, the multi-source sensing module 5 includes an abrasive particle detection unit 2 and an oil physical and chemical property monitoring unit 4. The sensors monitor seven parameters: dielectric constant, emulsified water content, acid value, temperature, viscosity, abrasive particle concentration, and material. Specifically:

[0107] Abrasive particle detection unit 2, real-time acquisition of abrasive particle concentration C and abrasive particle material classification results;

[0108] The oil physical and chemical property monitoring unit 4 includes:

[0109] The multi-frequency dielectric constant sensor integrates a temperature compensation circuit to monitor the oil dielectric constant ε in real time. By synchronously detecting the difference between the dielectric constant and loss factor at high and low frequencies, the emulsified water content M is calculated in combination with a calibration formula and a temperature dynamic compensation model.

[0110] Acid value sensor, using potentiometric titration to determine the total acid value TAN;

[0111] Viscosity sensor, measures dynamic viscosity and calculates viscosity change rate Δμ / Δt.

[0112] In this embodiment, the sensors in the oil physical and chemical property monitoring unit 4 are integrated into a valve block to improve integration and shield interference; a serial port is also provided to transmit data to the data analysis module via serial communication;

[0113] In this embodiment, the data analysis module 3 stores the demulsibility prediction model trained in Example 1: the maximum depth of GBDT is 7 layers, the learning rate is 0.05, the subsample ratio is 0.8, the SVM model adopts the radial basis kernel function (RBF kernel), the penalty factor C = 1.5, gamma = 0.01; the operating parameters collected by the multi-source sensing module 5 are used as input, the demulsibility prediction value is output, and it is transmitted to the cloud warning module 1.

[0114] The cloud warning module 1 includes a real-time data dashboard that dynamically displays the demulsibility index, wear risk level and abrasive distribution heat map; the graded warning logic is that if t sep>30 minutes, the "anti-emulsification failure" alarm is triggered. It is set according to ASTM D1401 and GB / T 8022-2019 standards and meets the actual needs of industrial equipment lubricants; maintenance decision push is made and operation instructions are generated according to the warning level.

[0115] In this embodiment, taking a certain mineral lubricant as an example and the parameters monitored at a certain moment as an example, the system collects parameters such as the emulsified water content (1.18%), dielectric constant change (+2.7), acid value (0.053mgKOH / g), temperature (68°C), viscosity change rate (-0.06cP / s), abrasive concentration (level 8) and material (mainly Fe) of the lubricant in real time through a multi-source sensing module.

[0116] The data analysis module 3 processes the data collected by the multi-source sensing module 5 through sliding window filtering, Min-Max / Z-Score hybrid normalization and PCA dimensionality reduction to eliminate noise interference and compress redundant information, thereby generating a standardized input vector.

[0117] The preprocessed features are input into the GBDT-SVM integrated model pre-trained based on the above parameters. GBDT weights the features, and SVM calculates the anti-emulsification prediction value t through RBF kernel function mapping. sep = 21.7 minutes. The actual laboratory measurement during the same period was 22 minutes, with an error of only 0.3 minutes and a relative error of 1.36%, meeting the test criteria. Based on the predicted results, the system triggered a "normal" status flag and recommended continued monitoring.

[0118] The continuous prediction results of the embodiment are as follows Figure 3 As shown, it should be noted that in order to make the difference between the predicted value and the actual value of demulsibility more obvious, the y-axis adopts a logarithmic scale and the x-axis still maintains a linear scale. Figure 3 As can be seen, as the number of measurements increases over time, the oil-water separation time (demulsibility) of the lubricant increases, leading to the conclusion that the lubricant's demulsibility gradually weakens or even fails. This is consistent with the actual physical phenomenon that as the oil's operating time progresses, parameters such as emulsified water content, abrasive concentration, and viscosity continuously change, leading to a continuous weakening of the oil's demulsibility. Furthermore, the error between the predicted and true values ​​is minimal, meeting the test criteria, demonstrating that the method and system established by this invention can be used for real-time monitoring and early warning of lubricant oil demulsibility.

[0119] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A multi-source sensing method for predicting the real-time demulsibility of lubricating oil, characterized in that: The following steps are involved: (1) A demulsibility prediction model consisting of a gradient boosted decision tree (GBDT) and a support vector machine (SVM) was constructed and pre-trained using historical data: the oil type and operating conditions in the historical data were used as input features, and the corresponding measured lubricant demulsibility values ​​were used as training targets. The importance weights of the input features were extracted through GBDT, and nonlinear regression was performed on the input features based on the kernel function mapping of the support vector machine (SVM) combined with the importance weights to obtain the trained demulsibility prediction model. (2) Real-time collection of lubricant oil type and operating conditions as input to the trained demulsibility prediction model, which then outputs the demulsibility prediction value; (3) According to the comparison result of the demulsibility prediction value and the preset threshold value, the current state of the lubricant is judged as a failure state or a normal state.

2. The multi-source sensing lubricant real-time demulsibility prediction method according to claim 1, characterized in that: In steps (1) and (2), the working conditions include: dielectric constant, emulsified water content, acid value, temperature, viscosity and abrasive concentration, and the oil type is the result of abrasive material classification.

3. The multi-source sensing lubricant real-time demulsibility prediction method according to claim 1, characterized in that: In step (2), the oil type and operating conditions are filtered, normalized or feature fused to generate a standardized input vector which is input into the trained demulsibility prediction model.

4. The multi-source sensing lubricant real-time demulsibility prediction method according to claim 3, characterized in that: Methods for filtering, normalizing, or fusing features of oil type and operating conditions include: Filtering: Sliding window filtering is performed on the original signals of dielectric constant and viscosity, and abnormal points are replaced by the median value within the window; a second-order Butterworth low-pass filter is used on the wear particle concentration signal to suppress high-frequency noise; Normalization adopts a hybrid normalization strategy: the Min-Max normalization method is adopted for the emulsified water content M and the acid value TAN; the Z-Score normalization is adopted for the viscosity change rate Δμ / Δt; and the logarithmic compression is adopted for the dielectric constant change Δε; Feature fusion processing, time series analysis of viscosity change rate Δμ / Δt is performed to extract statistical features within the sliding time window, including mean, variance and slope; vector encoding is performed on the abrasive material classification results to generate material feature vector N=[n1,n2,...,n i ], where n i Represents the probability weight of the i-th type of material.

5. The multi-source sensing lubricant real-time demulsibility prediction method according to claim 1, characterized in that: In step (3), the method for determining whether the current lubricating oil is in a failed state or a normal state includes: Level 1 warning: Demulsibility prediction value t sep >Preset threshold, judged as anti-emulsification failure, marked as failure state; Normal state: If t sep ≤ the preset threshold, marked as normal status.

6. The multi-source sensing lubricant real-time demulsibility prediction method according to claim 5, characterized in that: The method comprises: S4. After determining whether the current state of the lubricant is a failure state or a normal state, an anti-emulsification failure warning is triggered and a maintenance decision is generated: The current lubricating oil is in a failed state, triggering an audible and visual alarm and generating a forced oil change instruction; The current lubricating oil is in normal condition and the oil status is continuously monitored; S5. The current status of the lubricant and maintenance decisions are pushed to the operation and maintenance terminal through the Industrial Internet of Things (IIoT) platform and recorded in the cloud database.

7. A multi-source sensing lubricant real-time demulsibility prediction system using the method according to any one of claims 1 to 6, characterized in that: The system comprises: The multi-source sensing module is used to monitor and collect lubricating oil parameters in real time, including dielectric constant, emulsified water content, acid value, temperature, viscosity, wear particle concentration, and oil type, and transmit the information to the data processing and analysis module; The data processing and analysis module stores the trained demulsibility prediction model, uses the multi-source sensor module to monitor and collect lubricant parameters in real time as input, outputs the demulsibility prediction value, and transmits it to the cloud-based early warning module; The cloud-based early warning module determines whether the current state of the lubricant is a failure state or a normal state based on the comparison result of the anti-emulsification prediction value with the preset threshold, triggers the anti-emulsification failure warning, and generates maintenance decisions.

8. The multi-source sensing lubricant oil real-time demulsibility prediction system according to claim 7, characterized in that: The multi-source sensing module includes: The multi-frequency dielectric constant sensor integrates a temperature compensation circuit to monitor the oil dielectric constant ε in real time. By synchronously detecting the difference between the dielectric constant and loss factor at high and low frequencies, the emulsified water content M is calculated in combination with a calibration formula and a temperature dynamic compensation model. Acid value sensor, using potentiometric titration to determine the total acid value TAN; Viscosity sensor, which measures dynamic viscosity and calculates the viscosity change rate Δμ / Δt; The abrasive particle detection unit obtains the abrasive particle concentration C and abrasive particle material classification results in real time.

9. The multi-source sensing lubricant oil real-time demulsibility prediction system according to claim 7, characterized in that: The multi-frequency dielectric constant sensor integrated temperature compensation circuit, acid value sensor and viscosity sensor in the multi-source sensing module are integrated into a sensor valve block, and the communication method adopts serial communication; the sensor valve block adopts electromagnetic shielding coating, and signal transmission adopts differential signal line.

10. The multi-source sensing lubricant oil real-time demulsibility prediction system according to claim 7, characterized in that: The cloud-based early warning module includes a real-time data dashboard that dynamically displays demulsibility indicators, wear risk levels, and abrasive distribution thermal maps.

Images