Method and device for on-line analysis of the quality of oil in a gas turbine oil system

Abstract

This disclosure relates to an online oil quality analysis method and apparatus for a gas turbine oil system. The method includes: acquiring oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine; determining a comprehensive oil quality score based on the oil monitoring data; performing conflict detection between the comprehensive oil quality score and the equipment performance parameters using a pre-set conflict detection rule library; when a conflict is detected, using a pre-set physicochemical failure model library to deduce and diagnose the root cause of the conflict, determining the oil degradation mode leading to the conflict; correcting the parameters of the corresponding physicochemical failure model based on the oil degradation mode, and regenerating the comprehensive oil quality score using the corrected physicochemical failure model. This solution improves the reliability of the gas turbine oil system operation.

Description

Technical Field

[0001] This disclosure relates to the field of gas turbine technology, and in particular to a method and apparatus for online analysis of oil quality in a gas turbine oil system. Background Technology

[0002] In related technologies, lubricating oil and control oil systems are key components of gas turbines, providing lubrication and cooling for bearings and supplying working media for hydraulic valves. Currently, existing technologies mainly rely on periodic shutdowns for sampling and offline testing to monitor oil quality. This method typically only detects basic parameters such as viscosity, acid value, and moisture content, and these parameters are isolated, making it difficult to comprehensively and in real-time reflect the overall health of the oil. Furthermore, due to the inability to achieve continuous online monitoring, existing technologies struggle to detect slow oil degradation processes in a timely manner, such as the risk of servo valve jamming caused by varnish formation. This lag not only increases the risk of unplanned equipment downtime but also fails to provide data support for predictive maintenance, leading to higher safety risks and economic losses for gas turbines during long-term operation. Summary of the Invention

[0003] To overcome the problems existing in related technologies, this disclosure provides a method and apparatus for online analysis of oil quality in a gas turbine oil system.

[0004] According to a first aspect of the present disclosure, an online oil quality analysis method for a gas turbine oil system is provided, comprising:

[0005] Acquire oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and determine a comprehensive oil quality score based on the oil monitoring data;

[0006] The comprehensive oil quality score and the equipment performance parameters are checked for discrepancies using a pre-defined discrepancy detection rule base.

[0007] When the result of the contradiction detection is that a contradiction exists, the root cause of the contradiction is inferred and diagnosed using a preset physical and chemical failure model library to determine the oil deterioration mode that leads to the contradiction.

[0008] Based on the oil degradation mode, the parameters of the corresponding physicochemical failure model are corrected, and the comprehensive oil quality score is regenerated using the corrected physicochemical failure model.

[0009] According to a second aspect of the present disclosure, an online oil quality analysis device for a gas turbine oil system is provided, comprising:

[0010] The scoring unit is used to acquire oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and to determine a comprehensive oil quality score based on the oil monitoring data.

[0011] The detection unit is used to perform conflict detection between the comprehensive oil quality score and the equipment performance parameters using a preset conflict detection rule library;

[0012] The diagnostic unit is used to, when the result of the contradiction detection is that there is a contradiction, use a preset physical and chemical failure model library to reason and diagnose the root cause of the contradiction, and determine the oil deterioration mode that leads to the contradiction.

[0013] The generation unit is used to modify the parameters of the corresponding physicochemical failure model according to the oil deterioration mode, and regenerate the comprehensive oil quality score using the modified physicochemical failure model.

[0014] According to a third aspect of the present disclosure, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as described in any one of the first aspects.

[0015] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of the first aspects.

[0016] According to a fifth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method as described in any one of the first aspects.

[0017] The technical solutions provided by the embodiments of this disclosure can include the following beneficial effects: by acquiring oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, a multi-dimensional data foundation is provided for subsequent evaluation; a comprehensive oil quality score is determined based on the oil monitoring data, integrating multiple isolated parameters into a quantitative indicator to intuitively reflect the oil health status; a preset contradiction detection rule library is used to detect contradictions between the comprehensive oil quality score and the equipment performance parameters, enabling timely detection of inconsistencies between the oil status and equipment operation performance, avoiding misjudgment from a single data source; when the contradiction detection result indicates a contradiction, a preset physicochemical failure model library is used to deduce and diagnose the root cause of the contradiction, determine the oil degradation mode leading to the contradiction, correct the parameters of the corresponding physicochemical failure model based on the oil degradation mode, and regenerate the comprehensive oil quality score using the corrected physicochemical failure model, thus constructing a closed-loop self-correction mechanism from data to model to physical facts, enabling the monitoring system to have self-learning and adaptive capabilities. Through the above technical solutions, this disclosure significantly improves the reliability of gas turbine oil system operation.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0020] Figure 1 This is a flowchart illustrating an online oil quality analysis method for a gas turbine oil system according to an exemplary embodiment.

[0021] Figure 2 This is a block diagram illustrating an online oil quality analysis device for a gas turbine oil system according to an exemplary embodiment.

[0022] Figure 3 This is a block diagram illustrating an apparatus for an online oil quality analysis method for a gas turbine oil system, according to an exemplary embodiment.

[0023] Figure Labels

[0024] 201-Scoring unit; 202-Detection unit; 203-Diagnostic unit; 204-Generation unit; 300-Device; 302-Processing component; 304-Memory; 306-Power component; 308-Multimedia component; 310-Audio component; 312-I / O interface; 314-Sensor component; 316-Communication component; 320-Processor. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0026] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. The singular forms “a” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0027] It should be understood that although the terms first, second, third, etc., may be used to describe various information in embodiments of this disclosure, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of embodiments of this disclosure, and similarly, second information may also be referred to as first information. Depending on the context, the words “if” and “suppose” as used herein may be interpreted as “when”, “when”, or “in response to a determination”.

[0028] Furthermore, various forms of processes shown in the embodiments of this disclosure can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and no limitation is imposed herein.

[0029] It should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0030] Figure 1 This is a flowchart illustrating an online oil quality analysis method for a gas turbine oil system according to an exemplary embodiment, such as... Figure 1 As shown, it should be noted that the online oil quality analysis method for a gas turbine oil system according to this embodiment is applied in an online oil quality analysis device for a gas turbine oil system. For example... Figure 1 As shown, the method may include the following steps:

[0031] Step 101: Obtain oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and determine the comprehensive oil quality score based on the oil monitoring data.

[0032] In this embodiment, two types of basic data need to be acquired from the multi-source data acquisition system: one is oil monitoring data reflecting the physicochemical properties of the oil itself, and the other is equipment performance parameters reflecting the operating status of key components of the gas turbine. These two types of data characterize the health status of the system from the two dimensions of the oil system and the equipment itself, respectively, forming the data foundation for the entire process of subsequent comprehensive evaluation, contradiction detection, root cause diagnosis, and model correction.

[0033] In some embodiments of this disclosure, the oil monitoring data includes at least two of the following: viscosity index, acid value, base value, moisture content, particulate contamination level, dielectric constant, varnish index, and oil temperature. The equipment performance parameters include at least one or more of the following: servo valve response time, bearing vibration amplitude, bearing temperature, and lubricating oil differential pressure.

[0034] As an example of a possible implementation, a multi-parameter integrated oil sensor can be installed on the main return oil line of the lubricating oil pipeline. This sensor integrates a viscometer, an infrared spectroscopy analysis module, a capacitive moisture sensor, and a particle counter, and can output the aforementioned parameters in real time. The sensor is connected to a data acquisition terminal via an RS485 bus, and the data acquisition frequency can be configured to 1 time / minute. Equipment performance parameters are acquired through a DCS (Distributed Control System) or PLC (Programmable Logic Controller) and read in real time via the OPCUA (OPC Unified Architecture) protocol.

[0035] In some embodiments of this disclosure, the oil monitoring data is smoothed before determining the overall oil quality score based on the oil monitoring data.

[0036] Specifically, the following formula is used to smooth the oil monitoring data:

[0037] z t = ·x t + (1-λ)·z t-1

[0038] Where, x t The original monitoring value at the current moment, z t z is the smoothed value at the current time. t-1 This is the smoothed value from the previous time step. λ is a smoothing factor. As an example, for parameters with large fluctuations such as vibration, λ can be taken as 0.15; for relatively stable parameters such as viscosity, λ can be taken as 0.3.

[0039] In this embodiment, by smoothing the oil monitoring data before determining the comprehensive oil quality score, random noise and transient interference introduced during sensor acquisition can be effectively filtered out, improving the signal-to-noise ratio and stability of the input data. The exponentially weighted moving average method is used to recursively calculate the original monitoring values, assigning higher weights to recent data. This preserves the dynamic characteristics of data changes while eliminating abnormal jumps caused by electromagnetic interference or sensor jitter. The smoothed time series data can more realistically reflect the long-term trend of oil parameters, avoiding drastic fluctuations in the comprehensive oil quality score due to anomalies in single-point data, thereby improving the accuracy and reliability of the score calculation.

[0040] Furthermore, differentiated smoothing factors can be configured for the characteristics of different parameters. For example, a smaller smoothing factor can be used for vibration-related parameters with large fluctuations to respond quickly to changes, while a larger smoothing factor can be used for relatively stable parameters such as viscosity to enhance trend smoothing, thus enabling the preprocessing process to have adaptive capabilities. Through the above technical means, this disclosure provides a high-quality data foundation for subsequent comprehensive oil quality scoring, contradiction detection, and root cause diagnosis, significantly improving the anti-interference capability and evaluation accuracy of the online oil quality monitoring system.

[0041] In some embodiments of this disclosure, determining the comprehensive oil quality score based on oil monitoring data includes calculating the comprehensive oil quality score M according to the following formula:

[0042]

[0043] Wherein, VI is the viscosity index, w1 is the weighting coefficient of VI, TAN is the acid value (mgKOH / g), w2 is the weighting coefficient of TAN, TBN is the base value (mgKOH / g), w3 is the weighting coefficient of TBN, a is the moisture content, w4 is the weighting coefficient of a, d is the operating time (h) since the last oil quality monitoring passed, w5 is the weighting coefficient of d, c is the dielectric constant, and w6 is the weighting coefficient of c. The coating film index, W7 is... The weighting coefficients are t, where t is the oil temperature (°C) and w8 is the weighting coefficient of t.

[0044] It should be noted that the above formula uses 100 points as the full score benchmark. Each item is subtracted from a penalty term, the size of which depends on the deviation between the current measured value and the ideal benchmark value. The ideal benchmark value for viscosity index VI is set at 90, which is the typical design value for most gas turbine lubricants. The ideal benchmark value for TAN (Total Acid Number) is set at 0.2 mgKOH / g; exceeding this value indicates that the oil has begun to oxidize. The ideal benchmark value for TBN (Total Base Number) is set at 9 mgKOH / g; below this value indicates the consumption of additives. The ideal benchmark value for moisture content a is set at 300 ppm; exceeding this value may affect oil film strength. The penalty term for operating time d is in the form of (d-4)², indicating that after 4 hours, the impact of time on the score increases quadratically. The penalty term for oil temperature t is multiplied by the operating time d, representing the coupling effect of high temperature and long-term operation.

[0045] In some embodiments of this disclosure, the weighting coefficients w1 to w8 are determined using the following steps:

[0046] Step a1: Obtain historical oil monitoring data of the gas turbine oil system within a preset time window, and perform data standardization processing on each monitoring indicator among VI, TAN, TBN, a, d, c, ΔE, and t.

[0047] In this embodiment, historical oil monitoring data within a preset time window (e.g., the past 30 or 60 days) can be acquired to form a dataset containing multiple time point samples and multiple monitoring indicators. Since different monitoring indicators have different dimensions and orders of magnitude—for example, viscosity index (VI) is typically around 90, acid value (TAN) is in the range of 0-1 mgKOH / g, and moisture content (a) may be as high as hundreds of ppm—if the raw values ​​are directly used for subsequent weight calculations, indicators with larger orders of magnitude will dominate the calculation results, leading to distorted weight allocation. Therefore, it is necessary to perform data standardization processing on each monitoring indicator separately to eliminate the influence of dimensions and map the values ​​of each indicator to a unified numerical range (usually [0,1]). As an example, the min-max standardization method can be used. For each indicator, its minimum and maximum values ​​within the preset time window are found. The minimum value is subtracted from the value of each sample point, and then divided by the range, so that the standardized data all fall between 0 and 1.

[0048] The standardized data retains the distribution characteristics and fluctuation information of the original data, but eliminates dimensional differences, making different indicators comparable and laying a unified scale foundation for subsequent calculations of sample proportions, information entropy, and dynamic weights for each indicator. This step ensures that the entropy weight method can objectively reflect the degree of variation of each indicator, avoiding subjective bias caused by dimensional differences.

[0049] Step a2: Based on the standardized data, calculate the proportion of the sample value at each time point under each monitoring indicator to the total value of all samples under that indicator, and use this as the proportion of each time point under the monitoring indicator.

[0050] In one embodiment, for each monitoring indicator (e.g., TAN, TBN, etc.), the standardized values ​​of all time-point samples under that indicator are summed to obtain the total value of that indicator. For each time-point sample under that indicator, its standardized value is divided by the total value of that indicator to obtain the proportion of that sample under that indicator. This proportion reflects the share or weight of the sample value at a specific time point in the entire indicator data sequence within a preset time window.

[0051] Understandably, the sum of the proportions of all time-point samples under the same indicator is 1, thus forming the probability distribution of that indicator. By calculating the proportions, the absolute values ​​of the original data are transformed into relative contributions, laying the probabilistic foundation for the subsequent calculation of information entropy. The essence of information entropy is precisely the measurement of the uncertainty or dispersion of the probability distribution. The larger the proportion, the more important the value at that time point occupies in the whole; the more uniform the proportion distribution, the smaller the data fluctuation and the less information the indicator contains; the greater the difference in the proportion distribution, the greater the data fluctuation and the richer the information contained in the indicator. Through this step, the standardized data is transformed into probabilistic proportions, enabling subsequent objective evaluation of the information content of each indicator based on information theory methods.

[0052] Step a3: Calculate the information entropy of each monitoring indicator based on its weight; information entropy is used to characterize the dispersion of the indicator data.

[0053] In one embodiment, information entropy is a core indicator in information theory used to measure the uncertainty of random variables. Its value reflects the degree of dispersion or disorder of the data. For a certain monitoring indicator, if the proportion distribution of samples at each time point is more uniform, that is, the smaller the numerical difference between samples and the more gradual the fluctuation, the greater the information entropy of the indicator, indicating that the indicator provides less new information within a preset time window and the data changes tend to be stable. Conversely, if the proportion distribution of samples at each time point under the indicator is significantly different, that is, the values ​​at some time points deviate significantly from the overall level and the data fluctuates violently, the smaller the information entropy of the indicator, indicating that the indicator contains richer information about changes and can more sensitively reflect the dynamic evolution of the system state. In specific calculation, for each monitoring indicator, the proportion of samples at each time point is taken as a probability distribution. The natural logarithm of each proportion is taken and multiplied by the proportion. Then, the products of all time points are summed and the negative value is taken to obtain the information entropy of the indicator.

[0054] Understandably, information entropy ranges from 0 to 1. When all samples have equal weight, the information entropy reaches its maximum value of 1, indicating that the indicator is completely disordered and has zero information content. When the weight of a sample is close to 1 while the weights of the remaining samples are close to 0, the information entropy approaches 0, indicating that the indicator is highly ordered and has a very large amount of information. This step quantifies the original fluctuation characteristics of each monitoring indicator into a unified information entropy index, providing a scientific basis for the subsequent objective evaluation of the importance of each indicator. Indicators with lower information entropy and greater data fluctuation should be given higher attention and weight.

[0055] Step a4: Determine the difference coefficient for each monitoring indicator based on information entropy; the difference coefficient is the difference between 1 and information entropy.

[0056] In one embodiment, to make the importance of an indicator positively correlated with its information content—that is, indicators with greater data fluctuations and richer information content should receive higher weights—information entropy needs to be converted into a difference coefficient.

[0057] It should be noted that the coefficient of variation, which is 1 minus the information entropy, measures the "difference" or "information richness" of the indicator data. When the information entropy of an indicator is high (close to 1), its coefficient of variation is low (close to 0), indicating that the indicator data fluctuates smoothly and has low discriminative power. When the information entropy of an indicator is low (close to 0), its coefficient of variation is high (close to 1), indicating that the indicator data fluctuates dramatically and can effectively distinguish the state differences at different points in time.

[0058] This transformation converts the inverse measure of information entropy into a positive measure of the difference coefficient. This allows for direct judgment of the importance of each indicator based on the magnitude of the difference coefficient; a larger difference coefficient indicates that the indicator provides more information within a preset time window and should occupy a more important position in the comprehensive evaluation. This step completes the crucial mapping from information entropy to the difference coefficient, laying a direct numerical foundation for subsequent dynamic weight calculations.

[0059] Step a5: Normalize the difference coefficients of each monitoring indicator to obtain the dynamic weight of each monitoring indicator, and use the dynamic weight of each monitoring indicator as the weight coefficients w1 to w8.

[0060] It is understandable that the difference coefficient reflects the information content of each indicator within a preset time window, but the sum of the difference coefficients of each indicator is usually not equal to 1, and cannot be directly used as a weighting coefficient.

[0061] Therefore, it is necessary to sum the difference coefficients of all monitoring indicators and use the sum as the normalization denominator. For each monitoring indicator, its difference coefficient is divided by the sum of the difference coefficients of all indicators to obtain the normalized weight value of that indicator.

[0062] This normalization process ensures that the sum of the weights of all indicators is 1, satisfying the basic mathematical constraints of the weight coefficients. The normalized weight values ​​are the dynamic weights of each indicator, and their magnitudes are positively correlated with the degree of data fluctuation of that indicator within a preset time window. Using the calculated dynamic weights as weight coefficients w1 to w8 in the comprehensive oil quality scoring formula enables adaptive adjustment of the scoring model: when an indicator (such as acid value or paint film index) fluctuates significantly over a period of time, its weight automatically increases, making the scoring model more sensitive to the current major risks; when the indicator tends to stabilize, its weight decreases accordingly, avoiding excessive focus on stable parameters.

[0063] Through the above steps, the entire process of dynamically determining weights using the entropy weight method is completed, enabling the scoring model to automatically optimize weight allocation based on the inherent information content of the data itself. This avoids bias caused by subjective weighting and significantly improves the objectivity and adaptability of oil quality assessment.

[0064] Understandably, the entropy weight method dynamically adjusts weights based on the volatility characteristics of the data itself. When a certain indicator fluctuates drastically over a period of time, its weight will automatically increase, making the scoring model more sensitive to the current main risks.

[0065] As an example, oil quality monitoring data can be collected within a certain time window (such as the last 30 days) to form a data matrix X={ } where i=1,2,…,m represents the number of samples at time points, and j=1,2,…,n represents the monitoring indicators (such as viscosity index, acid value, etc., a total of 8 indicators). Data standardization is performed on each indicator to eliminate the influence of dimensions. Since some indicators have positive (higher values ​​are better) or negative (lower values ​​are better) characteristics, min-max standardization is used to scale the data to the [0, 1] interval:

[0066]

[0067] Where min( ) and max( ) represents the minimum and maximum values ​​of indicator j across all samples. For negative indicators such as acid value (TAN), the reciprocal or reverse calculation can be performed first, but for simplicity, the original values ​​are used directly for standardization here because the entropy weight method is based on data variability and does not depend on the directionality of the indicator. The weight of each sample for each indicator is calculated. :

[0068]

[0069] Calculate information entropy:

[0070] For each index j, calculate its information entropy. :

[0071] Where k = 1 / ln(m) is the normalization factor, ensuring ∈[0,1]. When When all are equal, the entropy is maximized. =1) indicates that the data is highly disordered and contains little information.

[0072] Calculate weights: Calculate the coefficient of variation for each indicator. =1 . The larger the value, the greater the fluctuation of indicator j, and the more information it provides.

[0073] Final weight for:

[0074]

[0075] Weight It will be updated dynamically based on the latest data regularly (e.g., daily) and incorporated into the oil quality scoring formula.

[0076] Step 102: Use the preset contradiction detection rule library to perform contradiction detection on the comprehensive oil quality score and equipment performance parameters.

[0077] It should be noted that in this embodiment, it is necessary to determine whether there is an inconsistency, or "contradiction," between the oil condition and the actual operating condition of the equipment. This inconsistency may stem from blind spots in the oil quality scoring model, sensor drift, or the fact that certain oil degradation factors not directly monitored have begun to affect equipment performance. Through this step, the previously isolated dimensions of oil monitoring and equipment performance monitoring are integrated, achieving system-level cross-validation.

[0078] In some embodiments of this disclosure, the contradiction detection rule base includes at least one of the following rules:

[0079] Rule R1: If the overall oil quality score is higher than the first threshold and the servo valve response time exceeds the set percentage of the historical baseline value, then a contradiction is determined.

[0080] As an example, the first threshold can be set to 75 points, and the percentage can be set to 20%. That is, if the oil quality score is ≥75 points, but the servo valve response time exceeds 20% of the historical baseline value, a contradiction is considered. The physical basis of this rule is that the servo valve is extremely sensitive to the cleanliness of the oil. If the oil quality score shows good but the response is slow, it may mean that there are unmonitored factors such as varnish.

[0081] Rule R2: If the decrease in the overall oil quality score does not exceed the set threshold, and the bearing vibration amplitude increases beyond the set threshold within a set time, and a characteristic frequency component corresponding to the preset fault type is detected in the vibration signal, then a contradiction is determined.

[0082] In rule R2, "decline not exceeding the set threshold" means that the oil quality score has not decreased significantly (e.g., a decrease of less than 5 points), "increase exceeding the set threshold within a set time" means, for example, the vibration amplitude increases by more than 80% of the alarm threshold within 1 hour, and "characteristic frequency components" means, for example, the frequency of rolling bearing failure. This rule can detect situations where the oil quality has not deteriorated significantly but the equipment has already shown abnormal signs.

[0083] Rule R3: If the overall oil quality score remains stable for a preset period of time, and the bearing temperature continues to rise above the set value, then a contradiction is determined.

[0084] In Rule R3, the "continuous preset duration" can be set to 7 days, and the "set value" can be set to 5℃. That is, if the oil quality score is stable above 80 points for 7 consecutive days, but the bearing temperature continues to rise above 5℃, there is a contradiction in the judgment.

[0085] When any of the above rules is triggered, the system determines that there is a contradictory event and proceeds to step 103 for root cause diagnosis.

[0086] In this embodiment, a cross-validation mechanism between oil condition and equipment performance is established, which can promptly detect potential risks that cannot be captured by a single data source, and solve the blind spot problem of traditional monitoring systems that only report data but do not verify it.

[0087] Step 103: When the result of the contradiction detection is that there is a contradiction, the root cause of the contradiction is inferred and diagnosed by using the preset physical and chemical failure model library, and the oil deterioration mode that leads to the contradiction is determined.

[0088] In this embodiment, a physicochemical failure model library is required to conduct in-depth analysis of contradictory events and identify the root causes of the contradictions. These models are built based on the physicochemical mechanisms of oil degradation and can explain the causal relationship between changes in oil parameters and equipment performance at the mechanistic level. This allows us not only to know that a contradiction has occurred, but also to understand why the contradiction occurred.

[0089] In some embodiments of this disclosure, the physicochemical failure model library includes an oxidation failure model, a moisture intrusion model, and a particulate contamination / wear model. Specifically: the oxidation failure model describes the relationship between oil temperature, operating time, and changes in acid value and paint film index; the moisture intrusion model describes the relationship between moisture content and changes in dielectric constant and oil film strength; and the particulate contamination / wear model describes the relationship between particle count, vibration characteristics, and degree of wear.

[0090] In some embodiments of this disclosure, step 103 may specifically include the following sub-steps:

[0091] Step b1: In response to the result of the contradiction detection indicating the existence of a contradiction, obtain the oil monitoring data and equipment performance parameters corresponding to the contradiction to obtain the data to be diagnosed.

[0092] Specifically, for the contradictory event triggered in step 102, the system will extract the oil monitoring data and equipment performance parameters at the trigger time and for a period of time before and after (e.g., 1 hour before and after) as data to be diagnosed.

[0093] Step b2: Extract abnormal device performance features from the data to be diagnosed.

[0094] Abnormal characteristics include, but are not limited to: the increment of servo valve response time, the rate of change of vibration amplitude, the energy increase of a specific frequency band in the vibration spectrum, and the rate of increase of bearing temperature.

[0095] Step b3: Input the oil monitoring data and abnormal equipment performance characteristics from the data to be diagnosed into multiple candidate failure models for simulation and deduction, and obtain the simulation and deduction results of each candidate failure model.

[0096] As an example, for the contradiction triggered by rule R1, the current moisture content, acid value, paint film index, particle size, and other data, as well as the abnormal characteristics of the servo valve response time, can be input into the oxidation failure model, moisture intrusion model, and particle contamination / wear model respectively, and parallel simulation can be performed to obtain the theoretical output values ​​of each model.

[0097] In one embodiment, the simulation results obtained in step b3 may refer to the theoretical prediction sequence output by each model after inputting the oil monitoring data in the data to be diagnosed and the abnormal features extracted from the equipment performance parameters into multiple candidate failure models such as oxidation failure model, water intrusion model, and particulate contamination / wear model, and performing parallel simulation calculations based on its built-in physicochemical mechanism equations.

[0098] Each candidate failure model corresponds to a specific oil degradation path. For example, the oxidation failure model, based on oxidation kinetics, calculates the theoretical trends of acid value and varnish index based on the input oil temperature and operating time; the moisture intrusion model calculates the theoretical value of dielectric constant based on the nonlinear mapping relationship between moisture content and dielectric constant; and the particulate contamination / wear model calculates the theoretical characterization of wear degree based on the correlation between particle count and vibration characteristic frequency. These simulation results are not generated out of thin air, but rather are quantitative answers from each failure model to the assumption of "how the oil parameters should change if the current degradation mode is the type represented by this model".

[0099] By simultaneously inputting the same set of data to be diagnosed into multiple models for parallel extrapolation, theoretical output sequences under different degradation assumptions can be obtained, providing a direct basis for comparison with actual monitoring values. This enables root cause diagnosis to select the one that best matches the actual data from a variety of possible degradation modes.

[0100] Step b4: Calculate the matching degree between each simulation result and the corresponding oil monitoring data in the data to be diagnosed.

[0101] As an example of a possible implementation, the degree of matching can be measured using root mean square error (RMSE) or coefficient of determination (COP). The smaller the RMSE or the closer the COP is to 1, the better the simulation results of the model match the actual data.

[0102] Step b5: Output the oil deterioration mode corresponding to the candidate failure model with the highest matching degree as the diagnostic result; the diagnostic result includes the confidence probability corresponding to the candidate failure model with the highest matching degree.

[0103] In some embodiments of this disclosure, it is necessary to select the oil deterioration mode corresponding to the model with the highest matching degree as the final diagnostic conclusion based on the matching degree between each candidate failure model calculated in step b4 and the actual data.

[0104] It should be noted that the highest matching degree indicates that the physicochemical mechanism assumptions of the model are most consistent with the current actual monitoring data, that is, the degradation path represented by the model can best explain the currently observed changes in oil parameters and abnormal equipment performance.

[0105] For example, if the simulation results of the oxidation failure model match the actual acid value and paint film index change trends most closely, the diagnosis result is "accelerated oil oxidation leads to an increase in paint film tendency"; if the moisture intrusion model matches most closely, the diagnosis result is "slight increase in moisture leads to a decrease in oil film strength".

[0106] In one embodiment, the diagnostic result includes not only a qualitative description of the degradation pattern but also a confidence probability of the model. The confidence probability is a quantitative value derived from a matching index (such as root mean square error or coefficient of determination), reflecting the reliability of the diagnostic conclusion. The higher the matching degree, the greater the confidence probability, indicating that the diagnostic result is more credible.

[0107] As an example, the coefficient of determination R² can be directly used as the confidence probability output, i.e., R²=0.85 corresponds to an 85% confidence level.

[0108] This step transforms the data-driven matching degree calculation results into physically meaningful and interpretable diagnostic conclusions, providing clear fault location information for operation and maintenance personnel and pointing the way for subsequent model parameter correction, thus achieving a key leap from "anomaly detection" to "root cause diagnosis".

[0109] For example, if the oxidation failure model has the highest matching degree, the diagnostic conclusion is output as: "Accelerated oil oxidation leads to increased paint film tendency (confidence 85%)"; if the moisture intrusion model has the highest matching degree, the conclusion is output as: "Slight increase in moisture leads to decreased oil film strength (confidence 72%)".

[0110] Step 104: Based on the oil deterioration mode, modify the parameters of the corresponding physicochemical failure model, and regenerate the comprehensive oil quality score using the modified physicochemical failure model.

[0111] In this embodiment, the failure model is optimized in a closed loop based on the diagnostic results (i.e., the oil degradation mode) to make the model more closely match the actual degradation state of the current equipment and output a corrected, more reliable oil quality score. This allows the model to learn from each contradictory event and continuously adjust its parameters to adapt to the personalized degradation trajectory of the equipment.

[0112] In some embodiments of this disclosure, parameter correction of the corresponding physicochemical failure model includes at least one of the following methods:

[0113] Method 1: Adjust the kinetic parameters of the physicochemical failure model based on the diagnostic results.

[0114] For example, if the diagnosis indicates accelerated oxidation, the system automatically adjusts the reaction rate constant in the oxidation failure model to better reflect the actual degradation rate of the equipment. Specifically, the system can use the least squares method to fit recent data, back-calculate the optimal model parameters, and update them.

[0115] Method 2: Based on the diagnostic results, perform data compensation or mark suspicious tags on the sensors involved in the resulting contradictions.

[0116] For example, if a diagnosis reveals that the moisture sensor has a long-term drift (manifested as a systematic deviation between the simulated value and the measured value of the moisture model), the system can initiate a model-based soft calibration: using the dielectric constant to reverse-calculate the "theoretical value" of moisture through the moisture intrusion model, and compensate for the reading of the moisture sensor, while marking the sensor as "suspicious" to prompt maintenance personnel to check it.

[0117] Method 3: Based on the diagnostic results, increase the weight of the corresponding monitoring indicators in the comprehensive oil quality score within a preset time period.

[0118] For example, if the diagnosis is that oxidation is the main cause, the weights of the acid value and paint film index calculated by the entropy weight method will be temporarily increased in the scoring calculation for a period of time, so that the scoring model can more sensitively reflect the current main problem.

[0119] After correction, the system regenerates the comprehensive oil quality score using the revised physicochemical failure model, and includes diagnostic optimization notes in the output. For example: "The current score has been sensitively optimized based on the detected oxidation acceleration trend. Main risk: varnish formation; it is recommended to pay attention to the efficiency of the oil filter and consider oil testing."

[0120] In this embodiment, the self-learning and adaptive capabilities of the monitoring system are realized, enabling the model to be continuously optimized as the equipment operates, thus solving the problem of decreased accuracy of traditional fixed models during long-term operation.

[0121] Through steps 101 to 104 above, this embodiment realizes a complete closed-loop process from data acquisition, comprehensive scoring, contradiction detection, root cause diagnosis to model self-correction, which significantly improves the accuracy and reliability of online oil quality monitoring.

[0122] In some embodiments, oil data sensors can be used to collect multi-dimensional raw monitoring data of the gas turbine oil system in real time. Subsequently, oil data preprocessing performs smoothing, filtering, and standardization on the raw data to eliminate noise interference. Based on this, oil health assessment calculates a comprehensive oil quality score based on the preprocessed data to quantitatively evaluate the oil health status. When the score is abnormal or a contradiction is detected, fault diagnosis and localization call the physical and chemical failure model library for root cause diagnosis to determine the oil deterioration mode. Based on the diagnosis results, maintenance management based on the predicted data generates corresponding maintenance suggestions, including starting the oil filter, taking samples for testing, or issuing early warnings. Finally, oil maintenance execution triggers specific maintenance operations according to the suggestions, forming a complete closed-loop process from data acquisition, status assessment, fault diagnosis to maintenance execution.

[0123] In some embodiments of this disclosure, based on the time series of historical oil monitoring data, a pre-trained deep learning model is used to predict the remaining service life of the oil, and the predicted value of the remaining service life and its prediction range are output; the deep learning model is a long short-term memory network model or a Transformer model.

[0124] In some embodiments of this disclosure, the deep learning model takes a multivariate time series as input with dimensions [T, N], where T is the time step and N is the feature dimension; the model also outputs the remaining useful life predictions for the 5th, 50th and 95th quantiles.

[0125] As an example, complete historical oil monitoring data from multiple gas turbines of the same model were collected. Each data point must start from new oil and continue until the oil is scrapped or replaced, forming a complete lifecycle sequence. A sliding window method was used to construct training samples, using multi-parameter data from the past 60 days as input features and the time from the current moment to the failure point as the label RUL (Remaining Useful Life).

[0126] Feature extraction can be performed using a two-layer LSTM (Long Short-Term Memory) network, followed by a fully connected output layer that outputs three scalar values, representing the 5th quantile, median, and 95th quantile of the RUL. A quantile loss function is used to teach the model to predict different percentiles.

[0127] In real-time operation, the model uses preprocessed and dynamically weighted multi-parameter time series data as input to directly predict the remaining service life of the oil and outputs the prediction range. For example, the model outputs: "Predicted remaining service life: 45 days (90% confidence interval: 38-52 days)".

[0128] This embodiment extends the monitoring system's capabilities from "current status assessment" to "future trend prediction," providing a direct basis for predictive maintenance decisions.

[0129] In some embodiments, complete historical oil monitoring data from multiple gas turbines of the same model can be collected. Each data point should start from new oil and continue until the oil is scrapped or replaced, forming a complete "lifecycle" sequence. A sliding window method is used to construct training samples. For a complete lifecycle sequence of length L, at any time t, data from the past T days is taken as input features, and the time length from time t to the failure point is taken as the label RUL_t. This generates a large number of (input sequence, RUL label) training pairs. The RUL label has a large value in the early stages of the sequence and approaches 0 near the failure point, showing a monotonically decreasing trend.

[0130] In some embodiments of this disclosure, the deep learning model employs an encoder-decoder architecture based on a Long Short-Term Memory (LSTM) network or a Transformer to predict remaining useful life. Specifically, the model input layer receives a multivariate time series of shape [T, N], where T is the time step (e.g., a historical window of the past 60 days), and N is the feature dimension (including key oil parameters such as viscosity index, acid value, moisture content, particle size, and paint film index after preprocessing and dynamic weighting). The input data is first processed by 2 to 3 layers of LSTM or Transformer encoders for feature extraction. These encoder layers can effectively learn the complex nonlinear interactions between parameters and the long-term dependency patterns of parameters over time. Subsequently, the feature aggregation layer uses an attention mechanism or global average pooling in the time dimension to filter out the most critical information for the prediction task from the high-dimensional features output by the encoder, achieving feature dimensionality reduction and fusion. The output layer is a fully connected layer that outputs three scalar values, representing the 5th quantile, median (50th quantile), and 95th quantile of the remaining useful life, respectively, thus providing not only point prediction values ​​but also the uncertainty range of the prediction.

[0131] During the model training phase, the quantile loss function is used as the optimization objective, and its mathematical expression is:

[0132] Loss = Σ [ (q·max(0, y_true - y_pred_q)) + ((1-q)·max(0, y_pred_q- y_true)) ]

[0133] Where Loss is the loss value, q is the quantile (values ​​are 0.05, 0.5, and 0.95), y_true is the true remaining useful life, and y_pred_q is the corresponding quantile predicted by the model. The unique advantage of this loss function is its ability to simultaneously train median and interval predictions. When q = 0.5, the loss function drives the model to learn median regression; when q = 0.05 and 0.95, the loss function guides the model to learn the upper and lower boundaries of the prediction interval. By minimizing the quantile loss, the model can learn to output a reasonable prediction range at different confidence levels, providing more comprehensive information for operational decisions.

[0134] In the training and validation phases, the collected complete lifecycle datasets of multiple gas turbines of the same model were divided into training, validation, and test sets according to unit or time period. A sliding window method was used to construct training samples, using multi-parameter data from the past T days as input features and the time length from the current moment to the failure point as the label RUL. An adaptive moment estimation Adam optimizer was used for gradient descent training, and the loss value on the validation set was monitored after each training cycle to prevent overfitting. Key validation metrics included: root mean square error and mean absolute percentage error of RUL predictions, used to evaluate the accuracy of point predictions; and prediction interval coverage, i.e., whether the proportion of true RUL values ​​falling within the 90% prediction interval is close to 90%, used to evaluate the reliability of interval predictions. Through this training process, the model can output statistically significant prediction intervals while ensuring point prediction accuracy, providing a quantitative basis for predictive maintenance decisions.

[0135] Based on Embodiment 1 or 2, this embodiment can further optimize the system architecture and adopt an edge-cloud collaborative architecture.

[0136] Specifically, the method is based on an edge-cloud collaborative architecture, which includes the edge side deployed at the gas turbine site and the remote cloud.

[0137] The edge-side configuration performs steps 101 to 104, including data acquisition, comprehensive oil quality score calculation, contradiction detection, and inference diagnosis, and generates real-time alarms and maintenance suggestions. Because the edge-side is deployed at the equipment site, it enables millisecond-level local response and rapid closed-loop correction.

[0138] The cloud and edge communication are connected. The cloud is configured to receive data synchronized from the edge, perform cross-group big data aggregation and analysis, train and optimize the global prediction model, and then send the updated model parameters to the edge. The cloud uses its powerful computing capabilities to train and optimize the deep learning model, and then sends the updated model parameters to each edge node to achieve continuous model evolution.

[0139] As an example, the edge devices synchronize raw data and diagnostic results to the cloud data center via an encrypted channel. The cloud performs cross-unit and cross-plant big data aggregation and analysis, trains and optimizes global predictive models (such as the Long Short-Term Memory (LSTM) network model), and then distributes the updated model parameters to each edge node. The cloud also implements blockchain-based notarization of oil quality data to ensure data immutability.

[0140] This architecture resolves the conflict between real-time requirements and computational complexity in industrial settings, enabling the maximization of data value and the system's evolvability.

[0141] In some embodiments of this disclosure, the method may further include intelligent diagnosis and maintenance strategy recommendation functions. Specifically, when the overall oil quality score decreases by more than a set range, or when the deep learning model predicts that the remaining service life is lower than a preset threshold, the system automatically triggers a root cause analysis process. This process, based on the oil deterioration mode determined in step 103, further analyzes the characteristic combinations of each monitoring indicator deviating from the normal mode.

[0142] For example, a significant increase in acid value accompanied by a change in dielectric constant suggests oxidation as the degradation mode; increased moisture content accompanied by rising particulate contamination suggests mixed contamination caused by a cooler leak. Based on the matching results of feature combinations, the system outputs precise diagnostic conclusions and actionable maintenance recommendations, such as: "Main degradation mode: oil oxidation; Recommendation: Check system sealing; assess antioxidant additive consumption." This function elevates the system's capabilities from "condition monitoring" to "intelligent diagnosis and decision support," providing operation and maintenance personnel with direct engineering insights.

[0143] It should be noted that the core method proposed in this disclosure, "multi-parameter fusion + dynamic weighting + physical model closed-loop correction + data-driven prediction," has broad applicability. This framework is not only applicable to the lubricating and control oil systems of gas turbines, but its methodology can also be transferred to the oil systems of other key rotating equipment such as steam turbines, water turbines, compressors, and large gearboxes. In practical applications, a configurable general process can be designed: sensor types can be selected according to the specific equipment monitoring requirements; the benchmark value in the scoring formula can be adjusted according to different oil specifications; and the initial weights of the entropy weighting method can be preset based on historical data of the equipment type. However, the core data processing logic, dynamic weight calculation, contradiction detection mechanism, physicochemical failure model reasoning, and deep learning prediction framework remain unchanged. Through this configurable design, the technical solution of this disclosure can be quickly adapted to different industrial scenarios, forming a general platform for oil health management covering various rotating machinery.

[0144] The online oil quality analysis method for gas turbine oil systems proposed in this disclosure provides a multi-dimensional data foundation for subsequent evaluation by acquiring oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine. A comprehensive oil quality score is determined based on the oil monitoring data, integrating multiple isolated parameters into a single quantitative indicator to intuitively reflect the oil health status. A pre-set contradiction detection rule library is used to detect contradictions between the comprehensive oil quality score and equipment performance parameters, enabling timely detection of inconsistencies between oil condition and equipment operation, avoiding misjudgments from a single data source. When a contradiction is detected, a pre-set physicochemical failure model library is used to deduce and diagnose the root cause of the contradiction, identifying the oil degradation mode leading to the contradiction. Based on the oil degradation mode, the parameters of the corresponding physicochemical failure model are corrected, and the comprehensive oil quality score is regenerated using the corrected physicochemical failure model. This constructs a closed-loop self-correction mechanism from data to model to physical facts, enabling the monitoring system to have self-learning and adaptive capabilities. Through the above technical solution, this disclosure significantly improves the reliability of gas turbine oil system operation.

[0145] Figure 2 This is a block diagram illustrating an online oil quality analysis device for a gas turbine oil system according to an exemplary embodiment. (Refer to...) Figure 2 The device includes a scoring unit 201, a detection unit 202, a diagnostic unit 203, and a generation unit 204.

[0146] Among them, the scoring unit 201 is used to acquire oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and to determine the comprehensive oil quality score based on the oil monitoring data.

[0147] The detection unit 202 is used to perform conflict detection on the comprehensive oil quality score and equipment performance parameters using a preset conflict detection rule library;

[0148] The diagnostic unit 203 is used to deduce and diagnose the root cause of the contradiction by using a preset physical and chemical failure model library when the contradiction detection result is that there is a contradiction, and to determine the oil deterioration mode that leads to the contradiction.

[0149] The generation unit 204 is used to modify the parameters of the corresponding physicochemical failure model according to the oil deterioration mode, and to regenerate the comprehensive oil quality score using the modified physicochemical failure model.

[0150] In some embodiments of this disclosure, the oil monitoring data includes at least two of the following: viscosity index, acid value, base value, moisture content, particulate contamination level, dielectric constant, varnish index, and oil temperature.

[0151] In some embodiments of this disclosure, the scoring unit 201 may specifically be used for:

[0152] The overall oil quality score M is calculated using the following formula:

[0153]

[0154] Where VI is the viscosity index, TAN is the acid value, TBN is the base value, a is the water content, d is the operating time since the last oil quality monitoring passed, and c is the dielectric constant. t represents the paint film index, t represents the oil temperature, and w1 to w8 represent the weighting coefficients of each index.

[0155] In some embodiments of this disclosure, the scoring unit 201 may specifically be used for:

[0156] The following steps are used to determine w1 to w8:

[0157] Acquire historical oil monitoring data of the gas turbine oil system within a preset time window, and analyze VI, TAN, TBN, a, d, c, Each monitoring indicator in t undergoes data standardization processing;

[0158] Based on the standardized data, the proportion of the sample value at each time point under each monitoring indicator to the total value of all samples under that indicator is calculated, which is used as the proportion of each time point under the monitoring indicator.

[0159] The information entropy of each monitoring indicator is calculated based on its proportion; information entropy is used to characterize the degree of dispersion of the indicator data.

[0160] The difference coefficient for each monitoring indicator is determined based on information entropy; the difference coefficient is the difference between 1 and the information entropy.

[0161] The difference coefficients of each monitoring indicator are normalized to obtain the dynamic weights of each monitoring indicator, and the dynamic weights of each monitoring indicator are used as weight coefficients w1 to w8.

[0162] In some embodiments of this disclosure, the apparatus may further include a smoothing processing unit for:

[0163] The following formula is used to smooth the oil monitoring data:

[0164] z t = ·x t +(1- )·z t-1

[0165] Where, x t The original monitoring value at the current moment, z t z is the smoothed value at the current time. t-1 This is the smoothed value from the previous time step. This is a smoothing factor.

[0166] In some embodiments of this disclosure, the contradiction detection rule base includes at least one of the following rules:

[0167] If the overall oil quality score is higher than the first threshold, and the servo valve response time exceeds the set percentage of the historical baseline value, then a contradiction is determined; and / or

[0168] If the overall oil quality score decreases by less than a set threshold, and the bearing vibration amplitude increases by more than a set threshold within a set time, and a characteristic frequency component corresponding to a preset fault type is detected in the vibration signal, then a contradiction is determined; and / or

[0169] If the overall oil quality score remains stable for a preset period of time, and the bearing temperature continues to rise above the set value, then a contradiction is identified.

[0170] In some embodiments of this disclosure, the physicochemical failure model library includes oxidation failure models, moisture intrusion models, and particulate contamination / wear models;

[0171] Among them, the oxidation failure model is used to describe the relationship between oil temperature, running time and acid value, and paint film index; the moisture intrusion model is used to describe the relationship between moisture content and dielectric constant, and oil film strength; and the particle contamination / wear model is used to describe the relationship between particle count and vibration characteristics, and wear degree.

[0172] In some embodiments of this disclosure, the diagnostic unit 203 may specifically be used for:

[0173] In response to the result of the contradiction detection, if a contradiction exists, the corresponding oil monitoring data and equipment performance parameters are obtained to obtain the data to be diagnosed.

[0174] Extract abnormal device performance characteristics from the data to be diagnosed;

[0175] The oil monitoring data and abnormal equipment performance characteristics in the data to be diagnosed are input into multiple candidate failure models for simulation and deduction, and the simulation and deduction results of each candidate failure model are obtained.

[0176] Calculate the matching degree between each simulation result and the corresponding oil monitoring data in the data to be diagnosed;

[0177] The oil deterioration mode corresponding to the candidate failure model with the highest matching degree is output as the diagnostic result; the diagnostic result includes the confidence probability corresponding to the candidate failure model with the highest matching degree.

[0178] In some embodiments of this disclosure, parameter correction of the corresponding physicochemical failure model includes at least one of the following:

[0179] Adjust the kinetic parameters of the physicochemical failure model based on the diagnostic results;

[0180] Based on the diagnostic results, data compensation or suspicious labeling is performed on the sensors involved in the resulting contradictions.

[0181] Based on the diagnostic results, the weight of the corresponding monitoring indicators in the comprehensive oil quality score will be increased within a preset time period.

[0182] In some embodiments of this disclosure, the apparatus may further include a prediction unit for predicting the remaining service life of the oil based on the time series of historical oil monitoring data using a pre-trained deep learning model, and outputting the predicted value of the remaining service life and its prediction range; the deep learning model is a Long Short-Term Memory (LSTM) network model or a Transformer model.

[0183] In some embodiments of this disclosure, the device is implemented based on an edge-cloud collaborative architecture, which includes an edge side deployed at the gas turbine site and a remote cloud.

[0184] The edge-side configuration is configured to perform the following steps: acquiring oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine; determining the comprehensive oil quality score based on the oil monitoring data; using a preset contradiction detection rule library to detect contradictions between the comprehensive oil quality score and equipment performance parameters; and when the contradiction detection result indicates the existence of contradictions, using a preset physicochemical failure model library to deduce and diagnose the root cause of the contradictions, and generating real-time alarms and maintenance suggestions.

[0185] The cloud and edge communication are connected. The cloud is configured to receive data synchronized from the edge, perform cross-unit big data aggregation analysis, train and optimize the global prediction model, and send the updated model parameters to the edge.

[0186] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0187] The online oil quality analysis device for a gas turbine oil system proposed in this disclosure acquires oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, providing a multi-dimensional data foundation for subsequent evaluation. Based on the oil monitoring data, a comprehensive oil quality score is determined, integrating multiple isolated parameters into a single quantitative indicator to intuitively reflect the oil health status. A pre-set contradiction detection rule library is used to detect contradictions between the comprehensive oil quality score and equipment performance parameters, enabling timely detection of inconsistencies between oil condition and equipment operation, avoiding misjudgments from a single data source. When a contradiction is detected, a pre-set physicochemical failure model library is used to deduce and diagnose the root cause of the contradiction, identifying the oil degradation mode leading to the contradiction. Based on the oil degradation mode, the parameters of the corresponding physicochemical failure model are corrected, and the comprehensive oil quality score is regenerated using the corrected physicochemical failure model. This constructs a closed-loop self-correction mechanism from data to model to physical facts, enabling the monitoring system to have self-learning and adaptive capabilities. Through the above technical solution, this disclosure significantly improves the reliability of gas turbine oil system operation.

[0188] Figure 3 This is a block diagram illustrating an apparatus for an online oil quality analysis method for a gas turbine oil system, according to an exemplary embodiment. For example, apparatus 300 may be an electronic device, such as a mobile phone, computer, digital broadcasting terminal, messaging device, tablet device, personal digital assistant, etc.

[0189] Reference Figure 3 The device 300 may include one or more of the following components: processing component 302, memory 304, power component 306, multimedia component 308, audio component 310, I / O interface 312, sensor component 314, and communication component 316.

[0190] Processing component 302 typically controls the overall operation of device 300, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 302 may include one or more processors 320 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 302 may include one or more modules to facilitate interaction between processing component 302 and other components. For example, processing component 302 may include a multimedia module to facilitate interaction between multimedia component 308 and processing component 302.

[0191] Memory 304 is configured to store various types of data to support the operation of device 300. Examples of such data include instructions for any application or method operating on device 300, contact data, phonebook data, messages, pictures, videos, etc. Memory 304 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0192] The power supply component 306 provides power to the various components of the device 300. The power supply component 306 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device 300.

[0193] Multimedia component 308 includes a screen that provides an output interface between the device 300 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 308 includes a front-facing camera and / or a rear-facing camera. When the device 300 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0194] Audio component 310 is configured to output and / or input audio signals. For example, audio component 310 includes a microphone (MIC) configured to receive external audio signals when device 300 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 304 or transmitted via communication component 316. In some embodiments, audio component 310 also includes a speaker for outputting audio signals.

[0195] I / O interface 312 provides an interface between processing component 302 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, start buttons, and lock buttons.

[0196] Sensor assembly 314 includes one or more sensors for providing status assessments of various aspects of device 300. For example, sensor assembly 314 may detect the on / off state of device 300, the relative positioning of components such as the display and keypad of device 300, changes in the position of device 300 or a component of device 300, the presence or absence of user contact with device 300, the orientation or acceleration / deceleration of device 300, and temperature changes of device 300. Sensor assembly 314 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 314 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 314 may also include an accelerometer, a gyroscope, a magnetometer, a pressure sensor, or a temperature sensor.

[0197] Communication component 316 is configured to facilitate wired or wireless communication between device 300 and other devices. Device 300 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 316 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 316 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0198] In an exemplary embodiment, the apparatus 300 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0199] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 304 including instructions, which can be executed by a processor 320 of the device 300 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0200] In an exemplary embodiment, a computer program product is also provided, including a computer program that implements the above-described method when executed by the processor 320 of the device 300.

[0201] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the appended claims.

[0202] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An on-line method of oil quality analysis of a gas turbine oil system, characterized by, include: Acquire oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and determine a comprehensive oil quality score based on the oil monitoring data; The comprehensive oil quality score and the equipment performance parameters are checked for discrepancies using a pre-defined discrepancy detection rule base. When the result of the contradiction detection is that there is a contradiction, the root cause of the contradiction is inferred and diagnosed by using a preset physical and chemical failure model library to determine the oil deterioration mode that leads to the contradiction. The physical and chemical failure model library includes oxidation failure model, water intrusion model and particulate contamination / wear model. Based on the oil degradation mode, the parameters of the corresponding physicochemical failure model are corrected, and the comprehensive oil quality score is regenerated using the corrected physicochemical failure model. The parameter correction of the corresponding physicochemical failure model includes at least one of the following: The kinetic parameters of the physicochemical failure model are adjusted based on the diagnostic results. Based on the diagnostic results, data compensation or suspicious labeling is performed on the sensors involved in the resulting contradictions. Based on the diagnostic results, increase the weight of the corresponding monitoring indicators in the comprehensive oil quality score within a preset time period; The contradiction detection rule base includes at least one of the following rules: If the overall oil quality score is higher than the first threshold, and the servo valve response time exceeds the set percentage of the historical baseline value, then a contradiction is determined; and / or If the overall oil quality score decreases by less than a set threshold, and the bearing vibration amplitude increases by more than a set threshold within a set time, and a characteristic frequency component corresponding to a preset fault type is detected in the vibration signal, then a contradiction is determined; and / or If the overall oil quality score remains stable for a preset period of time, and the bearing temperature continues to rise above the set value, then a contradiction is determined. When the result of the contradiction detection indicates the existence of a contradiction, a preset physicochemical failure model library is used to deduce and diagnose the root cause of the contradiction, and to determine the oil degradation mode leading to the contradiction, including: In response to the result of the contradiction detection indicating the existence of a contradiction, the oil monitoring data and equipment performance parameters corresponding to the contradiction are obtained to obtain the data to be diagnosed; Extract abnormal device performance characteristics from the data to be diagnosed; The oil monitoring data and the abnormal performance characteristics of the equipment in the data to be diagnosed are input into multiple candidate failure models for simulation and deduction, and the simulation and deduction results of each candidate failure model are obtained. Calculate the matching degree between each simulation result and the corresponding oil monitoring data in the data to be diagnosed; The oil deterioration mode corresponding to the candidate failure model with the highest matching degree is output as the diagnostic result; the diagnostic result includes the confidence probability corresponding to the candidate failure model with the highest matching degree. The comprehensive oil quality score is determined based on the oil monitoring data, including: The overall oil quality score M is calculated using the following formula: Wherein VI is a viscosity index, w1 is a weight coefficient of VI, TAN is an acid value, w2 is a weight coefficient of TAN, TBN is a base value, w3 is a weight coefficient of TBN, a is a moisture content, w4 is a weight coefficient of a, d is a running time after the last oil quality monitoring qualified, w5 is a weight coefficient of d, c is a dielectric constant, w6 is a weight coefficient of c, is a paint film index, w7 is a weight coefficient of the paint film index, t is an oil temperature, and w8 is a weight coefficient of t.

2. The method of claim 1, wherein, The oil monitoring data includes at least two of the following: viscosity index, acid value, alkalinity, moisture content, particulate contamination, dielectric constant, varnish index, and oil temperature.

3. The method of claim 1, wherein, The weighting coefficients w1 to w8 are determined using the following steps: obtaining historical oil monitoring data of the gas turbine oil system in a preset time window, performing data standardization processing on each of VI, TAN, TBN, a, d, c, , and t monitoring indexes. Based on the standardized data, the proportion of the sample value at each time point under each monitoring indicator to the total value of all samples under that indicator is calculated, which is used as the proportion of each time point under the monitoring indicator. The information entropy of each monitoring indicator is calculated based on its proportion; the information entropy is used to characterize the dispersion of the indicator data. The difference coefficient for each monitoring indicator is determined based on the information entropy; the difference coefficient is the difference between 1 and the information entropy. The difference coefficients of each monitoring indicator are normalized to obtain the dynamic weights of each monitoring indicator, and the dynamic weights of each monitoring indicator are used as the weight coefficients w1 to w8.

4. The method of claim 1, wherein, Before determining the overall oil quality score based on the oil monitoring data, the following steps are also included: The oil monitoring data is smoothed using the following formula: z t = ·x t +(1- )·z t-1 wherein x t is the original monitoring value at the current time, z t is the smoothed value at the current time, z t-1 is the smoothed value at the previous time, and is a smoothing factor.

5. The method of claim 1, wherein, The physicochemical failure model library includes oxidation failure models, moisture intrusion models, and particulate contamination / wear models. The oxidation failure model describes the relationship between oil temperature, operating time, acid value, and paint film index; the moisture intrusion model describes the relationship between moisture content, dielectric constant, and oil film strength; and the particulate contamination / wear model describes the relationship between particle count, vibration characteristics, and wear degree.

6. The method according to claim 1, characterized in that, Also includes: Based on the time series of historical oil monitoring data, a pre-trained deep learning model is used to predict the remaining service life of the oil, and the predicted value of the remaining service life and its prediction range are output; the deep learning model is a Long Short-Term Memory Network (LSTM) model or a Transformer model.

7. The method according to claim 1, characterized in that, The method is based on an edge-cloud collaborative architecture, which includes an edge deployed at the gas turbine site and a remote cloud. The edge side is configured to perform the following steps: acquiring oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine; determining a comprehensive oil quality score based on the oil monitoring data; using a preset contradiction detection rule library to perform contradiction detection between the comprehensive oil quality score and the equipment performance parameters; and when the contradiction detection result indicates a contradiction, using a preset physicochemical failure model library to deduce and diagnose the root cause of the contradiction, and generating real-time alarms and maintenance suggestions. The cloud is connected to the edge side. The cloud is configured to receive data synchronized from the edge side, perform cross-group big data aggregation analysis, train and optimize the global prediction model, and send the updated model parameters to the edge side.

8. An online oil quality analysis device for a gas turbine oil system, characterized in that, include: The scoring unit is used to acquire oil monitoring data of the gas turbine oil system and equipment performance parameters of the gas turbine, and to determine a comprehensive oil quality score based on the oil monitoring data. The comprehensive oil quality score M is calculated according to the following formula: Wherein, VI is the viscosity index, w1 is the weighting coefficient of VI, TAN is the acid value, w2 is the weighting coefficient of TAN, TBN is the base value, w3 is the weighting coefficient of TBN, a is the moisture content, w4 is the weighting coefficient of a, d is the operating time since the last oil quality monitoring passed, w5 is the weighting coefficient of d, c is the dielectric constant, and w6 is the weighting coefficient of c. The coating film index, W7 is... The weighting coefficients are t, where t is the oil temperature and w8 is the weighting coefficient of t. The detection unit is used to perform conflict detection on the comprehensive oil quality score and the equipment performance parameters using a preset conflict detection rule base, wherein the conflict detection rule base includes at least one of the following rules: If the overall oil quality score is higher than the first threshold, and the servo valve response time exceeds the set percentage of the historical baseline value, then a contradiction is determined; and / or If the overall oil quality score decreases by less than a set threshold, and the bearing vibration amplitude increases by more than a set threshold within a set time, and a characteristic frequency component corresponding to a preset fault type is detected in the vibration signal, then a contradiction is determined; and / or If the overall oil quality score remains stable for a preset period of time, and the bearing temperature continues to rise above the set value, then a contradiction is determined. The diagnostic unit is used to deduce and diagnose the root cause of the contradiction by using a preset physical and chemical failure model library when the result of the contradiction detection is contradictory, and to determine the oil deterioration mode that leads to the contradiction. The physical and chemical failure model library includes oxidation failure model, water intrusion model and particulate contamination / wear model. The generation unit is used to modify the parameters of the corresponding physicochemical failure model according to the oil deterioration mode, and regenerate the comprehensive oil quality score using the modified physicochemical failure model. The diagnostic unit is used to obtain oil monitoring data and equipment performance parameters corresponding to the contradiction in response to the result of the contradiction detection being a contradiction, and to obtain the data to be diagnosed. Extract abnormal device performance characteristics from the data to be diagnosed; The oil monitoring data and the abnormal performance characteristics of the equipment in the data to be diagnosed are input into multiple candidate failure models for simulation and deduction, and the simulation and deduction results of each candidate failure model are obtained. Calculate the matching degree between each simulation result and the corresponding oil monitoring data in the data to be diagnosed; The oil deterioration mode corresponding to the candidate failure model with the highest matching degree is output as the diagnostic result; the diagnostic result includes the confidence probability corresponding to the candidate failure model with the highest matching degree. The generation unit is used to modify the parameters of the corresponding physicochemical failure model, including at least one of the following: The kinetic parameters of the physicochemical failure model are adjusted based on the diagnostic results. Based on the diagnostic results, data compensation or suspicious labeling is performed on the sensors involved in the resulting contradictions. Based on the diagnostic results, the weight of the corresponding monitoring indicators in the comprehensive oil quality score will be increased within a preset time period.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images