Oil liquid quality detection method and system

Abstract

The invention discloses an oil quality detection method and system, and relates to the technical field of industrial equipment state monitoring. The method comprises the following steps: regularly collecting operation condition and environment data in the operation process of the hydraulic system, performing oil quality abnormity prediction, and generating a predicted oil abnormity probability; when the predicted oil abnormal probability exceeds a preset probability threshold value, oil sampling is triggered to obtain an oil sample; calling a historical predicted oil abnormal probability sequence in the sliding window to analyze an oil quality deterioration trend, and obtaining an oil quality deterioration rate; and generating an adaptive oil quality detection scheme by combining the current detection time interval and the oil quality deterioration rate, detecting the oil sample according to the adaptive oil quality detection scheme, and outputting a result. According to the method, the conversion from fixed period maintenance to state prediction maintenance is realized, the problem that excessive detection and insufficient detection coexist in a traditional method is effectively solved, and the accuracy, timeliness and resource utilization efficiency of oil quality monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment condition monitoring technology, specifically to a method and system for detecting oil quality. Background Technology

[0002] During the long-term operation of hydraulic systems and various industrial equipment, the performance of lubricating oil (hydraulic oil) directly determines transmission accuracy, component wear rate, and overall system reliability. To ensure safe and stable equipment production, continuous monitoring of lubricating oil quality and timely replacement are crucial.

[0003] However, in current industrial practice, oil quality monitoring mainly relies on fixed-cycle offline testing or simple sensor alarms. While these methods can reflect the oil condition to some extent, their response mechanisms are static and passive, failing to deeply correlate testing with real-time equipment operating load and the dynamic deterioration process of the oil. Specifically, existing technologies employ a uniform testing cycle and fixed indicator combination, implementing a "one-size-fits-all" approach to testing across all equipment and operating stages. This leads to two drawbacks: firstly, it easily results in over-testing and resource waste for equipment operating smoothly; secondly, it easily leads to under-testing and delayed warnings for equipment operating under harsh conditions or with accelerated deterioration, increasing the risk of failure. Furthermore, the fixed model cannot adapt to the differentiated deterioration rates under different equipment and conditions, causing a large number of tests to be disconnected from actual risks, making it difficult to optimize the overall maintenance economy and safety. Summary of the Invention

[0004] This invention addresses the technical problems of existing fixed-cycle testing methods, such as high testing blindness, unreasonable resource allocation, and inability to adapt to dynamic operating conditions and differentiated degradation rates, by providing an oil quality testing method and system.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for detecting oil quality, comprising: During the operation of the hydraulic system, operating condition data and operating environment data are collected periodically to predict abnormal lubricating oil quality and generate a predicted probability of abnormal oil quality. If the predicted probability of oil abnormality is greater than the preset probability threshold, then oil sampling is performed to obtain oil samples. The historical predicted oil quality anomaly probability sequence within the sliding window is used to analyze the trend of oil quality deterioration and obtain the rate of oil quality deterioration. Based on the current detection time interval and the oil quality deterioration rate, an appropriate oil quality detection scheme is generated, the oil sample is subjected to quality detection, and the oil quality detection results are output.

[0006] Secondly, the present invention provides an oil quality detection system, comprising: The data acquisition and prediction module is used to periodically collect operating condition data and operating environment data during the operation of the hydraulic system to predict abnormal lubricating oil quality and generate a predicted probability of abnormal oil quality. An anomaly triggering and sampling module is used to perform oil sampling to obtain oil samples if the predicted oil anomaly probability is greater than a preset probability threshold. The trend analysis and rate calculation module is used to call the historical predicted oil anomaly probability sequence in the sliding window to perform oil quality deterioration trend analysis and obtain the oil quality deterioration rate. The adaptive detection and output module is used to generate an adapted oil quality detection scheme based on the current detection time interval and the oil quality deterioration rate, perform quality detection on the oil sample, and output the oil quality detection results.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention first predicts the probability of oil anomalies based on real-time operating conditions and environmental data, enabling on-demand triggering of detection actions and overcoming the blindness of fixed-period sampling. Secondly, by analyzing the trends of historical prediction sequences to obtain the deterioration rate, it can dynamically assess the urgency of oil condition degradation. Thirdly, by adaptively generating detection schemes based on the current operating time and deterioration rate, the combination and depth of detection indicators can accurately match the actual risk level. Finally, the above steps together form an intelligent closed loop, avoiding over-detection of stable states while ensuring focused attention and timely warnings of risks accelerating deterioration, thereby improving the timeliness, accuracy, and resource utilization efficiency of oil quality monitoring overall. Attached Figure Description

[0008] Figure 1 A schematic flowchart of an oil quality testing method provided by the present invention; Figure 2 This is a schematic diagram of the structure of an oil quality detection system provided by the present invention.

[0009] In the attached diagram, the components represented by each number are as follows: The module includes: data acquisition and prediction module 11, anomaly triggering and sampling module 12, trend analysis and rate calculation module 13, and adaptive detection and output module 14. Detailed Implementation

[0010] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for detecting oil quality, including: S10: During the operation of the hydraulic system, periodically collect operating condition data and operating environment data to predict abnormal lubricating oil quality and generate a predicted probability of abnormal oil quality. First, during the operation of the hydraulic system, operating condition data and environmental data are collected periodically. A hydraulic system refers to a complete set of devices that uses hydraulic oil as the working medium and completes specific mechanical actions through power components, control components, and actuators. Examples include the hydraulic system of an excavator in construction machinery, the hydraulic feed system in machine tools, or the hydraulic pressing system in the metallurgical industry. During the operation of this hydraulic system, operating condition data and environmental data are collected periodically; these data are a series of parameters reflecting the system's working status and external conditions.

[0011] Since the deterioration process of hydraulic oil is closely related to the actual working load of the system and its environment, drastic changes in operating conditions or deterioration of environmental conditions will accelerate oil oxidation, contamination, and additive loss. Therefore, based on the operating condition data and operating environment data, oil quality anomalies can be predicted, generating a predicted probability of oil anomalies. The predicted probability of oil anomalies is a quantifiable value used to characterize the likelihood that an abnormal state of oil quality has occurred or is about to occur under given operating data conditions. Specifically, the principle of this step is to establish a correlation model between operating parameters and oil state, transforming the originally passive, periodic physical detection into an active, operating state-based risk assessment, thereby providing an intelligent decision-making basis for whether to initiate actual sampling and testing.

[0012] Specifically, operating condition data and operating environment data are collected periodically to predict lubricating oil quality anomalies, generating predicted probabilities of oil anomalies, including: During the operation of the hydraulic system, starting from the lubricating oil replacement node, the operating condition data and operating environment data of the hydraulic system are collected periodically according to the predetermined monitoring frequency to obtain the operating condition data sequence and the operating environment data sequence, and the number of monitoring nodes of the data is recorded. Activate the adaptive oil quality predictor according to the number of monitoring nodes, perform oil quality anomaly prediction based on the operating condition data sequence and operating environment data sequence, and generate the predicted oil anomaly probability. The operating condition data includes system working pressure and pressure pulsation data, hydraulic oil temperature data, electrical power parameters of the main pump and motor, operating frequency and load cycle data of the actuators, system leakage indication parameters, and vibration data of key mechanical components. The operating environment data includes ambient temperature and humidity data, ambient particulate matter concentration data, and temperature and humidity data of the breathing air from the hydraulic oil tank.

[0013] First, during the operation of the hydraulic system, the time point of the last lubricating oil change is used as the starting point for monitoring. Starting from this starting point, according to a pre-set monitoring frequency, such as once per hour or once per shift, the operating condition data and operating environment data of the hydraulic system are collected periodically to form a sequence of operating condition data and operating environment data over time.

[0014] The collected operating condition data comprehensively covers key system parameters that affect the state of the hydraulic fluid, including system working pressure and pressure pulsation data that characterizes pressure stability, hydraulic oil temperature data that directly affects the oxidation rate of the hydraulic fluid, electrical power parameters of the main pump and motor, operating frequency and load cycle data of the actuators that reflect the working intensity of the actuators, system leakage indication parameters that indicate the sealing condition of the system, and vibration data of key mechanical components that reflect the mechanical wear condition of the components.

[0015] Meanwhile, the collected operating environment data focuses on the potential impact of external conditions on the hydraulic fluid, specifically including environmental temperature and humidity data that may accelerate the intrusion or oxidation of water into the hydraulic fluid, environmental particulate matter concentration data that affects the degree of hydraulic fluid contamination, and temperature and humidity data of the air breathing into the hydraulic oil tank that is in direct contact with the hydraulic fluid in the tank.

[0016] Simultaneously, the number of data points successfully collected since the starting node is recorded; this number represents the number of monitoring nodes. This number of monitoring nodes characterizes the length of the data sequence, and its core function is to match an oil quality predictor with the corresponding input dimensions, ensuring that the data structure input to the model remains consistent with that used during training, thereby guaranteeing the accuracy and reliability of the prediction.

[0017] Specifically, based on the number of monitoring nodes currently recorded, an oil quality predictor is activated that is adapted to the number of nodes. This oil quality predictor is a pre-built analytical model optimized for different data sequence lengths.

[0018] Specifically, the construction steps of the adapted oil quality predictor include: With the number of monitoring nodes as a constraint, historical operating records of similar hydraulic systems are retrieved, sample operating condition data sequence sets and sample operating environment data sequence sets are collected, and the proportion of oil quality abnormality events of similar hydraulic systems in subsequent historical time zones for different sample operating condition data sequences and sample operating environment data sequences is counted and recorded as the sample oil abnormality probability, and the sample oil abnormality probability set is obtained. The sample operating condition data sequence set, sample operating environment data sequence set, and sample oil abnormality probability set are used as training data, and K-fold cross-partitioning with replacement is performed to obtain K sample training sets. Using the K sample training sets, long short-term memory networks are trained to convergence to generate K oil quality prediction units. These units are then integrated and constructed according to the mean fusion strategy to create an adapted oil quality predictor.

[0019] First, the current number of monitoring nodes, i.e., the required length of the input data sequence, is used as the core constraint. Based on this constraint, historical operating records of similar hydraulic systems are retrieved. For each historical operating record that meets the conditions, a sample operating condition data sequence and a sample operating environment data sequence with the same length as the constraint are collected, starting from a certain lubricating oil replacement node, thus forming a sample operating condition data sequence set and a sample operating environment data sequence set.

[0020] Subsequently, for each sample operating condition data sequence set and sample operating environment data sequence set corresponding to the same type of hydraulic system, the percentage of oil quality abnormality events occurring within a specific subsequent historical time zone after the data sequence was collected, such as the following 15 days, was statistically analyzed. This percentage was defined as the sample oil abnormality probability, and the probability values ​​of all sample operating condition data sequences and sample operating environment data sequences constituted the sample oil abnormality probability set.

[0021] The final sample operating condition data sequence set and sample operating environment data sequence set, along with their corresponding sample oil abnormality probability set, together constitute complete training sample data, establishing a quantitative correlation between specific operating state modes and subsequent short-term oil deterioration results.

[0022] Furthermore, the obtained sample operating condition data sequence set, sample operating environment data sequence set, and their corresponding sample oil anomaly probability set are used together as training data. This training data is processed using a K-fold cross-partitioning method with replacement. This method randomly divides the original training data into K mutually exclusive subsets, constructing K sample training sets through sampling with replacement. Each sample training set contains a subset of samples used for model training and a subset of samples used for validation. This K-fold cross-partitioning method helps to fully utilize limited data and evaluate model stability in subsequent model training.

[0023] Finally, using the obtained K sample training sets, K Long Short-Term Memory (LSTM) network models were independently trained. LSM is a special type of recurrent neural network, adept at handling and predicting long-term dependencies in time-series data, and suitable for analyzing operational data sequences reflecting continuous changes in oil condition. Each LSM network was trained until performance convergence, generating K oil quality prediction units with a certain degree of independence and differentiation in prediction. The convergence condition was set according to preset training objectives and stopping criteria; for example, the model was considered to have reached convergence when the model's loss function value on the validation set no longer decreased for 10 consecutive training cycles, or when the root mean square value of the prediction error was below 5‰.

[0024] During the deployment phase, the obtained K oil quality prediction units are integrated into a unified oil quality predictor using a mean fusion strategy. The mean fusion strategy is an ensemble learning model combination method where, for a new input sequence, each of the K units provides a prediction probability, and the final output is the arithmetic mean of all probabilities. This mean fusion strategy helps smooth the prediction errors of individual oil quality prediction units, reduces the impact of model overfitting or random fluctuations, and thus improves the robustness and accuracy of the overall prediction.

[0025] For example, each oil quality prediction unit mainly consists of an input layer, a long short-term memory (LSTM) network layer, and a probability output layer. The input layer receives standardized operating condition data sequences and operating environment data sequences, which together form a multi-channel temporal feature vector. The LSM network layer adopts a multi-layer stacked structure, with each layer containing a specific number of memory units. It retains and transmits long-term state information related to oil degradation through a gating mechanism. A fully connected layer is connected after the LSM network layer, and a sigmoid activation function is used to map the final abstracted temporal feature into a continuous value between 0 and 1, which serves as the predicted probability of oil anomalies.

[0026] During training, key hyperparameters included a learning rate of 0.0005, 100 training epochs, and a batch size of 32. The learning rate setting helped the model maintain stable convergence in learning complex temporal patterns, the number of training epochs ensured the model fully learned the evolution of oil states over time, and the batch size balanced training efficiency with the reliability of gradient updates.

[0027] Based on the collected sample operating condition data sequence set, sample operating environment data sequence set, and sample oil anomaly probability set, a K-fold cross-partitioning method with replacement is used to divide the overall data into K sample training sets. For each sample training set, its contained temporal feature vector is used as input, and the corresponding sample oil anomaly probability is used as the supervision signal. The network weight parameters are iteratively optimized through backpropagation algorithm combined with Adam optimizer. The binary cross-entropy loss function is used to measure the deviation between the predicted probability and the true anomaly probability label, and the training process is monitored through a reserved validation subset within each training set. When the validation set loss function value no longer decreases for 20 consecutive training epochs, or the average absolute error between the predicted probability and the true label is less than 0.02, the oil quality prediction unit is considered to have converged, and training is terminated. This process is repeated to generate K independent oil quality prediction units that have been trained to convergence.

[0028] Finally, the K oil quality prediction units are integrated into an adapted oil quality predictor using a mean fusion strategy. For a new input data sequence, all oil quality prediction units are invoked for parallel inference, and the arithmetic mean of their output probabilities is used as the final predicted oil quality anomaly probability. This integration strategy effectively combines the knowledge learned by different units on diverse data subsets, smoothing the prediction fluctuations of individual models and improving the robustness and accuracy of the overall prediction.

[0029] Further, the adaptive oil quality predictor is activated according to the number of monitoring nodes, and oil quality anomaly prediction is performed based on the operating condition data sequence and operating environment data sequence to generate a predicted oil anomaly probability, including: The product of the ratio of the preset standard number of monitoring nodes to the number of monitoring nodes and the adjustment coefficient L is used as the compensation coefficient for the first unit. The coefficients of variation for multiple data types in the operating condition data sequence and the operating environment data sequence are calculated separately, and then weighted and summed to obtain the coefficient of variation for the operating state. The ratio of the coefficient of variation of the operating state to the preset benchmark coefficient of variation of the operating state is set as the compensation coefficient for the second unit selection. The overall unit selection compensation coefficient is obtained by weighted calculation based on the compensation coefficient selected by the first unit and the compensation coefficient selected by the second unit. The product of the overall unit selection compensation coefficient and the preset standard unit selection quantity P is rounded down to obtain the number of adaptation units selected J, where P is the rounded down value of K / 3. If J is less than 1, J is equal to 1; if J is greater than K, J is equal to K. J oil quality prediction units are randomly selected from the K oil quality prediction units of the adapted oil quality predictor. Oil quality anomaly prediction is performed according to the operating condition data sequence and the operating environment data sequence, and the average of the J prediction results is used as the predicted oil anomaly probability.

[0030] First, the ratio of the preset standard number of monitoring nodes to the current actual number of monitoring nodes is obtained. This ratio is then multiplied by an adjustment coefficient L, and the product is used as the compensation coefficient for the first unit. The preset standard number of monitoring nodes is a pre-defined empirical or statistical benchmark value, representing the length of the most representative input sequence used by the model during the training phase, or the minimum amount of data required to ensure the prediction reaches the desired confidence level. This preset standard number of monitoring nodes is determined comprehensively based on the specific hydraulic system type, typical operating cycle, and historical data analysis results. For example, it might be set to 30 monitoring nodes, corresponding to the data accumulated from approximately 30 hours or 30 work shifts of continuous system operation.

[0031] The adjustment coefficient L serves to prevent the ratio from being too large or too small, which could lead to an unreasonable range of values ​​in the final calculation result. Its value can be dynamically adjusted according to the range of the ratio, for example, by setting it through a preset mapping table. Specifically, when the actual number of monitoring nodes is less than the standard number of monitoring nodes, i.e., the ratio is greater than 1, the adjustment coefficient L is set between 0 and 1, and the value of L gradually decreases as the ratio increases, in order to mitigate the impact of insufficient data. When the actual number of monitoring nodes is more than the standard number of monitoring nodes, i.e., the ratio is less than 1, the adjustment coefficient L is set greater than 1, and the value of L gradually decreases as the ratio increases, in order to adapt to scenarios with richer information.

[0032] Overall, the first compensation coefficient reflects the impact of the length of the monitoring data sequence on the sufficiency of the prediction information. The more monitoring nodes there are, the richer the accumulated historical information, and the fewer model units need to be called for prediction, thus saving computing resources and improving prediction efficiency. Conversely, if there are fewer monitoring nodes and the information is relatively insufficient, more model units need to be called for collaborative prediction to compensate for the possible biases of a single model and ensure the robustness of the prediction.

[0033] Secondly, the coefficients of variation (COPs) for each type of data in the operating condition data sequence and the operating environment data sequence are calculated separately. The COP is the ratio of the standard deviation to the mean, used to measure the dispersion of data in each dimension. The calculated COPs are then weighted and summed to obtain a comprehensive operating state COP. Specifically, the operating state COP equals the sum of the products of the COPs for each data type and their corresponding weight coefficients. The weight coefficients for each data type are set based on prior knowledge of the impact of that type of data on oil quality deterioration or statistical analysis results of historical data. For example, fluctuations in system operating pressure and oil temperature have a more significant impact on the oil oxidation rate, and their weight coefficients are usually set higher.

[0034] The coefficient of variation of the operating state is finally obtained to reflect the overall fluctuation level of the hydraulic system's operating state over a period of time. The greater the fluctuation, the more unstable the operation and the higher the risk of abnormal oil condition.

[0035] Furthermore, the calculated coefficient of variation of the operating state is divided by the preset baseline coefficient of variation of the operating state, and this ratio is set as the compensation coefficient for the second unit selection. The preset baseline coefficient of variation of the operating state is a statistical average based on long-term observation data of similar hydraulic systems under normal and stable operating conditions. This coefficient is calculated based on the degree of variation of data under stable operating conditions in a large number of historical operating records, and for example, a value of 0.15. This compensation coefficient for the second unit selection quantifies the degree of fluctuation of the current operating state relative to the baseline state. Greater fluctuations in the operating state indicate that the system may be in an unstable or abnormal operating condition, increasing the uncertainty of the prediction. Therefore, it is necessary to call more model units for comprehensive judgment to improve the reliability and robustness of the prediction results under complex and variable operating conditions.

[0036] Furthermore, the overall unit selection compensation coefficient is obtained by weighting the compensation coefficients selected in the first and second units. Overall unit selection compensation coefficient = First unit selection compensation coefficient × a + Second unit selection compensation coefficient × b. The weighting coefficients a and b are preset empirical values, set based on the relative contributions of data sufficiency and operational volatility to prediction uncertainty. For example, they are determined based on expert experience or through regression analysis of historical prediction errors. Weight a is set to 0.4, and weight b is set to 0.6.

[0037] The overall unit selection compensation coefficient integrates information from two dimensions: data sufficiency and operational volatility. Together, they determine the final number of model units to be called. The larger the overall unit selection compensation coefficient, the higher the uncertainty of the prediction. Correspondingly, more prediction units need to be called to integrate the collective calculation results and ensure the stability of the prediction results.

[0038] Finally, the overall unit selection compensation coefficient is multiplied by the preset standard unit selection quantity P, and the product is rounded down to obtain the actual number of adaptive units J to be selected in this prediction. Here, the standard unit selection quantity P is the total number of units K contained in the oil predictor divided by three and rounded down. To prevent the calculation results from exceeding a reasonable range, the following rule is set: if the calculated J is less than 1, then J is forced to be equal to 1; if J is greater than K, then J is forced to be equal to K.

[0039] Finally, J oil quality prediction units are randomly selected from the K oil quality prediction units included in the oil quality predictor. The current operating condition data sequence and the operating environment data sequence are respectively input into the J oil quality prediction units. Each oil quality prediction unit independently performs oil quality anomaly prediction and outputs a probability value. Finally, the arithmetic mean of the outputs of the J prediction units is used as the final predicted oil anomaly probability. Specifically, through this dynamic selection and mean fusion mechanism, the prediction scale is adaptively adjusted according to data conditions and operating status, optimizing the allocation efficiency of computing resources while ensuring prediction accuracy.

[0040] S20: If the predicted probability of oil abnormality is greater than the preset probability threshold, then oil sampling is performed to obtain oil samples. Once the predicted probability of hydraulic fluid anomaly is generated, it is compared with a pre-set probability threshold. This pre-set probability threshold is a critical action value determined based on a comprehensive analysis of hydraulic system reliability requirements, maintenance cost tolerance, and historical fault data; for example, it can be set to 70%. If the predicted probability of hydraulic fluid anomaly exceeds this pre-set probability threshold, it is determined that there is a significant risk of hydraulic fluid anomaly, reaching a level requiring intervention for physical verification and precise diagnosis. At this point, an automatic command will be generated or issued to trigger hydraulic fluid sampling.

[0041] Specifically, oil sampling refers to collecting a small, representative sample of in-use lubricating oil under specified operating conditions using a pre-set sampling valve or dedicated sampling port in the hydraulic system. The obtained oil sample serves as the basis for subsequent quality analysis and testing.

[0042] S30: Call the historical predicted oil anomaly probability sequence in the sliding window to analyze the oil quality deterioration trend and obtain the oil quality deterioration rate; Specifically, the historical predicted oil quality anomaly probability sequence within the sliding window is used to analyze the oil quality deterioration trend and obtain the oil quality deterioration rate, including: Set N consecutive monitoring time nodes as a sliding window, and call the historical predicted oil anomaly probability sequence within the sliding window, where N is an integer greater than or equal to 10; Calculate the probability difference between any two adjacent data in the historical predicted oil anomaly probability sequence according to the chronological order from first to last, and obtain the historical predicted oil anomaly probability difference sequence. The slope of the historical predicted oil anomaly probability difference sequence is calculated as the rate of oil quality deterioration.

[0043] First, N consecutive monitoring time points are used as a sliding window, where N is an integer greater than or equal to 10. For example, setting N to 15 represents the time period corresponding to the 15 most recent consecutive monitoring points. A sliding window refers to a fixed-length continuous data observation interval that moves along the time axis in time series analysis. All historical predicted oil anomaly probabilities stored within this sliding window are retrieved, forming a historical predicted oil anomaly probability sequence arranged chronologically. This historical predicted oil anomaly probability sequence records the dynamic trajectory of recent oil anomaly risks. Its core function is to filter out long-term baseline fluctuations or occasional instantaneous noise by analyzing probability change patterns within local time periods, thereby more accurately capturing the directional deterioration trend of oil quality.

[0044] Secondly, based on the time sequence, the difference between adjacent data points in the historical predicted oil anomaly probability sequence is calculated. Specifically, following the chronological order, the difference between the predicted oil anomaly probability value at each subsequent time point in the historical predicted oil anomaly probability sequence is calculated, resulting in a new sequence, namely the historical predicted oil anomaly probability difference sequence. This historical predicted oil anomaly probability difference sequence eliminates the absolute level of probability, highlighting the change in oil anomaly probability between adjacent monitoring points, i.e., the instantaneous increase or decrease in risk.

[0045] Finally, based on the historical predicted oil anomaly probability difference sequence obtained above, the slope of this historical predicted oil anomaly probability difference sequence over time is calculated using linear fitting methods such as the least squares method. This slope is the desired rate of oil quality deterioration. This rate of oil quality deterioration is a quantitative trend indicator; a positive value indicates that the oil anomaly probability shows an upward trend within the sliding window, meaning that the oil quality is deteriorating; its absolute value directly reflects the speed of the deterioration trend.

[0046] By acquiring the rate of oil quality deterioration, the urgency of oil degradation can be assessed from a dynamic perspective, rather than relying solely on the probability value of a single point. This provides crucial time-series trend information for the subsequent development of differentiated testing solutions.

[0047] S40: Generate an appropriate oil quality detection scheme based on the current detection time interval and the oil quality deterioration rate, perform quality detection on the oil sample, and output the oil quality detection results.

[0048] First, the time interval between the current detection time point and the lubricating oil replacement time point is calculated as the current detection time interval.

[0049] Specifically, first, a key timing parameter reflecting the cumulative operating time of the hydraulic system since the last maintenance is obtained, namely the current detection time interval. The current detection time node refers to the specific moment when step S20 is triggered and an oil sample is successfully obtained, and quality testing is about to begin. The lubricating oil replacement node is the specific time point when the hydraulic system last completed a complete lubricating oil replacement operation. The difference between the two time points is calculated, and the resulting time length is the current detection time interval.

[0050] From the perspective of equipment maintenance, lubricant degradation is a gradual process strongly correlated with operating time. Generally speaking, the longer the operating time, the more severe the oil's exposure to heat, oxidation, contamination, and additive depletion, thus increasing the basic risk of quality abnormalities. Therefore, the current testing interval serves as a fundamental risk dimension, reflecting the oil's service age over time and providing a fundamental time scale for comprehensively assessing its condition. Longer intervals typically indicate a higher level of background risk, requiring greater attention when developing testing plans.

[0051] Specifically, an adapted oil quality detection scheme is generated based on the current detection time interval and the oil quality deterioration rate, including: Configure the basic test index set and supplementary test index sequence for oil quality testing; The ratio of the current detection time interval to the average historical oil quality anomaly time span is used as the first anomaly risk coefficient. The ratio of the oil quality deterioration rate to the preset benchmark oil quality deterioration rate is used as the second abnormal risk coefficient, and the overall abnormal risk coefficient is calculated by combining it with the first abnormal risk coefficient. The product of the overall anomaly risk coefficient and the initial number of supplementary detection indicators M is rounded down to obtain the number of adaptive supplementary detection indicators Q. The first Q supplementary detection indicators of the supplementary detection indicator sequence are selected as the supplementary detection indicator set. The basic and supplementary detection index sets are used as the adapted oil quality detection scheme.

[0052] The basic test index set includes solid particle count, moisture content, kinematic viscosity, and acid value.

[0053] First, a basic set of testing indicators and a supplementary sequence of testing indicators are configured for oil quality testing. The basic set of testing indicators includes core, mandatory items for evaluating the basic health status of the oil, typically including particle count, water content, kinematic viscosity, and acid value. These basic indicators effectively reflect the degree of contamination, oxidative aging, and basic physicochemical properties of the oil. The supplementary sequence of testing indicators is a list arranged in descending order of importance and specificity, containing more in-depth or specialized analytical items.

[0054] Specifically, the methods for constructing supplementary detection index sequences include: Multiple supplementary detection indicators are configured, including Fourier transform infrared spectroscopy analysis, elemental spectroscopy analysis, pollution level, PQ index, air release value, demulsibility, foam characteristics, rotating oxygen bomb value, zinc content and phosphorus content; The additional detection indicators are sorted from highest to lowest degree of correlation with oil quality to generate a sequence of additional detection indicators.

[0055] First, several supplementary testing indicators are configured. These supplementary testing indicators are for deeper, more specialized, or more specific failure mode-targeting tests of lubricating oils, including Fourier transform infrared spectroscopy (FTIR), elemental spectroscopy, contamination level, PQ index, air release value, demulsibility, foaming characteristics, rotating bomb oxygen value, zinc content, and phosphorus content. Specifically, FTIR is used to identify chemical changes in the oil, such as oxidation, nitration, and additive degradation; elemental spectroscopy is used to detect the content of wear metals and contaminants; contamination level and PQ index are used to more precisely assess solid contamination; air release value, demulsibility, and foaming characteristics are used to evaluate the physical and performance properties of the oil; the rotating bomb oxygen value is used to measure the residual oxidation stability of the oil; and zinc and phosphorus content are used to monitor the consumption of key anti-wear additives.

[0056] Subsequently, all the added detection indicators were prioritized according to their correlation with the oil quality deterioration process and potential failure modes, from highest to lowest. The correlation degree could be determined based on domain expert experience, the contribution analysis of different detection indicators in historical failure cases, or mechanistic studies of the impact of each indicator on key oil performance. For example, items directly reflecting severe wear or additive depletion might be given higher priority. Through this prioritization, an ordered sequence of added detection indicators was ultimately generated.

[0057] The principle behind constructing this supplementary detection indicator sequence is to ensure that when additional detection items are needed, the detection indicators that best reveal the nature of the current risk and have the highest diagnostic value can be prioritized. The order of this supplementary detection indicator sequence is fixed. Based on the calculated number of suitable supplementary detection indicators Q, by simply selecting the first Q items from front to back, a set of supplementary detection indicators matching the risk level can be quickly and consistently generated, achieving specialization of the detection plan and automation of the generation process.

[0058] Furthermore, the calculated current detection time interval is divided by the average historical time span of similar hydraulic systems from oil change to the first occurrence of oil quality anomalies, and the resulting ratio is used as the first anomaly risk coefficient. This first anomaly risk coefficient quantifies the relative risk level based on the operating time dimension. A first anomaly risk coefficient value greater than 1 indicates that the current oil service time has exceeded the average safe period, and the risk has increased.

[0059] Simultaneously, the calculated rate of oil quality deterioration is divided by a preset benchmark rate of oil quality deterioration, and the resulting ratio is used as the second anomaly risk coefficient. The preset benchmark rate of oil quality deterioration is based on the typical average growth rate of the probability of oil anomalies over time in similar hydraulic systems under normal and stable operating conditions. This rate is calculated statistically based on the trend slope of stable deterioration phases from a large amount of historical operating data, and is, for example, set to 0.5% per day. This second anomaly risk coefficient quantifies the relative risk urgency based on the recent deterioration trend. A second anomaly risk coefficient value greater than 1 indicates that the deterioration rate exceeds the normal benchmark, the risk is accumulating rapidly, and closer monitoring and more in-depth detection are required.

[0060] Furthermore, by combining the first and second abnormal risk coefficients mentioned above, and through weighted summation or averaging, the overall abnormal risk coefficient is obtained. This overall abnormal risk coefficient comprehensively characterizes the overall risk level of the current oil quality.

[0061] Finally, the overall anomaly risk coefficient is multiplied by an initially set number of supplementary detection indicators M, and the product is rounded down to obtain the actual number of supplementary detection indicators Q required for this detection. Based on the order of the supplementary detection indicator sequence, the first Q indicators are selected to form the supplementary detection indicator set for this detection.

[0062] Finally, the mandatory basic testing indicator set and the supplementary testing indicator set determined dynamically based on risk are merged to form the appropriate oil quality testing plan implemented this time, matching the current risk level. In summary, this method realizes the transformation from fixed testing items to on-demand, tiered testing, ensuring in-depth and comprehensive diagnosis during high-risk periods and cost-effective basic screening during low-risk periods, thereby optimizing the allocation efficiency of testing resources.

[0063] In summary, the embodiments of this application have at least the following technical effects: This invention constructs a closed-loop proactive oil quality monitoring system by integrating intelligent prediction of operational data, dynamic analysis of deterioration trends, and adaptive generation of detection schemes. First, it predicts the probability of anomalies based on real-time operating conditions and environmental data, enabling a shift from fixed-period sampling to on-demand triggering, overcoming the blindness of timed detection. Second, by analyzing the trends of historical prediction sequences to obtain the deterioration rate, it can accurately quantify the urgency of oil condition deterioration, providing dynamic temporal evidence for risk assessment. Third, by combining the detection interval representing runtime with the deterioration rate reflecting the rate of change, it adaptively generates detection schemes matching the current risk level, allowing detection resources to be precisely focused on high-risk aspects, avoiding over-detection in low-risk areas and under-detection in high-risk areas.

[0064] Ultimately, the synergistic effect of these technologies improves the timeliness and accuracy of hydraulic condition warnings while ensuring the safe operation of the hydraulic system, and optimizes the allocation efficiency of maintenance resources, achieving a balance between economy and reliability.

[0065] Example 2, as Figure 2 As shown, based on the same inventive concept as the oil quality detection method provided in Embodiment 1, this embodiment of the invention also provides an oil quality detection system, including: The data acquisition and prediction module 11 is used to periodically collect operating condition data and operating environment data during the operation of the hydraulic system to predict the abnormality of lubricating oil quality and generate the predicted probability of oil abnormality. The anomaly triggering and sampling module 12 is used to perform oil sampling to obtain oil samples if the predicted oil anomaly probability is greater than a preset probability threshold. The trend analysis and rate calculation module 13 is used to call the historical predicted oil anomaly probability sequence in the sliding window to perform oil quality deterioration trend analysis and obtain the oil quality deterioration rate. The adaptive detection and output module 14 is used to generate an adapted oil quality detection scheme based on the current detection time interval and the rate of deterioration of the oil quality, perform quality detection on the oil sample, and output the oil quality detection results.

[0066] Specifically, the data acquisition and prediction module 11 is used for: Regularly collect operating condition data and operating environment data to predict lubricating oil quality anomalies and generate predicted probabilities of oil anomalies, including: During the operation of the hydraulic system, starting from the lubricating oil replacement node, the operating condition data and operating environment data of the hydraulic system are collected periodically according to the predetermined monitoring frequency to obtain the operating condition data sequence and the operating environment data sequence, and the number of monitoring nodes of the data is recorded. Activate the adaptive oil quality predictor according to the number of monitoring nodes, perform oil quality anomaly prediction based on the operating condition data sequence and operating environment data sequence, and generate the predicted oil anomaly probability. The operating condition data includes system working pressure and pressure pulsation data, hydraulic oil temperature data, electrical power parameters of the main pump and motor, operating frequency and load cycle data of the actuators, system leakage indication parameters, and vibration data of key mechanical components. The operating environment data includes ambient temperature and humidity data, ambient particulate matter concentration data, and temperature and humidity data of the breathing air from the hydraulic oil tank.

[0067] Specifically, the construction steps of the adapted oil quality predictor include: With the number of monitoring nodes as a constraint, historical operating records of similar hydraulic systems are retrieved, sample operating condition data sequence sets and sample operating environment data sequence sets are collected, and the proportion of oil quality abnormality events of similar hydraulic systems in subsequent historical time zones for different sample operating condition data sequences and sample operating environment data sequences is counted and recorded as the sample oil abnormality probability, and the sample oil abnormality probability set is obtained. The sample operating condition data sequence set, sample operating environment data sequence set, and sample oil abnormality probability set are used as training data, and K-fold cross-partitioning with replacement is performed to obtain K sample training sets. Using the K sample training sets, long short-term memory networks are trained to convergence to generate K oil quality prediction units. These units are then integrated and constructed according to the mean fusion strategy to create an adapted oil quality predictor.

[0068] Specifically, the appropriate oil quality predictor is activated according to the number of monitoring nodes, and oil quality anomaly prediction is performed based on the operating condition data sequence and the operating environment data sequence to generate a predicted oil anomaly probability, including: The product of the ratio of the preset standard number of monitoring nodes to the number of monitoring nodes and the adjustment coefficient L is used as the compensation coefficient for the first unit. The coefficients of variation for multiple data types in the operating condition data sequence and the operating environment data sequence are calculated separately, and then weighted and summed to obtain the coefficient of variation for the operating state. The ratio of the coefficient of variation of the operating state to the preset benchmark coefficient of variation of the operating state is set as the compensation coefficient for the second unit selection. The overall unit selection compensation coefficient is obtained by weighted calculation based on the compensation coefficient selected by the first unit and the compensation coefficient selected by the second unit. The product of the overall unit selection compensation coefficient and the preset standard unit selection quantity P is rounded down to obtain the number of adaptation units selected J, where P is the rounded down value of K / 3. If J is less than 1, J is equal to 1; if J is greater than K, J is equal to K. J oil quality prediction units are randomly selected from the K oil quality prediction units of the adapted oil quality predictor. Oil quality anomaly prediction is performed according to the operating condition data sequence and the operating environment data sequence, and the average of the J prediction results is used as the predicted oil anomaly probability.

[0069] The exception triggering and sampling module 12 is specifically used for: If the predicted probability of oil abnormality is greater than a preset probability threshold, then oil sampling is performed to obtain oil samples.

[0070] Specifically, the trend analysis and rate calculation module 13 is used for: The historical predicted oil quality anomaly probability sequence within the sliding window is used to analyze the oil quality deterioration trend and obtain the oil quality deterioration rate, including: Set N consecutive monitoring time nodes as a sliding window, and call the historical predicted oil anomaly probability sequence within the sliding window, where N is an integer greater than or equal to 10; Calculate the probability difference between any two adjacent data in the historical predicted oil anomaly probability sequence according to the chronological order from first to last, and obtain the historical predicted oil anomaly probability difference sequence. The slope of the historical predicted oil anomaly probability difference sequence is calculated as the rate of oil quality deterioration.

[0071] The adaptive detection and output module 14 is specifically used for: First, the time interval between the current detection time point and the lubricating oil replacement time point is calculated as the current detection time interval.

[0072] Specifically, an adapted oil quality detection scheme is generated based on the current detection time interval and the oil quality deterioration rate, including: Configure the basic test index set and supplementary test index sequence for oil quality testing; The ratio of the current detection time interval to the average historical oil quality anomaly time span is used as the first anomaly risk coefficient. The ratio of the oil quality deterioration rate to the preset benchmark oil quality deterioration rate is used as the second abnormal risk coefficient, and the overall abnormal risk coefficient is calculated by combining it with the first abnormal risk coefficient. The product of the overall anomaly risk coefficient and the initial number of supplementary detection indicators M is rounded down to obtain the number of adaptive supplementary detection indicators Q. The first Q supplementary detection indicators of the supplementary detection indicator sequence are selected as the supplementary detection indicator set. The basic and supplementary detection index sets are used as the adapted oil quality detection scheme.

[0073] The basic test index set includes solid particle count, moisture content, kinematic viscosity, and acid value.

[0074] Specifically, the methods for constructing supplementary detection index sequences include: Multiple supplementary detection indicators are configured, including Fourier transform infrared spectroscopy analysis, elemental spectroscopy analysis, pollution level, PQ index, air release value, demulsibility, foam characteristics, rotating oxygen bomb value, zinc content and phosphorus content; The additional detection indicators are sorted from highest to lowest degree of correlation with oil quality to generate a sequence of additional detection indicators.

Claims

1. A method for detecting the quality of oil, characterized in that, The methods include: During the operation of the hydraulic system, operating condition data and operating environment data are collected periodically to predict abnormal lubricating oil quality and generate a predicted probability of abnormal oil quality. If the predicted probability of oil abnormality is greater than the preset probability threshold, then oil sampling is performed to obtain oil samples. The historical predicted oil quality anomaly probability sequence within the sliding window is used to analyze the trend of oil quality deterioration and obtain the rate of oil quality deterioration. Based on the current detection time interval and the oil quality deterioration rate, an appropriate oil quality detection scheme is generated, the oil sample is subjected to quality detection, and the oil quality detection results are output.

2. The method for detecting oil quality according to claim 1, characterized in that, Regularly collect operating condition data and operating environment data to predict lubricating oil quality anomalies and generate predicted probabilities of oil anomalies, including: During the operation of the hydraulic system, starting from the lubricating oil replacement node, the operating condition data and operating environment data of the hydraulic system are collected periodically according to the predetermined monitoring frequency to obtain the operating condition data sequence and the operating environment data sequence, and the number of monitoring nodes of the data is recorded. Activate the adaptive oil quality predictor according to the number of monitoring nodes, perform oil quality anomaly prediction based on the operating condition data sequence and operating environment data sequence, and generate the predicted oil anomaly probability. The operating condition data includes system working pressure and pressure pulsation data, hydraulic oil temperature data, electrical power parameters of the main pump and motor, operating frequency and load cycle data of the actuators, system leakage indication parameters, and vibration data of key mechanical components. The operating environment data includes ambient temperature and humidity data, ambient particulate matter concentration data, and temperature and humidity data of the breathing air from the hydraulic oil tank.

3. The method for detecting oil quality according to claim 2, characterized in that, The construction steps of the adapted oil quality predictor include: With the number of monitoring nodes as a constraint, historical operating records of similar hydraulic systems are retrieved, sample operating condition data sequence sets and sample operating environment data sequence sets are collected, and the proportion of oil quality abnormality events of similar hydraulic systems in subsequent historical time zones for different sample operating condition data sequences and sample operating environment data sequences is counted and recorded as the sample oil abnormality probability, and the sample oil abnormality probability set is obtained. The sample operating condition data sequence set, sample operating environment data sequence set, and sample oil abnormality probability set are used as training data, and K-fold cross-partitioning with replacement is performed to obtain K sample training sets. Using the K sample training sets, long short-term memory networks are trained to convergence to generate K oil quality prediction units. These units are then integrated and constructed according to the mean fusion strategy to create an adapted oil quality predictor.

4. The method for detecting oil quality according to claim 3, characterized in that, The adaptive oil quality predictor is activated according to the number of monitoring nodes. Based on the operating condition data sequence and the operating environment data sequence, oil quality anomaly prediction is performed, generating a predicted probability of oil anomaly, including: The product of the ratio of the preset standard number of monitoring nodes to the number of monitoring nodes and the adjustment coefficient L is used as the compensation coefficient for the first unit. The coefficients of variation for multiple data types in the operating condition data sequence and the operating environment data sequence are calculated separately, and then weighted and summed to obtain the coefficient of variation for the operating state. The ratio of the coefficient of variation of the operating state to the preset benchmark coefficient of variation of the operating state is set as the compensation coefficient for the second unit selection. The overall unit selection compensation coefficient is obtained by weighted calculation based on the compensation coefficient selected by the first unit and the compensation coefficient selected by the second unit. The product of the overall unit selection compensation coefficient and the preset standard unit selection quantity P is rounded down to obtain the number of adaptation units selected J, where P is the rounded down value of K / 3. If J is less than 1, J is equal to 1; if J is greater than K, J is equal to K. J oil quality prediction units are randomly selected from the K oil quality prediction units of the adapted oil quality predictor. Oil quality anomaly prediction is performed according to the operating condition data sequence and the operating environment data sequence, and the average of the J prediction results is used as the predicted oil anomaly probability.

5. The method for detecting oil quality according to claim 1, characterized in that, The historical predicted oil quality anomaly probability sequence within the sliding window is used to analyze the oil quality deterioration trend and obtain the oil quality deterioration rate, including: Set N consecutive monitoring time nodes as a sliding window, and call the historical predicted oil anomaly probability sequence within the sliding window, where N is an integer greater than or equal to 10; Calculate the probability difference between any two adjacent data in the historical predicted oil anomaly probability sequence according to the chronological order from first to last, and obtain the historical predicted oil anomaly probability difference sequence. The slope of the historical predicted oil anomaly probability difference sequence is calculated as the rate of oil quality deterioration.

6. The method for detecting oil quality according to claim 1, characterized in that, The time interval between the current detection time point and the lubricating oil replacement time point is calculated as the current detection time interval.

7. The method for detecting oil quality according to claim 1, characterized in that, Based on the current detection time interval and the oil quality deterioration rate, a suitable oil quality detection scheme is generated, including: Configure the basic test index set and supplementary test index sequence for oil quality testing; The ratio of the current detection time interval to the average historical oil quality anomaly time span is used as the first anomaly risk coefficient. The ratio of the oil quality deterioration rate to the preset benchmark oil quality deterioration rate is used as the second abnormal risk coefficient, and the overall abnormal risk coefficient is calculated by combining it with the first abnormal risk coefficient. The product of the overall anomaly risk coefficient and the initial number of supplementary detection indicators M is rounded down to obtain the number of adaptive supplementary detection indicators Q. The first Q supplementary detection indicators of the supplementary detection indicator sequence are selected as the supplementary detection indicator set. The basic and supplementary detection index sets are used as the adapted oil quality detection scheme.

8. The method for detecting oil quality according to claim 7, characterized in that, The basic set of test indicators includes solid particle count, moisture content, kinematic viscosity, and acid value.

9. The method for detecting oil quality according to claim 7, characterized in that, Methods for constructing supplementary detection index sequences include: Multiple supplementary detection indicators are configured, including Fourier transform infrared spectroscopy analysis, elemental spectroscopy analysis, pollution level, PQ index, air release value, demulsibility, foam characteristics, rotating oxygen bomb value, zinc content and phosphorus content; The additional detection indicators are sorted from highest to lowest degree of correlation with oil quality to generate a sequence of additional detection indicators.

10. An oil quality detection system, characterized in that, A method for performing an oil quality testing method according to any one of claims 1-9, comprising: The data acquisition and prediction module is used to periodically collect operating condition data and operating environment data during the operation of the hydraulic system to predict abnormal lubricating oil quality and generate a predicted probability of abnormal oil quality. An anomaly triggering and sampling module is used to perform oil sampling to obtain oil samples if the predicted probability of oil anomaly is greater than a preset probability threshold. The trend analysis and rate calculation module is used to call the historical predicted oil anomaly probability sequence in the sliding window to perform oil quality deterioration trend analysis and obtain the oil quality deterioration rate. The adaptive detection and output module is used to generate an adapted oil quality detection scheme based on the current detection time interval and the oil quality deterioration rate, perform quality detection on the oil sample, and output the oil quality detection results.

Images