Intelligent detection and maintenance method for oil in large gearboxes

Abstract

This application relates to the field of oil detection technology and proposes an intelligent detection and maintenance method for oil in large gearboxes. The method includes: collecting oil parameters from the gearbox; constructing an oil parameter sequence, whereby the oil parameters include viscosity, dielectric constant, density, temperature, moisture content, and particle number; calculating the first independent component score, quantization error, and intrinsic deviation index at the time of data collection; calculating the particle number residual, binary eigenvalue assignment, and moisture wear synergy at the time of data collection; determining the dynamic contamination rate parameter based on the moisture wear synergy; and using the isolated forest algorithm to obtain the gearbox oil detection results based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density, and particle number at the time of data collection. This application can improve the accuracy of intelligent oil detection.

Description

Technical Field

[0001] This application relates to the field of oil detection technology, specifically to an intelligent detection and maintenance method applicable to oil in large gearboxes. Background Technology

[0002] The lubricating oil inside large gearboxes plays a crucial role in lubrication and friction reduction, heat dissipation, sealing and protection, shock absorption, and corrosion and rust prevention. The quality and deterioration of this oil directly impact the overall operational stability, service life, and safety of the industrial transmission system. Therefore, oil condition is a core monitoring target for equipment operation and maintenance. Currently, gearbox oil maintenance and monitoring technology integrates a multi-parameter sensor monitoring architecture online. Leveraging IoT data transmission technology and edge computing capabilities, it enables dynamic real-time monitoring, trend analysis, and predictive maintenance management of oil deterioration and component wear, significantly improving the intelligence level of gearbox operation and maintenance.

[0003] However, current intelligent gearbox oil detection models still have core technical flaws. Industrial gear oils of different grades and specifications, with different additive formulations, exhibit significant inherent differences in key parameters such as basic dielectric constant, viscosity index, and physicochemical properties. Most current mainstream detection models are trained and fitted based on historical monitoring data of a single, fixed grade of gear oil. These models have highly specific identification thresholds and judgment logic, making them extremely sensitive to fluctuations in the physicochemical properties of the oil itself. In actual industrial operation and maintenance, regular gearbox oil changes, replacement of different batches of oil, or mixing of oils under emergency conditions are commonplace. Existing detection models cannot distinguish between the inherent differences in oil formulation characteristics and the actual wear and tear or oil deterioration fault characteristics of the equipment. This easily leads to misjudging fluctuations in basic parameters caused by oil changes as abnormal equipment wear faults, resulting in frequent false alarms and missed critical faults. This seriously interferes with normal gearbox operation and maintenance scheduling, significantly reducing the operational reliability and practical application value of intelligent oil detection and predictive maintenance systems. Summary of the Invention

[0004] This application provides an intelligent detection and maintenance method for large gearbox oils to solve the problem that intelligent gearbox oil detection models are sensitive to oil changes and cannot achieve stable and accurate detection when changing oils or mixing different gearbox oils. The specific technical solution adopted is as follows: One embodiment of this application provides an intelligent detection and maintenance method for oil in large gearboxes, the method comprising the following steps: The oil parameters in the gearbox are collected, and a sequence of oil parameters at each collection time is constructed. The oil parameters include the viscosity, dielectric constant, density, temperature, moisture content, and number of particles of the oil. Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, calculate the first independent component score and quantization error at the acquisition time, and calculate the intrinsic deviation index at the acquisition time based on the first independent component score and quantization error. Based on the intrinsic deviation index, temperature, and particle number, the particle number residual at the time of sampling is calculated. Based on the particle number residual at the time of sampling and moisture, a number pair is established, and the binary eigenvalues ​​of the number pair are assigned. Combined with the particle number residual, the moisture wear synergy at the time of sampling is calculated. Based on the moisture wear synergy at the time of sampling, the dynamic contamination rate parameter at the time of sampling is determined. The dynamic contamination rate parameter at the time of sampling is used as the adaptive value of the contamination rate parameter. Using the isolated forest algorithm, the detection results of the gearbox oil are obtained based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density and particle number at the time of sampling.

[0005] Furthermore, the specific calculation method for the quantization error is as follows: Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, a multi-dimensional oil feature vector is constructed at the acquisition time. Using the self-organizing map algorithm, the Euclidean distance between the multi-dimensional oil features and the weight vector of the best matching neuron in the vector self-organizing map network is obtained, which is denoted as the quantization error at the acquisition time.

[0006] Furthermore, the specific calculation method for the intrinsic deviation index at the acquisition time is as follows: Calculate the second deviation of the acquisition time based on the quantization error at the acquisition time; The weighted sum of the first deviation and the second deviation at the acquisition time is denoted as the intrinsic deviation index at the acquisition time, where the weighted sum of the first deviation and the second deviation at the acquisition time is 1.

[0007] Furthermore, the second deviation at the acquisition time is the ratio of the quantization error at the acquisition time to the maximum value of the quantization errors at all acquisition times prior to the acquisition time.

[0008] Furthermore, the specific calculation method for the first deviation at the acquisition time is as follows: The first independent component score at the time of collection is divided by the sum of the oil quality benchmark and the preset minimum positive parameter, and the result is recorded as the first ratio at the time of collection. The absolute value of the difference between the first ratio at the time of collection and the number 1 is recorded as the first deviation at the time of collection.

[0009] Furthermore, the specific calculation method for the particle number residual at the acquisition time is as follows: An intrinsic deviation index sequence at the acquisition time is established. The intrinsic deviation index sequence and temperature sequence at the acquisition time are used as inputs and independent variables, and the particle number sequence at the acquisition time is used as inputs and dependent variables. The random forest regression algorithm is used to obtain the residuals corresponding to the particle number, which are denoted as the particle number residuals at the acquisition time.

[0010] Furthermore, the specific method for assigning binary eigenvalues ​​to the logarithmic pairs includes: The particle number residual at the sampling time is combined with the moisture content to form a data pair at the sampling time. All data pairs at the sampling time and within a sliding window of a preset time length before the sampling time are clustered to identify outliers in the data pairs. The minimum absolute value of the difference between the number of particles in the number of pairs and the residuals of particle number in all cluster centers is denoted as the particle number residual difference of the number of pairs. The minimum absolute value of the difference between the number of particles in the number of pairs and the residuals of moisture in all cluster centers is denoted as the moisture difference of the number of pairs. When the difference in particle number residuals of a pair is greater than twice the difference in moisture content, and the pair is an outlier, the binary eigenvalue of the pair is assigned a value of 1; otherwise, the binary eigenvalue of the pair is assigned a value of 0.

[0011] Furthermore, the specific calculation method for the moisture abrasion synergy at the time of collection is as follows: The Z-score normalized result of the particle number residual at the collection time is calculated and the maximum value of the number 0 is calculated. The product of the maximum value and the value of the binary eigenvalue of the number pair at the collection time is recorded as the moisture wear synergy at the collection time.

[0012] Furthermore, the specific method for determining the dynamic contamination rate parameter at the time of data collection is as follows: The sum of the moisture wear synergy at the time of collection and the number 1 is used as the denominator, and the moisture wear synergy at the time of collection is used as the numerator. The product of the fraction value and the preset adjustment coefficient is recorded as the characteristic product at the time of collection. The sum of the characteristic product at the time of collection and the number 1 is used as the first multiplier. The product of the first multiplier and the preset initial baseline contamination rate is recorded as the dynamic contamination rate parameter at the time of collection.

[0013] Furthermore, the method for obtaining the gearbox oil detection results using the isolated forest algorithm based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density, and particle number at the time of data collection includes the following specific methods: The intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density and particle number at the time of collection are arranged in order to establish a six-dimensional feature vector at the time of collection. The dynamic contamination rate parameter at the time of collection is used as the adaptive value of the contamination rate parameter. The isolated forest algorithm is used to process the six-dimensional feature vector at the time of collection to obtain the anomaly score at the time of collection. When the abnormal score exceeds the preset abnormal score threshold, the gearbox oil is maintained.

[0014] The beneficial effects of this application are: This application extracts the amplitudes of the most significant independent components from the viscosity, dielectric constant, and density sequences at the acquisition time, using them as projections of the inherent properties of the oil at the acquisition time under no wear interference. Combined with the overall characteristic deviation of the oil at the acquisition time relative to historical normal oil characteristics, an intrinsic deviation index is calculated. This intrinsic deviation index integrates characteristic shifts at the independent component level with pattern deviations in the overall feature space, and can be used to distinguish characteristic changes in the oil. Additional particle change after removing the influence of oil type is calculated to obtain the particle number residual at the acquisition time. Considering that the particle number residual may be affected by high moisture content, leading to spurious values ​​in the sensor data and making it difficult to reliably distinguish the failure modes of large gearboxes, the moisture wear synergy degree at the acquisition time is calculated by combining the particle number residual and moisture content. The greater the moisture wear synergy degree, the more likely the gearbox is to fail at the acquisition time. There is genuine abnormal wear. The lower the degree of synergy between moisture and wear, the more likely the anomaly at the time of sampling is due to changes in the oil's own properties or fluctuations in moisture content. Furthermore, the Isolation Forest algorithm is selected to perform unsupervised anomaly detection on abnormal oil samples. Considering that the performance of the Isolation Forest algorithm is highly dependent on a pre-set contamination rate parameter, a dynamic contamination rate parameter is determined based on the degree of synergy between moisture and wear at the time of sampling. This adapts to the dynamic evolution of oil parameters, avoiding the misclassification of normal parameter drift caused by changes in oil properties as anomalies caused by actual equipment wear. The Isolation Forest algorithm is then used to obtain the detection results for gearbox oil, addressing the problem that the intelligent gearbox oil detection model is sensitive to oil changes and cannot stably and accurately detect when changing oil or mixing different gearbox oils. This ensures normal operation and maintenance of the gearbox, improving the operational reliability and practical application value of intelligent oil detection and predictive maintenance. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic flowchart of an intelligent detection and maintenance method for oil in large gearboxes, provided as an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Please see Figure 1 The diagram illustrates a flowchart of an intelligent detection and maintenance method for large gearbox oil provided in one embodiment of this application. The method includes the following steps: Step S001: Collect the oil parameters of the oil in the gearbox and construct the oil parameter sequence at each collection time. The oil parameters include the viscosity, dielectric constant, density, temperature, moisture content, and number of particles of the oil.

[0019] A multi-parameter sensor group is integrated and installed in the oil circuit of a large gearbox. The multi-parameter sensor group specifically includes a viscosity sensor, a dielectric constant sensor, a density sensor, a temperature sensor, a moisture sensor, and a particle counter. It should be noted that all sensors in the multi-parameter sensor group are arranged in the return oil line of the large gearbox to ensure that the detected oil is representative.

[0020] A multi-parameter sensor array is used to collect oil parameters in the return oil pipeline, including oil viscosity, dielectric constant, density, temperature, moisture content, and particle number.

[0021] The units for viscosity, density, temperature, moisture content, and particle number are: mm² / s, g / cm³, ℃, mg / kg, and particles / ml, respectively.

[0022] In this embodiment, the acquisition frequency of all types of oil parameters is set to 1Hz, and the oil parameters of the same type within a sliding window of 3 preset time lengths before the acquisition time are normalized respectively.

[0023] It should be noted that this embodiment uses the maximum-minimum normalization method to calculate the normalized value. In the maximum-minimum normalization method, the sum of the difference between the maximum and minimum values ​​and a preset minimum positive parameter is used as the denominator to calculate the normalized value. In this embodiment, the preset minimum positive parameter is set to a value of [value missing]. In practical applications, implementers may use other methods from existing technologies, such as the tanh function or the sigmoid function, to calculate the normalized value; no restrictions are imposed here.

[0024] The same type of oil parameters within a sliding window of the acquisition time and a preset time length prior to the acquisition time are arranged sequentially to obtain the oil parameter sequence at the acquisition time. It can be understood that the oil parameter sequence includes viscosity sequence, dielectric constant sequence, density sequence, temperature sequence, moisture content sequence, and particle number sequence.

[0025] In this embodiment, the preset time length is set to 1 hour. It should be noted that if less than 3 preset time lengths are elapsed before the data collection time, oil level detection is not performed; only oil level parameters are collected.

[0026] At this point, the oil parameter sequence at each acquisition time has been obtained.

[0027] Step S002: Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, calculate the first independent component score and quantization error at the acquisition time. Based on the first independent component score and quantization error, calculate the intrinsic deviation index at the acquisition time.

[0028] During the long-term operation of large gearboxes, it is common to change gear oils of different grades and additive formulations, or to mix multiple types of gear oils. Existing large gearbox oil detection models cannot distinguish between the differences in intrinsic physicochemical properties such as dielectric constant and viscosity of the gear oil and the deterioration characteristics caused by actual abnormal wear of the equipment. They are prone to misjudging the fluctuations in inherent basic parameters caused by oil changes as internal wear faults in the gearbox, thus frequently triggering false alarms in the system, seriously affecting the accuracy and reliability of intelligent oil detection and maintenance.

[0029] The FastICA (Fast Independent Component Analysis) algorithm is used to map the feature vector at the current acquisition time using the unmixing matrix trained with a preset sample set. The viscosity, dielectric constant, and density sequences at the same acquisition time are centered and whitened to eliminate second-order correlations between variables, thereby obtaining the first independent component at the same acquisition time and calculating the score of the first independent component.

[0030] In this embodiment, the comparison function of the fast independent component analysis algorithm is set as the negative entropy maximization criterion, and the convergence tolerance is set to [value missing]. Independent component analysis algorithms are well-known techniques and will not be elaborated further.

[0031] Independent component analysis can separate statistically independent non-Gaussian source signals from multivariate observation signals. The first independent component score represents the amplitude of the most significant independent component in the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, which is the projection of the inherent properties of the oil at the acquisition time without wear interference.

[0032] A preset oil quality benchmark is established, representing the inherent property level of the oil when new oil is first added. Specifically, in this embodiment, after the gearbox is first filled with new oil and no abnormalities are confirmed, viscosity, dielectric constant, and density sequences are continuously collected at a preset number of sampling times. The first independent component score is calculated at these preset number of sampling times, and the average of the first independent component scores at these preset number of sampling times is recorded as the oil quality benchmark. In this embodiment, the preset number is set to 100.

[0033] The first independent component score at the time of collection is divided by the sum of the oil quality benchmark and the preset minimum positive parameter, and the result is recorded as the first ratio at the time of collection. The absolute value of the difference between the first ratio at the time of collection and the number 1 is recorded as the first deviation at the time of collection.

[0034] In this embodiment, the value of the minimum positive parameter is... The first ratio is a quantitative result of the relative shift in the intrinsic properties of the oil.

[0035] Using the absolute value of the difference between the first ratio at the time of data acquisition and the number 1 as the first deviation can prevent calculation distortion caused by an excessively large first ratio. The first deviation at the time of data acquisition quantifies the relative shift in the intrinsic characteristics of the oil.

[0036] Even when large gearboxes use the same brand of gear oil for extended periods, the oil will slowly age and deteriorate due to oxidation and gradual loss of additives during continuous service. If this is compounded by changes to different brands or additive formulations of gear oil during on-site maintenance, or by the mixing of different oils, the physicochemical properties of the oil will change significantly, leading to deviations in its fundamental properties that are difficult to quantify accurately. In actual service life, gearboxes commonly experience oil changes, oil mixing, and oil iterations. Existing gearbox oil testing methods cannot distinguish between the inherent physicochemical differences in dielectric constants, viscosity, etc., of different oils and the actual wear and tear characteristics of the equipment. This easily leads to misinterpretations of oil property deviations as abnormal gearbox wear faults, frequently triggering false alarms.

[0037] Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, a multi-dimensional oil feature vector is constructed at the acquisition time. The self-organizing map algorithm is used to process the multi-dimensional oil feature vector at the acquisition time to obtain the Euclidean distance between the multi-dimensional oil feature vector and the weight vector of the best matching neuron in the self-organizing map network. The Euclidean distance is recorded as the quantization error at the acquisition time.

[0038] In this embodiment, the two-dimensional mesh in the self-organizing map algorithm is set to 5. The hexagonal grid with a value of 5 is used, with an initial learning rate of 0.1, a neighborhood radius of 2, and 200 iterations. The self-organizing map algorithm is a well-known technique and will not be described in detail here.

[0039] The self-organizing map algorithm is used to quantize the Euclidean distance between the input vector and the weight vector of the best-matching neuron. The larger the quantization error, the less the multi-dimensional oil feature vector matches the topological structure obtained by the algorithm, and the higher the probability and degree of deviation from the normal pattern. The quantization error represents the degree of overall feature deviation of the oil at the time of acquisition relative to the historical normal oil features.

[0040] The ratio of the quantization error at the acquisition time to the maximum value of the quantization errors at all acquisition times prior to the acquisition time is denoted as the second deviation at the acquisition time. The weighted sum of the first deviation and the second deviation at the acquisition time is denoted as the intrinsic deviation index at the acquisition time. The weighted sum of the first deviation and the second deviation at the acquisition time is 1. In this embodiment, the weights of the first deviation and the second deviation at the acquisition time are both 0.5.

[0041] In the process of calculating the ratio, in order to avoid the denominator being zero, a preset value needs to be added to the denominator. In this embodiment, the preset value is 0.0001.

[0042] The intrinsic deviation index at the time of data collection integrates the characteristic shifts at the level of independent components with the pattern deviations in the overall feature space, and is used to distinguish the characteristic changes of oil products.

[0043] At this point, the intrinsic deviation index at the acquisition time is obtained.

[0044] Step S003: Calculate the particle number residual at the time of sampling based on the intrinsic deviation index, temperature, and particle number. Establish a data pair based on the particle number residual and moisture content at the time of sampling, assign values ​​to the binary eigenvalues ​​of the data pair, and calculate the moisture wear synergy at the time of sampling by combining the particle number residual.

[0045] The intrinsic properties of hydraulic fluids fluctuate normally with oil changes and long-term aging. Additionally, equipment wear can alter these properties. Traditional oil detection methods cannot effectively distinguish the causes of these parameter changes, making it difficult to determine whether the changes are due to differences in the oil's inherent properties or actual component wear, thus increasing the system's false alarm rate.

[0046] The intrinsic deviation indexes of each acquisition time within a sliding window of a preset time length before the acquisition time are arranged sequentially to establish an intrinsic deviation index sequence for each acquisition time. The intrinsic deviation index sequence and temperature sequence are used as inputs and independent variables, and the particle number sequence at each acquisition time is used as input and dependent variable. The random forest regression algorithm is used to obtain the residuals corresponding to the particle number, which are denoted as the particle number residuals at each acquisition time.

[0047] The residual corresponding to the particle number is: the residual between the particle number at the time of sampling and the normal particle number, under the residual oil characteristics and temperature at the corresponding sampling time. This particle number residual represents the particle number residual value after correcting for intrinsic oil properties and temperature fluctuations.

[0048] The particle number residual may be affected by high moisture content, causing spurious values ​​in the data collected by the sensor, making it difficult to reliably distinguish the failure modes of large gearboxes.

[0049] Therefore, the particle number residuals and moisture content at each sampling time are combined to form a data pair at that sampling time. The DBSCAN algorithm is used to cluster all data pairs within a sliding window of a preset time length before and after the sampling time to identify outliers in the data pairs. During the clustering process, the minimum absolute value of the difference between the data pair and the particle number residuals of all cluster centers is recorded as the particle number residual difference of the data pair, and the minimum absolute value of the difference between the data pair and the moisture content of all cluster centers is recorded as the moisture difference of the data pair. When the particle number residual difference of a data pair is greater than twice the moisture difference, and the data pair is an outlier, the binary eigenvalue of the data pair is assigned a value of 1; in all other cases, the binary eigenvalue of the data pair is assigned a value of 0.

[0050] In the process of using the DBSCAN algorithm for clustering, this embodiment sets the neighborhood radius to 0.3 and the minimum number of neighborhood samples to 5. Using the DBSCAN algorithm for clustering and for identifying outliers are well-known techniques and will not be described in detail here.

[0051] The Z-score normalized result of the particle number residual at the collection time is calculated and the maximum value of the number 0 is calculated. The product of the maximum value and the value of the binary eigenvalue of the number pair at the collection time is recorded as the moisture wear synergy at the collection time.

[0052] Z-score standardization is a well-known technique and will not be elaborated further. In this embodiment, Z-score standardization is implemented using data within a sliding window of three preset time lengths prior to the data acquisition time. The Z-score standardization result ensures that only the residual number of particles that deviates positively from the historical mean will result in a maximum value greater than 0 during the calculation process.

[0053] The moisture wear correlation coefficient at the time of sampling is only non-zero when the residual particle number exceeds the normal range, and both moisture and residual particle number are abnormal. This allows for the quantification of wear anomalies after eliminating inherent differences in the oil and moisture interference. The higher the moisture wear correlation coefficient at the time of sampling, the more likely the gearbox is to have genuine abnormal wear at that time; conversely, the lower the moisture wear correlation coefficient at the time of sampling, the more likely the anomaly is due to changes in the oil's inherent properties or fluctuations in moisture content.

[0054] At this point, the degree of moisture abrasion synergy at the time of collection is obtained.

[0055] Step S004: Based on the moisture wear synergy at the time of collection, determine the dynamic contamination rate parameter at the time of collection. Use the dynamic contamination rate parameter at the time of collection as the adaptive value of the contamination rate parameter. Using the isolated forest algorithm, obtain the detection results of the gearbox oil based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density and particle number at the time of collection.

[0056] In the process of intelligent detection and maintenance of large gearbox oils, the intrinsic physicochemical properties of different types of gear oils naturally differ. At the same time, factors such as long-term aging and deterioration of the oil and interference from moisture in the field can cause traditional anomaly detection methods to misjudge, identifying normal fluctuations in the oil's own parameters as abnormal wear of the gearbox, thus triggering a large number of false alarms.

[0057] This application selects the Isolation Forest algorithm for unsupervised anomaly detection of abnormal oil samples. However, the performance of the Isolation Forest algorithm is highly dependent on a pre-set contamination rate parameter, which is used to characterize the estimated proportion of abnormal samples in the dataset. In the scenario of intelligent detection of oil in large gearboxes, the operating status of the oil changes dynamically with service time. Coupled with factors such as oil replacement and natural aging of the oil, the characteristic parameters of normal oil samples will drift overall. Therefore, the Isolation Forest algorithm, which uses a fixed contamination rate configuration, cannot adapt to the dynamic evolution of oil parameters and is prone to incorrectly identifying normal parameter drifts caused by changes in the oil's own properties as abnormal states caused by actual equipment wear.

[0058] The dynamic contamination rate parameter at the time of collection is determined based on the moisture abrasion synergy at the time of collection.

[0059] Specifically: the sum of the moisture wear synergy at the collection time and the number 1 is used as the denominator, the moisture wear synergy at the collection time is used as the numerator, the product of the fraction value and the preset adjustment coefficient is recorded as the characteristic product at the collection time, the sum of the characteristic product at the collection time and the number 1 is used as the first multiplier, and the product of the first multiplier and the preset initial baseline contamination rate is recorded as the dynamic contamination rate parameter at the collection time.

[0060] The adjustment coefficient is used to control the adjustment intensity of the moisture wear synergy on the dynamic contamination rate parameter. The value of the adjustment coefficient should be greater than or equal to 0.5 and less than or equal to 1.5. In this embodiment, the value of the adjustment coefficient is 1.0. The value of the initial baseline contamination rate should be greater than or equal to 0.05 and less than or equal to 0.2. In this embodiment, the value of the initial baseline contamination rate is 0.1.

[0061] In the process of calculating the dynamic contamination rate parameter at the time of data collection, the fraction is used to normalize the moisture wear coordination degree at the time of data collection. This avoids excessive adjustment of the dynamic contamination rate parameter when the moisture wear coordination degree is large, and improves the sensitivity of the subsequent isolated forest algorithm to identify abnormal gear states. At the same time, when the moisture wear coordination degree is small, it maintains a low false alarm sensitivity and keeps the value close to the initial baseline contamination rate.

[0062] The intrinsic deviation index, moisture abrasion synergy, viscosity, dielectric constant, density, and particle number at each sampling time are arranged sequentially to establish a six-dimensional feature vector for that sampling time. The dynamic contamination rate parameter at each sampling time is used as an adaptive value for the contamination rate parameter. The isolated forest algorithm is then applied to process the six-dimensional feature vector at each sampling time to obtain anomaly scores.

[0063] In the process of using the isolated forest algorithm, the number of decision trees was set to 100, and the single subsampling size was set to 256 groups; the use of the isolated forest algorithm to obtain anomaly scores is a well-known technique and will not be described in detail here.

[0064] The 90th percentile of the abnormal scores of the sampling times within a sliding window of 3 preset time lengths prior to the sampling time is used as the abnormal score threshold. Sampling times with abnormal scores greater than the abnormal score threshold are judged as abnormal sampling times, that is, the gearbox oil is abnormal at the abnormal sampling time and needs to be maintained.

[0065] This completes the gearbox oil test.

[0066] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for intelligent monitoring and maintenance of large gear box oil suitable for, characterized by, The method includes the following steps: The oil parameters in the gearbox are collected, and a sequence of oil parameters at each collection time is constructed. The oil parameters include the viscosity, dielectric constant, density, temperature, moisture content, and number of particles of the oil. Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, calculate the first independent component score and quantization error at the acquisition time, and calculate the intrinsic deviation index at the acquisition time based on the first independent component score and quantization error. Based on the intrinsic deviation index, temperature, and particle number, the particle number residual at the time of sampling is calculated. Based on the particle number residual at the time of sampling and moisture, a number pair is established, and the binary eigenvalues ​​of the number pair are assigned. Combined with the particle number residual, the moisture wear synergy at the time of sampling is calculated. Based on the moisture wear synergy at the time of collection, the dynamic contamination rate parameter at the time of collection is determined. The dynamic contamination rate parameter at the time of collection is used as the adaptive value of the contamination rate parameter. Using the isolated forest algorithm, the detection results of gearbox oil are obtained based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density and particle number at the time of collection. The specific calculation method for the intrinsic deviation index at the acquisition time is as follows: Calculate the second deviation of the acquisition time based on the quantization error at the acquisition time; The weighted sum of the first deviation and the second deviation at the acquisition time is denoted as the intrinsic deviation index at the acquisition time, where the weighted sum of the first deviation and the second deviation at the acquisition time is 1. The specific calculation method for the moisture abrasion synergy at the time of collection is as follows: The Z-score normalized result of the particle number residual at the collection time is calculated and the maximum value of the number 0 is calculated. The product of the maximum value and the value of the binary eigenvalue of the number pair at the collection time is recorded as the moisture wear synergy at the collection time. The specific method for determining the dynamic contamination rate parameter at the time of data collection is as follows: The sum of the moisture wear synergy at the time of collection and the number 1 is used as the denominator, and the moisture wear synergy at the time of collection is used as the numerator. The product of the fraction value and the preset adjustment coefficient is recorded as the characteristic product at the time of collection. The sum of the characteristic product at the time of collection and the number 1 is used as the first multiplier. The product of the first multiplier and the preset initial baseline contamination rate is recorded as the dynamic contamination rate parameter at the time of collection.

2. The method for intelligent monitoring and maintenance of large gear box oil as claimed in claim 1 wherein, The specific method for calculating the quantization error is as follows: Based on the viscosity sequence, dielectric constant sequence, and density sequence at the acquisition time, a multi-dimensional oil feature vector is constructed at the acquisition time. Using the self-organizing map algorithm, the Euclidean distance between the multi-dimensional oil features and the weight vector of the best matching neuron in the vector self-organizing map network is obtained, which is denoted as the quantization error at the acquisition time.

3. The method as claimed in claim 1, wherein, The second deviation at the acquisition time is the ratio of the quantization error at the acquisition time to the maximum value of the quantization errors at all acquisition times prior to the acquisition time.

4. The method as claimed in claim 1, wherein, The specific calculation method for the first deviation at the acquisition time is as follows: The first independent component score at the time of collection is divided by the sum of the oil quality benchmark and the preset minimum positive parameter, and the result is recorded as the first ratio at the time of collection. The absolute value of the difference between the first ratio at the time of collection and the number 1 is recorded as the first deviation at the time of collection.

5. The method as claimed in claim 1, wherein, The specific calculation method for the particle number residual at the acquisition time is as follows: An intrinsic deviation index sequence at the time of data acquisition is established. The intrinsic deviation index sequence and temperature sequence at the time of data acquisition are used as independent variables, and the particle number sequence at the time of data acquisition is used as the dependent variable. The random forest regression algorithm is used to obtain the residuals corresponding to the particle number, which are denoted as the particle number residuals at the time of data acquisition.

6. The method for intelligent monitoring and maintenance of large gear box oil as claimed in claim 1 wherein, The specific methods for assigning binary eigenvalues ​​to the logarithmic pairs are as follows: The particle number residual at the sampling time is combined with the moisture content to form a data pair at the sampling time. All data pairs at the sampling time and within a sliding window of a preset time length before the sampling time are clustered to identify outliers in the data pairs. The minimum absolute value of the difference between the number of particles in the number of pairs and the residuals of particle number in all cluster centers is denoted as the particle number residual difference of the number of pairs. The minimum absolute value of the difference between the number of particles in the number of pairs and the residuals of moisture in all cluster centers is denoted as the moisture difference of the number of pairs. When the difference in particle number residuals of a pair is greater than twice the difference in moisture content, and the pair is an outlier, the binary eigenvalue of the pair is assigned a value of 1; otherwise, the binary eigenvalue of the pair is assigned a value of 0.

7. The method as claimed in claim 1, wherein, The isolated forest algorithm is used to obtain the gearbox oil detection results based on the intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density, and particle number at the time of data collection. The specific methods include: The intrinsic deviation index, moisture wear synergy, viscosity, dielectric constant, density and particle number at the time of collection are arranged in order to establish a six-dimensional feature vector at the time of collection. The dynamic contamination rate parameter at the time of collection is used as the adaptive value of the contamination rate parameter. The isolated forest algorithm is used to process the six-dimensional feature vector at the time of collection to obtain the anomaly score at the time of collection. When the abnormal score exceeds the preset abnormal score threshold, the gearbox oil is maintained.

Images