Machine learning based industrial equipment lubricant fluid life prediction method and system

Abstract

The application provides a kind of industrial equipment lubricating oil life prediction method and system based on machine learning, first obtain the lubricating oil state monitoring data set of industrial equipment in preset monitoring period, it contains multiple lubricating state monitoring sequences, each lubricating state monitoring sequence is constituted by lubricating oil property index data and equipment operating state data, then dynamic feature extraction is carried out to the lubricating oil state monitoring data set, the dynamic change feature of lubricating oil property and the equipment operating state association feature are obtained, then based on the preset lubricating life prediction model, the above-mentioned features are fused to generate the lubricating oil life prediction value, finally, according to the lubricating oil life prediction value and the preset lubricating life threshold, the lubricating maintenance decision instruction is generated, and is fed back to the industrial equipment maintenance terminal to trigger the lubricating maintenance operation, the effective prediction and maintenance decision of the life of industrial equipment lubricating oil are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, in particular to an industrial equipment lubricating oil life prediction method and system based on machine learning. BACKGROUND

[0002] In the field of industrial equipment operation and maintenance, lubricating oil as a key medium to ensure the normal operation of equipment, its performance state directly affects the reliability, efficiency and service life of the equipment. The traditional lubricating oil maintenance strategy mainly relies on regular replacement or passive maintenance mode based on simple threshold (such as total running time, mileage or fixed period). Such methods have significant limitations: first, the complexity of the dynamic change of lubricating oil performance with the running condition of the equipment is not fully considered, resulting in over-maintenance (waste of resources) or insufficient maintenance (causing equipment failure); second, existing monitoring technologies focus on offline detection of a single parameter (such as viscosity, acid value), lacking comprehensive analysis of the correlation between multi-dimensional properties of lubricating oil (such as oxidation degree, metal particle content, water content, etc.) and equipment operating conditions (such as temperature, load, speed, etc.); third, traditional methods are difficult to capture the nonlinear characteristics of lubricating oil performance degradation, and cannot early warn potential failure risks, resulting in maintenance decisions lagging behind actual needs.

[0003] In the prior art, some schemes attempt to monitor the state of lubricating oil in real time through sensors, but only stay at the data collection level, without realizing effective fusion and deep mining of multi-source heterogeneous data. For example, some systems only record the changes in physical properties of lubricating oil, without correlating real-time operating condition data of the equipment, resulting in one-sided analysis results; another scheme uses simple statistical models or threshold comparison for life prediction, but is limited by single data dimension and insufficient algorithm robustness, making it difficult to meet the needs of industrial-level application requirements. In addition, existing maintenance decision mechanisms are mostly based on static rules, lack dynamic adaptive ability, and cannot adjust the maintenance strategy according to the actual operating state of the equipment, resulting in low maintenance efficiency and high cost. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide an industrial equipment lubricating oil life prediction method based on machine learning, which comprises:

[0005] obtaining a set of lubricating oil state monitoring data of an industrial equipment within a preset monitoring period, the set of lubricating oil state monitoring data comprising a plurality of lubricating state monitoring sequences, each lubricating state monitoring sequence being composed of lubricating oil attribute index data at at least one lubricating state sampling time and corresponding equipment operating state data;

[0006] performing dynamic feature extraction processing on the lubricating oil state monitoring data set to obtain lubricating oil property dynamic change features and equipment operation state correlation features of each lubricating state monitoring sequence;

[0007] Based on the pre-set lubricating life prediction model, the lubricating oil property dynamic change features and the equipment operation state correlation features are fused and predicted to generate a lubricating oil life prediction value of the lubricating state monitoring sequence.

[0008] According to the lubricating oil life prediction value and a pre-set lubricating life threshold, a lubricating maintenance decision instruction is generated, and the lubricating maintenance decision instruction is fed back to an industrial equipment maintenance terminal to trigger a lubricating maintenance operation.

[0009] In still another aspect, the embodiment of the present application also provides an industrial equipment lubricating oil life prediction system based on machine learning, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above-mentioned method.

[0010] Based on the above aspects, the embodiment of the present application builds a multi-dimensional time series data set containing lubricating oil property indexes and equipment operation states, and in the dynamic feature extraction link, the non-linear law of the evolution of lubricating oil properties with time and the correlation features with the equipment operation states are analyzed, not only the progressive process of lubricating performance degradation is captured, but also the synergistic influence mechanism of working condition parameters such as equipment load and temperature on lubricating life is revealed, thereby the completeness and prediction accuracy of feature representation are significantly improved. Based on the pre-set lubricating life prediction model, multi-feature fusion prediction is performed, the hysteresis and one-sidedness of the traditional threshold judgment method are effectively overcome, the potential patterns in the time series data are deeply mined through the machine learning algorithm, and the accurate quantitative evaluation of the remaining life of the lubricating oil is realized. Finally, the intelligent lubricating maintenance decision instruction is generated through dynamic threshold comparison, not only the equipment failure risk caused by lubricating failure is reduced, but also the lubricating resource consumption is optimized through accurate maintenance timing selection, and the reliability of the industrial equipment operation is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is an execution flow schematic diagram of the industrial equipment lubricating oil life prediction method based on machine learning provided by the embodiment of the present application.

[0012] Figure 2 is a schematic diagram of exemplary hardware and software components of the industrial equipment lubricating oil life prediction system based on machine learning provided by the embodiment of the present application. DETAILED DESCRIPTION

[0013] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a method for predicting the life of lubricating oil in industrial equipment based on machine learning provided by an embodiment of the present invention. The method for predicting the life of lubricating oil in industrial equipment based on machine learning is introduced in detail below.

[0014] Step S110: Obtain a lubricating oil state monitoring data set of industrial equipment within a preset monitoring period, wherein the lubricating oil state monitoring data set includes multiple lubricating state monitoring sequences, and each lubricating state monitoring sequence is composed of lubricating oil property index data and corresponding equipment operating state data at at least one lubricating state sampling moment.

[0015] In this embodiment, taking a key processing equipment in a large manufacturing plant as an example, a preset monitoring period of 60 days is set. During these 60 days, comprehensive lubricating oil sampling and monitoring is conducted every three days. Each sampling operation obtains data on various lubricating oil properties, such as lubricating oil viscosity, acid value, moisture content, and oxidation stability. Equipment operating status data, such as equipment load, operating temperature, vibration amplitude, and rotational speed, are also recorded. For example, during sampling on the third day, the lubricating oil viscosity is 45 (unit: centistokes), the acid value is 0.4 (unit: mgKOH / g), the moisture content is 0.1%, the oxidation stability test time is 800 (unit: minutes); the equipment load is 480 (unit: kilowatts), the operating temperature is 58 (unit: degrees Celsius), the vibration amplitude is 0.05 (unit: mm), and the rotational speed is 1500 (unit: rpm). Based on this pattern, multiple such sampling data sets can be generated within 60 days. These sampling data sets constitute multiple lubrication status monitoring sequences, and thus form a complete lubricating oil status monitoring data set.

[0016] Step S120: performing dynamic feature extraction processing on the lubricating oil state monitoring data set to obtain dynamic change features of lubricating oil properties and equipment operation state correlation features of each lubricating oil state monitoring sequence.

[0017] Next, the processing process of the lubricating oil condition monitoring data set is described in detail.

[0018] Step S121: performing time series segmentation processing on the lubricating oil property index data in the lubrication state monitoring sequence to obtain a plurality of lubrication property change segments.

[0019] In this embodiment, taking lubricating oil viscosity data as an example, 20 sampling data points were obtained during the aforementioned 60-day monitoring period. These 20 sampling data points can then be segmented into segments of 5 samples each. For example, assuming the viscosity data points for these 20 samplings are 45, 47, 50, 48, 49, 51, 53, 55, 54, 56, 58, 60, 59, 61, 63, 62, 64, 66, 65, 67, in order. Then, the data for the first segment is 45, 47, 50, 48, 49; the data for the second segment is 51, 53, 55, 54, 56; the data for the third segment is 58, 60, 59, 61, 63; and the data for the fourth segment is 62, 64, 66, 65, 67. Through this time series segmentation, multiple segments of lubrication property changes are obtained.

[0020] Step S122: calling a preset lubrication characteristic encoder to perform dynamic trend encoding processing on the multiple lubrication property change segments to generate the dynamic change characteristics of the lubricating oil properties of the lubrication state monitoring sequence, wherein the dynamic change characteristics of the lubricating oil properties include the lubricating oil viscosity change rate, the acid value cumulative gradient and the impurity concentration fluctuation trend.

[0021] The sub-steps of this step are further described below:

[0022] Step S1221: for each lubrication property change segment, extract the initial property indicator sequence and corresponding monitoring time interval data in the lubrication property change segment.

[0023] For example, for the first viscosity change segments of 45, 47, 50, 48, and 49, the initial attribute indicator sequence is these five viscosity values, and the corresponding monitoring interval data is all 3 days. For acid value data, assuming that within the same sampling period, the first acid value change segments are 0.4, 0.42, 0.45, 0.43, and 0.44, the initial attribute indicator sequence is these five acid values, and the monitoring interval is also 3 days.

[0024] Step S1222: performing local fluctuation analysis on the initial attribute indicator sequence based on a sliding time window to generate multiple local fluctuation feature subsequences.

[0025] For example, if the sliding time window size is set to 3, for the first viscosity variation segment mentioned above, the first local fluctuation characteristic subsequence is 45, 47, and 50. The fluctuation is analyzed by calculating the difference or ratio of adjacent data. For example, the difference of adjacent data is calculated as 47-45 = 2 and 50-47 = 3. The second local fluctuation characteristic subsequence is 47, 50, and 48, and the difference is calculated as 50-47 = 3 and 48-50 = -2. The third local fluctuation characteristic subsequence is 50, 48, and 49, and the difference is 48-50 = -2 and 49-48 = 1. And so on, multiple local fluctuation characteristic subsequences are obtained. The same method is used to analyze the acid value variation segment.

[0026] Step S1223: calling the convolutional neural network layer in the lubrication feature encoder to perform convolution processing on the local fluctuation feature subsequence to obtain a primary lubrication feature vector.

[0027] For example, assume the convolutional neural network layer has a convolution kernel size of 2 and a stride of 1. A convolution calculation is performed on the first local viscosity fluctuation feature subsequence of 45, 47, and 50. For example, the first convolution kernel is subjected to some calculation (such as a weighted sum, assuming weights are 0.6 and 0.4, respectively) with the first two data points 45 and 47, resulting in 45 × 0.6 + 47 × 0.4 = 45.8. The convolution kernel is then moved to the next data set 47 and 50, and the same calculation is performed, resulting in 47 × 0.6 + 50 × 0.4 = 48.2. After these calculations, a primary lubrication feature vector is obtained, which contains the characteristic information of the local fluctuations in this segment. The same operation is performed on the local fluctuation feature subsequences of acid value and other properties.

[0028] Step S1224: Obtain a weight adjustment coefficient corresponding to the monitoring time interval data, wherein the weight adjustment coefficient is in a nonlinear inverse proportional relationship with the duration of the monitoring time interval data.

[0029] For example, the weight adjustment coefficient for a 3-day monitoring interval is calculated using a nonlinear function. Assuming the nonlinear function is f(x) = 1 / (x^2 + 2), when x = 3, the calculated weight adjustment coefficient is 1 / (3^2 + 2) = 1 / 11 ≈ 0.09. For different monitoring intervals, the corresponding weight adjustment coefficient is calculated using this nonlinear function.

[0030] Step S1225: multiplying the weight adjustment coefficient by the primary lubrication feature vector element by element to obtain a time-weighted lubrication feature vector.

[0031] In this embodiment, each weight adjustment coefficient can be multiplied with the element at the corresponding position in the primary lubrication feature vector. For example, the primary lubrication feature vector is [45.8, 48.2, …], and the weight adjustment coefficient is 0.09, then the first element of the time-weighted lubrication feature vector is 45.8 x 0.09 = 4.122, the second element is 48.2 x 0.09 = 4.338, and so on, to obtain the time-weighted lubrication feature vector.

[0032] Step S1226: calling the bidirectional long short-term memory network layer in the lubrication feature encoder to perform time sequence dependence modeling on the time-weighted lubrication feature vector, to generate a dynamic trend encoding result of the lubrication attribute change segment.

[0033] Further describe the sub-steps of this step:

[0034] Step S12261: inputting the time-weighted lubrication feature vector into the forward processing branch of the bidirectional long short-term memory network layer, performing forward sequence processing on the time-weighted lubrication feature vector in time sequence, to generate a forward hidden state sequence.

[0035] For example, the first element 4.122 of the time-weighted lubrication feature vector is input at the first time step, and after internal calculation of the network (such as multiplication with a weight matrix, addition of a bias term, and operation through an activation function), the first forward hidden state is obtained; the second element 4.338 is input at the second time step, and similar calculation is performed in combination with the first forward hidden state, to obtain the second forward hidden state, and so on, to generate the forward hidden state sequence.

[0036] Step S12262: synchronously inputting the time-weighted lubrication feature vector into the backward processing branch of the bidirectional long short-term memory network layer, performing backward sequence processing on the time-weighted lubrication feature vector in time reverse order, to generate a backward hidden state sequence.

[0037] That is, starting from the last time step, the calculation of similar forward processing is performed in reverse. For example, the last element of the time-weighted lubrication feature vector is input at the last time step, and the first backward hidden state is obtained through bidirectional long short-term memory network calculation; the second last element is input at the second last time step, and the second backward hidden state is obtained through calculation in combination with the first backward hidden state, to generate the backward hidden state sequence.

[0038] Step S12263: performing cross-sequence feature alignment on the forward hidden state sequence and the backward hidden state sequence, to obtain a splicing feature vector of the forward hidden state and the backward hidden state at each time step.

[0039] For example, the forward hidden state and the backward hidden state of the first time step are concatenated in dimension. Assuming that the forward hidden state is a 10-dimensional vector [a1, a2, …, a10] and the backward hidden state is a 10-dimensional vector [b1, b2, …, b10], the concatenated feature vector is [a1, a2, …, a10, b1, b2, …, b10], forming a new concatenated feature vector.

[0040] Step S12264: performing nonlinear activation and feature dimensionality reduction processing on the concatenated feature vector to generate a temporal dependency feature of the time-weighted lubrication feature vector at a corresponding time step.

[0041] The concatenated feature vector is processed using a nonlinear activation function, such as the ReLU function (which takes the larger of the input value and 0). For example, for each element in the concatenated feature vector, if the element value is less than 0, it is set to 0. Feature dimensionality reduction is then performed, such as using a fully connected layer to reduce the concatenated feature vector from high to low dimensions, for example, from 20 dimensions to 10 dimensions. The temporal dependency features of each time step are then calculated.

[0042] Step S12265: Aggregate the dynamic trend coding results of all lubrication property change segments to generate dynamic change characteristics of lubricating oil properties of the lubrication state monitoring sequence.

[0043] In this embodiment, the dynamic trend encoding results obtained from all segments can be spliced ​​together in terms of dimensions to form a dynamic change signature of lubricating oil properties, including information such as the lubricating oil viscosity change rate, acid value cumulative gradient, and impurity concentration fluctuation trend. For example, the dynamic trend encoding result of the first segment is a 10-dimensional vector, and the second segment is also a 10-dimensional vector. These are spliced ​​together to form a 20-dimensional vector, and so on, ultimately resulting in a complete dynamic change signature of lubricating oil properties.

[0044] Step S123: performing operating mode matching processing on the equipment operating status data in the lubrication status monitoring sequence to obtain equipment operating status correlation characteristics, wherein the equipment operating status correlation characteristics include at least one of the following: equipment load fluctuation characteristics, temperature correlation influence coefficient, and mechanical vibration frequency correlation.

[0045] The sub-steps of this step are further described below:

[0046] Step S1231: extracting the load fluctuation sequence, temperature change sequence and vibration frequency sequence from the equipment operation status data.

[0047] For example, the load fluctuation sequence is the equipment load value recorded at each sampling within 60 days, assuming it is 480, 490, 500, 495, 485, 510, 520, 515, 505, 530, 540, 535, 525, 550, 560, 555, 545, 570, 565, 575 in sequence; the temperature change sequence is the temperature value recorded at each sampling, such as 58, 60, 63, 61, 59, 62, 64, 63, 62, 65, 66, 65, 64, 67, 68, 67, 66, 69, 68, 70; the vibration frequency sequence is the mechanical vibration frequency value recorded at each sampling, such as 50, 52, 55, 53, 54, 56, 58, 60, 59, 61, 63, 62, 64, 65, 66, 65, 64, 67, 66, 68.

[0048] Step S1232: performing peak detection and valley alignment processing on the load fluctuation sequence to generate a device load fluctuation feature.

[0049] In this embodiment, a rule-based algorithm can be used to detect peaks and valleys in a load fluctuation sequence. For example, after calculation, the peaks are 575, 570, 565, 560, and 555, and the valleys are 480, 485, 490, 495, and 500. The peaks and valleys are then aligned and organized into a specific feature representation according to certain rules to generate a device load fluctuation signature. For example, the peaks and valleys can be combined into a new vector, or a characteristic value representing the load fluctuation, such as the difference between the peaks and valleys or the average value, can be calculated using certain statistical methods to form the device load fluctuation signature.

[0050] Step S1233: Matching the temperature change sequence with a preset temperature influence model to calculate a temperature correlation influence coefficient, wherein the temperature correlation influence coefficient is used to characterize the cumulative influence of temperature changes on lubricating oil property indicators.

[0051] For example, the sub-steps of step S1233 are as follows:

[0052] Step S12331: extracting historical temperature values ​​and corresponding monitoring timestamps in the temperature change sequence.

[0053] For example, for the temperature change sequence 58, 60, 63, 61, 59, 62, 64, 63, 62, 65, 66, 65, 64, 67, 68, 67, 66, 69, 68, 70, the corresponding monitoring timestamps start from the 3rd day and are 3, 6, 9, 12, 15, 18, 21, 24, 27, 30, 33, 36, 39, 42, 45, 48, 51, 54, 57, 60.

[0054] Step S12332: Segment the temperature change sequence into a plurality of temperature fluctuation time windows according to the reference temperature threshold and time window segmentation rule defined in the preset temperature influence model.

[0055] For example, assuming the reference temperature threshold is 60 degrees Celsius and the time window segmentation rule is 10 days per window, the first temperature fluctuation time window contains data 58, 60, 63, and 61, covering the 3rd to 12th day; the second window contains data 59, 62, 64, and 63, covering the 15th to 24th day; and so on, segmenting the entire temperature change sequence into multiple temperature fluctuation time windows.

[0056] Step S12333: For each temperature fluctuation time window, calculate the cumulative value of temperature deviations between all historical temperature values ​​in the window and the reference temperature threshold.

[0057] For example, for the first temperature fluctuation time window, the accumulated temperature deviation is (58-60)+(60-60)+(63-60)+(61-60)=-2+0+3+1=2. The accumulated temperature deviation is calculated in the same way for other windows.

[0058] Step S12334: input the temperature deviation cumulative value corresponding to each temperature fluctuation time window into a preset nonlinear influence function for mapping, and obtain the window influence factor of each temperature fluctuation time window.

[0059] For example, assuming the preset nonlinear influence function is f(x) = x^2 + 1, for the first window's accumulated temperature deviation value of 2, the resulting window influence factor after mapping is 2^2 + 1 = 5. This mapping operation is performed on the accumulated temperature deviation value of each window to obtain the window influence factor for each window.

[0060] Step S12335: performing time-weighted processing on the window impact factor of each temperature fluctuation time window according to the time attenuation coefficient configured in the temperature impact model to generate a weighted window impact factor sequence.

[0061] For example, assuming the time decay coefficient is 0.9 and the window impact factor of the first window is 5, the time weight is calculated based on the ratio of the window start time to the total monitoring period. For example, if the first window starts on the third day and the total monitoring period is 60 days, the time weight is 3 / 60 = 0.05. The weighted window impact factor is 5 × 0.9^0.05 ≈ 4.9. Using this method, the window impact factor of each window is time-weighted to generate a weighted window impact factor sequence.

[0062] Step S12336: cumulatively summing the weighted window influence factor sequence to generate a temperature-related influence coefficient of the temperature change sequence.

[0063] In this embodiment, all values ​​in the weighted window influence factor sequence may be added together. For example, the weighted window influence factor sequence is [4.9, 5.2, ...], and the temperature correlation influence coefficient is obtained by cumulative summation. Assume that the final result is 30.

[0064] Step S1234: determining a mechanical vibration frequency correlation according to the frequency domain energy distribution of the vibration frequency sequence, wherein the mechanical vibration frequency correlation is used to quantify the accelerated attenuation effect of mechanical vibration on the life of the lubricating oil.

[0065] First, the vibration frequency sequence is transformed into the frequency domain, for example, using a fast Fourier transform (FFT) to convert the time-domain frequency sequence into a frequency-domain representation. Assuming the vibration frequency sequence is 50, 52, 55, 53, 54, 56, 58, 60, 59, 61, 63, 62, 64, 65, 66, 65, 64, 67, 66, 68, the FFT transform yields the energy distribution in the frequency domain. Next, by analyzing certain characteristics of the frequency-domain energy distribution, such as the energy percentage within a specific frequency range and the location of the energy peak, the correlation between the mechanical vibration frequencies is determined. For example, if the energy percentage within a key frequency range is high, it indicates that vibration at that frequency has a significant impact on the lubricant life. Based on pre-defined rules, this correlation is quantified into a numerical value, assuming a value of 0.6, indicating that mechanical vibration has a certain degree of accelerated attenuation effect on the lubricant life.

[0066] Step S130: Based on a preset lubrication life prediction model, the dynamic change characteristics of the lubricating oil properties and the associated characteristics of the equipment operation status are fused and predicted to generate a lubricating oil life prediction value of the lubrication status monitoring sequence.

[0067] The sub-steps of this step are further described below:

[0068] Step S131: Dividing the dynamic change characteristics of the lubricating oil properties into a viscosity change characteristic subset, an acid value accumulation characteristic subset, and an impurity fluctuation characteristic subset.

[0069] For example, assuming the dynamic change characteristics of lubricating oil properties are a 30-dimensional vector, and the first 10 dimensions primarily reflect changes in lubricating oil viscosity, these 10 dimensions can be extracted to form a viscosity change feature subset. The middle 10 dimensions, primarily related to acid value accumulation, form the acid value accumulation feature subset after extraction. The final 10 dimensions reflect the fluctuation trend of impurity concentration, forming the impurity fluctuation feature subset. In this way, by dividing the dimensions of the dynamic change characteristics of lubricating oil properties, different aspects of the features can be separated, facilitating the subsequent targeted fusion of features associated with equipment operating status.

[0070] Step S132: Divide the equipment operation status associated features into a load associated feature subset, a temperature associated feature subset, and a vibration associated feature subset.

[0071] The equipment operating status correlation feature is assumed to be a 25-dimensional vector. The first 8 dimensions primarily describe the fluctuations in the equipment load and are extracted as the load-correlated feature subset. The middle 8 dimensions are related to the temperature-correlated influence coefficient and the combined impact of temperature changes on the equipment and lubricating oil, forming the temperature-correlated feature subset. The final 9 dimensions reflect the correlation between mechanical vibration frequency and other vibration-related features, forming the vibration-correlated feature subset. This division enables the effective matching and integrated analysis of different key factors of the equipment operating status with the lubricating oil properties.

[0072] Step S133: calling the feature cross layer in the lubrication life prediction model to cross-combine the viscosity change feature subset and the load correlation feature subset to generate a first cross feature vector.

[0073] The sub-steps of this step are further described below:

[0074] Step S1331: extracting the viscosity gradient sequence from the viscosity change feature subset and the load fluctuation amplitude sequence from the load association feature subset.

[0075] In the viscosity variation feature subset, the viscosity gradient sequence is obtained by calculating the difference between adjacent viscosity values. For example, the 10-dimensional data corresponding to the viscosity variation feature subset is [v1, v2, v3, v4, v5, v6, v7, v8, v9, v10], and the calculated viscosity gradient sequence is [v2-v1, v3-v2, v4-v3, v5-v4, v6-v5, v7-v6, v8-v7, v9-v8, v10-v9]. For the load-related feature subset, assuming the 8-dimensional data is [l1, l2, l3, l4, l5, l6, l7, l8], the load fluctuation amplitude sequence is obtained by calculating the absolute value of the difference between adjacent load values, such as [|l2-l1|, |l3-l2|, |l4-l3|, |l5-l4|, |l6-l5|, |l7-l6|, |l8-l7|].

[0076] Step S1332: Calculate the dynamic correlation coefficient matrix between the viscosity gradient sequence and the load fluctuation amplitude sequence.

[0077] In this embodiment, a dynamic correlation analysis method can be used to consider the changes in the viscosity gradient sequence and the load fluctuation amplitude sequence over time. For example, the correlation between corresponding elements of the two sequences is calculated within different time windows. Assuming the time window is 3, for the viscosity gradient sequence [g1, g2, g3, g4, g5] and the load fluctuation amplitude sequence [a1, a2, a3, a4, a5], the correlation between g1 and a1, g2 and a2, and g3 and a3 is calculated within the first time window, and so on. A dynamic correlation coefficient matrix is ​​obtained. This dynamic correlation coefficient matrix can reflect the changes in the degree of correlation between the viscosity gradient and the load fluctuation amplitude at different times.

[0078] Step S1333: performing weighted correction on the viscosity gradient sequence based on the dynamic correlation coefficient matrix to obtain a corrected viscosity gradient characteristic vector.

[0079] In this embodiment, the viscosity gradient sequence can be weighted using the value of each element in the dynamic correlation coefficient matrix as a weight. For example, if the elements at corresponding positions in the dynamic correlation coefficient matrix are [w1, w2, w3, w4, w5] and the viscosity gradient sequence is [g1, g2, g3, g4, g5], then the first element of the corrected viscosity gradient characteristic vector is g1×w1, the second element is g2×w2, and so on, resulting in a corrected viscosity gradient characteristic vector. This weighted correction takes into account the dynamic relationship between the viscosity gradient and the load fluctuation amplitude, allowing the viscosity gradient characteristic to more accurately reflect load-related information.

[0080] Step S1334: performing feature concatenation on the corrected viscosity gradient feature vector and the load fluctuation amplitude sequence to generate a first cross feature vector.

[0081] In this embodiment, the corrected viscosity gradient feature vector and the load fluctuation amplitude sequence can be spliced ​​in terms of dimension. Assuming that the corrected viscosity gradient feature vector is a 5-dimensional vector [m1, m2, m3, m4, m5] and the load fluctuation amplitude sequence is a 5-dimensional vector [a1, a2, a3, a4, a5], the first cross feature vector is [m1, m2, m3, m4, m5, a1, a2, a3, a4, a5]. Through this splicing method, the correlation characteristics between viscosity change and load fluctuation are integrated.

[0082] Step S134: performing nonlinear mapping processing on the acid value cumulative feature subset and the temperature associated feature subset to generate a second cross feature vector.

[0083] The sub-steps of this step are further described below:

[0084] Step S1341: Acquire the acid value accumulation rate data in the acid value accumulation feature subset and the temperature influence coefficient in the temperature correlation feature subset.

[0085] In this embodiment, within the acid value accumulation feature subset, acid value accumulation rate data is obtained by calculating the change in acid value within a certain time interval. For example, the 10-dimensional data corresponding to the acid value accumulation feature subset reflects the accumulation of acid values ​​at different time points. The difference between the acid values ​​at adjacent time points is calculated and divided by the time interval to obtain the acid value accumulation rate data. For the temperature-dependent feature subset, the calculated temperature influence coefficient is directly extracted.

[0086] Step S1342: inputting the acid value accumulation rate data into a preset exponential mapping function to obtain a nonlinear acid value accumulation characteristic.

[0087] In this embodiment, a preset exponential mapping function is assumed to be f(x) = exp(x). Each value in the acid value accumulation rate data is substituted into this function for calculation. For example, if the acid value accumulation rate data is [r1, r2, r3], the nonlinear acid value accumulation characteristic obtained after mapping is [exp(r1), exp(r2), exp(r3)]. This exponential mapping can highlight the changing trend of the acid value accumulation rate and convert the linear acid value accumulation rate into a nonlinear characteristic that better reflects its impact on the life of the lubricating oil.

[0088] Step S1343: multiplying the nonlinear acid value cumulative characteristic by the temperature influence coefficient element by element to obtain a temperature-weighted acid value characteristic vector.

[0089] In this embodiment, it is assumed that the nonlinear acid value accumulation feature is [n1, n2, n3], and the temperature influence coefficient is t, then the first element of the temperature weighted acid value feature vector is n1xt, the second element is n2xt, and the third element is n3xt. By this element-by-element multiplication method, the influence of temperature on acid value accumulation is integrated into the feature vector, which reflects the potential influence of the interaction between temperature and acid value accumulation on the service life of lubricating oil.

[0090] Step S1344: calling the full connection layer in the lubrication life prediction model to perform dimension reduction processing on the temperature weighted acid value feature vector to generate a second cross feature vector.

[0091] In this embodiment, the full connection layer performs matrix multiplication operation on the temperature weighted acid value feature vector through a weight matrix and adds a bias term, and then according to the set dimension reduction target, for example, reduces the vector from 3D to 2D. It is assumed that the temperature weighted acid value feature vector is [x1, x2, x3], the weight matrix is [[w11, w12], [w21, w22], [w31, w32]], the bias term is [b1, b2], and the second cross feature vector [y1, y2] after dimension reduction is obtained after calculation, wherein y1=x1×w11+x2×w21+x3×w31+b1, y2=x1×w12+x2×w22+x3×w32+b2.

[0092] Step S135: performing attention weight distribution processing on the impurity fluctuation feature subset and the vibration correlation feature subset to generate a third cross feature vector.

[0093] The sub-steps of this step are further described as follows:

[0094] Step S1351: extracting the impurity concentration fluctuation sequence in the impurity fluctuation feature subset and the vibration frequency correlation degree in the vibration correlation feature subset.

[0095] In this embodiment, in the impurity fluctuation feature subset, the sequence data reflecting the fluctuation of impurity concentration over time is directly extracted. For example, the 10-dimensional data corresponding to the impurity fluctuation feature subset reflects the fluctuation of impurity concentration at different sampling times, and the sequence is extracted as the impurity concentration fluctuation sequence. For the vibration correlation feature subset, the mechanical vibration frequency correlation degree value calculated before is extracted.

[0096] Step S1352: calling the attention mechanism layer in the lubrication life prediction model to calculate the attention weight distribution between the impurity concentration fluctuation sequence and the vibration frequency correlation degree.

[0097] In this embodiment, the attention mechanism layer determines the attention weight distribution by calculating a certain similarity or correlation between the impurity concentration fluctuation sequence and the vibration frequency correlation. For example, using the dot product attention mechanism, each element in the impurity concentration fluctuation sequence is dot-producted with the vibration frequency correlation, and then the above dot product result is normalized by the softmax function to obtain the attention weight distribution. Assuming that the impurity concentration fluctuation sequence is [i1, i2, i3] and the vibration frequency correlation is v, the dot product [i1×v, i2×v, i3×v] is calculated, and after processing with the softmax function, the attention weight distribution [w1, w2, w3] is obtained, where w1+w2+w3=1, and each weight value represents the importance of the impurity concentration fluctuation element relative to the vibration frequency correlation.

[0098] Step S1353: Dynamically weight the impurity concentration fluctuation sequence according to the attention weight distribution to generate a weighted impurity fluctuation feature.

[0099] In this embodiment, the attention weight distribution can be element-wise multiplied with the impurity concentration fluctuation sequence to obtain a weighted impurity fluctuation signature. For example, if the impurity concentration fluctuation sequence is [i1, i2, i3] and the attention weight distribution is [w1, w2, w3], then the weighted impurity fluctuation signature is [i1×w1, i2×w2, i3×w3]. This dynamic weighting can highlight the more important parts of the impurity concentration fluctuation when it is correlated with the vibration frequency.

[0100] Step S1354: performing feature fusion on the weighted impurity fluctuation feature and the vibration frequency correlation to generate a third cross feature vector.

[0101] In this embodiment, the weighted impurity fluctuation characteristics and the vibration frequency correlation can be spliced ​​in terms of dimensions. Assuming that the weighted impurity fluctuation characteristics are a 3D vector [j1, j2, j3] and the vibration frequency correlation is v, then the third cross feature vector is [j1, j2, j3, v]. Through this fusion method, the information of impurity fluctuation and vibration frequency correlation is integrated.

[0102] Step S136: Fusing the first cross feature vector, the second cross feature vector, and the third cross feature vector to generate a multi-dimensional lubricating oil feature set.

[0103] In this embodiment, the first cross-feature vector, the second cross-feature vector, and the third cross-feature vector can be concatenated in terms of dimension. Assuming that the first cross-feature vector is a 10-dimensional vector [f1, f2, ..., f10], the second cross-feature vector is a 2-dimensional vector [s1, s2], and the third cross-feature vector is a 4-dimensional vector [t1, t2, t3, t4], the multi-dimensional lubricating oil feature set is [f1, f2, ..., f10, s1, s2, t1, t2, t3, t4], forming a comprehensive feature set that includes multiple correlation features such as viscosity and load, acid value and temperature, impurities and vibration.

[0104] Step S137: calling the regression prediction layer in the lubrication life prediction model to perform life prediction on the multi-dimensional lubricating oil feature set to generate the lubricating oil life prediction value.

[0105] In this embodiment, the regression prediction layer uses pre-trained parameters and a pre-set regression algorithm, such as linear regression or nonlinear regression. Assuming a linear regression algorithm, the regression prediction layer includes a weight matrix W and a bias term b. The multi-dimensional lubricating oil feature set is an n-dimensional vector X = [x1, x2, …, xn]. The lubricating oil life prediction value y is calculated by calculating y = WX + b. The calculation process here involves matrix multiplication and addition operations. Each row of the weight matrix W corresponds to the weight of a feature dimension. The influence of different features on lubricating oil life is learned through pre-training. The bias term b is a constant term used to adjust the prediction result. For example, the weight matrix W is [[w11, w12, …, w1n], [w21, w22, …, w2n], …, [wm1, wm2, …, wmn]], the bias term b is [b1, b2, …, bm], and the calculated prediction value y is a scalar representing the predicted lubricating oil life.

[0106] Step S140: generating a lubrication maintenance decision instruction according to the lubricating oil life prediction value and a preset lubrication life threshold, and feeding back the lubrication maintenance decision instruction to the industrial equipment maintenance terminal to trigger a lubrication maintenance operation.

[0107] The sub-steps of this step are further described below:

[0108] Step S141: When the predicted value of the lubricating oil life is less than a first lubricating oil life threshold, a first lubricating maintenance decision instruction is generated, where the first lubricating maintenance decision instruction is used to instruct an immediate lubricating oil replacement operation.

[0109] In this embodiment, assume that the first lubrication life threshold is set at 30 days, and the lubrication life prediction model calculates a predicted lubricant life value of 20 days. Since 20 days is less than 30 days, a first lubrication maintenance decision instruction is generated. This first lubrication maintenance decision instruction is sent to the relevant operation terminal via the industrial equipment maintenance system, explicitly instructing the operator to immediately replace the lubricant in the industrial equipment to avoid equipment failure or damage due to severe degradation of the lubricant performance.

[0110] Step S142: When the predicted value of the lubricating oil life is between the first lubricating oil life threshold and the second lubricating oil life threshold, a second lubricating maintenance decision instruction is generated, and the second lubricating maintenance decision instruction is used to trigger an operation to increase the frequency of real-time monitoring of the lubricating oil state.

[0111] For example, the first lubrication life threshold is 30 days, the second lubrication life threshold is 45 days, and the predicted lubricant life is 35 days. Because 30 days < 35 days < 45 days, a second lubrication maintenance decision instruction can be generated. This second lubrication maintenance decision instruction causes the industrial equipment monitoring system to increase the frequency of real-time lubricant oil status monitoring from every three days to every day. This increased monitoring frequency allows for more timely monitoring of lubricant oil status changes, allowing for timely action if lubricant performance deteriorates significantly.

[0112] Step S143: When the predicted value of the lubricating oil life is greater than the second lubricating oil life threshold, a third lubricating maintenance decision instruction is generated, and the third lubricating maintenance decision instruction is used to maintain the current lubricating maintenance strategy and output remaining life prompt information.

[0113] In this embodiment, assume that the second lubrication life threshold is 45 days and the predicted lubricant life is 50 days. Since 50 days is greater than 45 days, a third lubrication maintenance decision instruction can be generated. This third lubrication maintenance decision instruction instructs the industrial equipment to continue operating according to the current lubrication maintenance strategy and simultaneously displays a remaining life prompt on the operation terminal or related monitoring interface, such as "The remaining life of the lubricant is approximately 50 days. The current lubrication maintenance strategy can continue." This allows operators to understand the remaining usable time of the lubricant and appropriately plan subsequent maintenance work.

[0114] In this embodiment, the lubrication maintenance decision instruction is sent to the industrial equipment maintenance terminal through the communication network of the industrial equipment. After receiving the lubrication maintenance decision instruction, the industrial equipment maintenance terminal can perform the corresponding operation according to the instruction type. If it is the first lubrication maintenance decision instruction, the terminal will pop up a prompt box, clearly displaying "Perform lubricating oil replacement operation immediately", and provide detailed operation steps and precautions to guide the operator to perform the replacement operation; if it is the second lubrication maintenance decision instruction, the terminal will automatically adjust the parameters of the monitoring system, increase the monitoring frequency of the lubricating oil status to the specified interval time, and record the relevant parameter adjustment information; if it is the third lubrication maintenance decision instruction, the terminal will display the remaining life prompt information on the main interface, while keeping the current lubrication maintenance policy settings unchanged, to ensure that the lubrication maintenance work of the industrial equipment can be carried out according to a reasonable plan.

[0115] Based on the above steps, the embodiment of the present application constructs a multi-dimensional time series data set containing lubricating oil property indicators and equipment operating status. The dynamic feature extraction link analyzes the nonlinear laws of the evolution of lubricating oil properties over time and the correlation characteristics with the equipment operating status, which not only captures the gradual process of lubrication performance degradation, but also reveals the synergistic influence mechanism of equipment load, temperature and other working parameters on lubrication life, thereby significantly improving the completeness of feature characterization and prediction accuracy. Multi-feature fusion prediction is performed based on the preset lubrication life prediction model, which effectively overcomes the lag and one-sidedness of the traditional threshold judgment method. The potential patterns in the time series data are deeply mined through machine learning algorithms to achieve accurate quantitative evaluation of the remaining life of the lubricating oil. Finally, intelligent lubrication maintenance decision instructions are generated through dynamic threshold comparison, which not only reduces the risk of equipment failure due to lubrication failure, but also optimizes lubrication resource consumption through accurate maintenance timing selection, significantly improving the reliability of industrial equipment operation.

[0116] Furthermore, the training process of the lubrication life prediction model includes the following steps:

[0117] Step S210: Acquire a sample monitoring data set, wherein the sample monitoring data set includes a plurality of sample monitoring sequences, each of which consists of lubricating oil property index data, equipment operating status data, and corresponding actual lubrication life annotation values ​​arranged in chronological order.

[0118] In the actual sample data collection process, sample monitoring data can be obtained from multiple pieces of similar industrial equipment. For example, operational data from 50 pieces of equipment is collected over a period of time. For each piece of equipment, sampling and monitoring are performed at regular intervals. Assume that the sampling interval is 2 days and monitoring is continuous for 90 days. Each sampling period records various lubricant properties, such as viscosity, acid value, moisture content, and additive content, as well as equipment operating status data, such as load, operating temperature, vibration frequency, and rotational speed. Furthermore, the actual lubricant life rating corresponding to each sampling period is determined based on actual equipment operation records and maintenance history. For example, for the first piece of equipment, at the time of sampling on the second day, the lubricant viscosity was 42 (centistokes), the acid value was 0.35 (mgKOH / g), the moisture content was 0.08%, and the additive content was 3%. The equipment load was 450 (kW), the operating temperature was 55 (°C), the vibration frequency was 48 (Hz), and the rotational speed was 1450 (rpm). The actual lubricant life rating was 60 days. In this way, complete data is recorded for each sampling of each device within 90 days, forming multiple sample monitoring sequences, and all the above sequences constitute the sample monitoring data set.

[0119] Step S220: performing dynamic trend coding processing on the lubricating oil property index data in each sample monitoring sequence to generate dynamic change characteristics of the sample lubricating oil properties.

[0120] This step is similar to the dynamic trend coding process of the lubricating oil property index data in the previous step S122, which may specifically include the following steps:

[0121] Step S221 : for each sample lubrication property change segment, extract the initial property indicator sequence and the corresponding monitoring time interval data in the lubrication property change segment.

[0122] For example, for the lubricating oil viscosity data in a sample monitoring sequence, within a specific time segment, the data are 42, 44, 46, 45, and 47. The initial attribute indicator sequence is these 5 viscosity values, and the corresponding monitoring time interval data are all 2 days.

[0123] Step S222: performing local fluctuation analysis on the initial attribute indicator sequence based on a sliding time window to generate multiple local fluctuation feature subsequences.

[0124] For example, if the sliding time window size is set to 3, for the above viscosity sequence, the first local fluctuation characteristic subsequence is 42, 44, and 46. The change rates of adjacent data are calculated, such as (44-42) / 42≈0.048 and (46-44) / 44≈0.045, to characterize the fluctuation within this subsequence. The second local fluctuation characteristic subsequence is 44, 46, and 45, and the change rates are calculated as (46-44) / 44≈0.045 and (45-46) / 46≈-0.022. The third local fluctuation characteristic subsequence is 46, 45, and 47, and the change rates are (45-46) / 46≈-0.022 and (47-45) / 45≈0.044. In this way, multiple local fluctuation characteristic subsequences are generated, which can fully reflect the fluctuation characteristics of lubricating oil property indicators in different local time periods.

[0125] Step S223: calling the convolutional neural network layer in the lubrication feature encoder to perform convolution processing on the local fluctuation feature subsequence to obtain a primary lubrication feature vector.

[0126] For example, assume the convolutional neural network layer has a convolution kernel size of 2 and a stride of 1. For the first local fluctuation feature subsequence 42, 44, and 46, a weighted calculation is performed on the first convolution kernel and the first two data points 42 and 44 (assuming weights are 0.7 and 0.3, respectively), resulting in 42 × 0.7 + 44 × 0.3 = 42.6. The convolution kernel is then moved to the next set of data points 44 and 46, and the same calculation is performed, resulting in 44 × 0.7 + 46 × 0.3 = 44.6. This convolution calculation is performed on all local fluctuation feature subsequences, ultimately resulting in a primary lubrication feature vector that integrates the key information in the local fluctuation feature subsequence.

[0127] Step S224: obtaining a weight adjustment coefficient corresponding to the monitoring time interval data, wherein the weight adjustment coefficient is in a nonlinear inverse proportional relationship with the duration of the monitoring time interval data.

[0128] For example, the function f(x) = 1 / (x^2 + 3) can be used to calculate the weight adjustment coefficient. When the monitoring interval x = 2 days, the calculated weight adjustment coefficient is 1 / (2^2 + 3) = 1 / 7 ≈ 0.143. For different monitoring intervals in different sample monitoring sequences, the corresponding weight adjustment coefficient is calculated based on this function to reflect the impact of the interval on the data importance.

[0129] Step S225: multiplying the weight adjustment coefficient by the primary lubrication feature vector element by element to obtain a time-weighted lubrication feature vector.

[0130] For example, assuming the primary lubrication eigenvector is [v1, v2, v3] and the weight adjustment coefficient is 0.143, then the first element of the time-weighted lubrication eigenvector is v1 × 0.143, the second element is v2 × 0.143, and the third element is v3 × 0.143. This element-by-element multiplication operation incorporates the influence of time on the eigenvector, allowing the eigenvector to more accurately reflect the dynamic changes in lubricating oil properties at different time scales.

[0131] Step S226: calling the bidirectional long short-term memory network layer in the lubrication feature encoder to perform temporal dependency modeling on the time-weighted lubrication feature vector to generate a dynamic trend encoding result of the lubrication attribute change segment.

[0132] Step S2261: Input the time-weighted lubrication feature vector into the forward processing branch of the bidirectional long short-term memory network layer, perform forward sequence processing on the time-weighted lubrication feature vector in chronological order, and generate a forward hidden state sequence.

[0133] For example, at the first time step, the first element of the time-weighted lubrication feature vector is input into the bidirectional long short-term memory network layer. After multiplication by the weight matrix within the bidirectional long short-term memory network layer, the bias term is added, and the activation function (such as the tanh function) is processed to obtain the first forward hidden state. At the second time step, the second element is input into the network along with the first forward hidden state for calculation to obtain the second forward hidden state. This process continues, generating a sequence of forward hidden states. In this process, the bidirectional long short-term memory network layer learns the chronological characteristic changes of the time-weighted lubrication feature vector.

[0134] Step S2262: synchronously inputting the time-weighted lubrication feature vector into the backward processing branch of the bidirectional long short-term memory network layer, performing backward sequence processing on the time-weighted lubrication feature vector in reverse time order, and generating a backward hidden state sequence.

[0135] For example, starting with the last element of the time-weighted lubrication feature vector, it is fed into the backward processing branch network. Following a similar computational process to the forward processing, the first backward hidden state is obtained. The second-to-last element is then fed into the network along with the first backward hidden state to calculate the second backward hidden state, thus generating a sequence of backward hidden states. Through backward processing, the network captures feature dependencies in reverse time.

[0136] Step S2263: performing cross-sequence feature alignment on the forward hidden state sequence and the backward hidden state sequence to obtain a concatenated feature vector of the forward hidden state and the backward hidden state at each time step.

[0137] For example, the forward hidden state sequence and the backward hidden state sequence are spliced at each time step. For example, at the first time step, the forward hidden state is an 8-dimensional vector [hi_1, hi_2, …, hi_8], and the backward hidden state is an 8-dimensional vector [h2_1, h2_2, …, h2_8], and the spliced feature vector is [hi_1, hi_2, …, hi_8, h2_1, h2_2, …, h2_8], forming a new vector containing forward and backward feature information.

[0138] Step S2264: performing nonlinear activation and feature dimension reduction processing on the spliced feature vector to generate the time-sequential dependence feature of the time- weighted lubrication feature vector at the corresponding time step.

[0139] For example, a nonlinear activation function such as a ReLU function is used to process the spliced feature vector, setting elements less than 0 to 0. Then, feature dimension reduction is performed through a fully connected layer, for example, reducing the 16-dimensional spliced feature vector to 10 dimensions, and calculating the time-sequential dependence feature at each time step. The above feature embodies the time-sequential dependence relationship of the time- weighted lubrication feature vector at different time steps.

[0140] Step S2265: aggregating the dynamic trend encoding results of all lubrication property change segments to generate the sample lubricating oil property dynamic change feature.

[0141] In this embodiment, the dynamic trend encoding results obtained after processing all lubrication property change segments are spliced. For example, there are 5 lubrication property change segments in a sample monitoring sequence, and the dynamic trend encoding result of each segment is a 10-dimensional vector. The 5 10-dimensional vectors are spliced in turn to form a 50-dimensional sample lubricating oil property dynamic change feature vector, which comprehensively reflects the dynamic change of the lubricating oil property in the sample.

[0142] Step S230: performing operating condition mode matching processing on the equipment operating state data in each sample monitoring sequence to generate sample equipment operating state association features.

[0143] This step is similar to the operating condition mode matching processing of the equipment operating state data in step S123.

[0144] Step S231: extracting the load fluctuation sequence, temperature change sequence, and vibration frequency sequence from the sample equipment operating state data.

[0145] For example, for a sample monitoring sequence, the equipment load fluctuation sequence over 90 days is 450, 460, 470, 465, 455, etc.; the temperature change sequence is 55, 57, 59, 58, 56, etc.; and the vibration frequency sequence is 48, 50, 52, 51, 49, etc. These sequences record the key operating parameters of the equipment at different time points.

[0146] Step S232: performing peak detection and valley alignment processing on the load fluctuation sequence to generate a sample device load fluctuation feature.

[0147] In this embodiment, a predefined algorithm can be used to detect peaks and valleys in a load fluctuation sequence. For example, calculations may reveal peaks at 470, 465, and valleys at 450, 455, and so on. The peaks and valleys are then aligned, for example, by arranging them in chronological order, or by calculating a statistical metric such as the difference or average between the peaks and valleys to form a sample device load fluctuation signature, which is used to describe the fluctuation characteristics of the device load.

[0148] Step S233: matching the temperature change sequence with a preset temperature influence model to calculate the sample temperature correlation influence coefficient.

[0149] Step S2331: extracting historical temperature values ​​and corresponding monitoring timestamps in the temperature change sequence.

[0150] Assume that the temperature change sequence is 55, 57, 59, 58, and 56, and the corresponding monitoring timestamps are 2, 4, 6, 8, and 10 days respectively.

[0151] Step S2332: Segment the temperature change sequence into a plurality of temperature fluctuation time windows according to the reference temperature threshold and time window segmentation rule defined in the preset temperature influence model.

[0152] Assume the reference temperature threshold is 57 degrees Celsius and the time window segmentation rule is 4 days per window. Then, the first temperature fluctuation time window contains data of 55 and 57, ranging from days 2 to 5; the second window contains data of 59 and 58, ranging from days 6 to 9; and the third window contains data of 56, ranging from day 10.

[0153] Step S2333: For each temperature fluctuation time window, calculate the cumulative value of temperature deviations between all historical temperature values ​​in the window and the reference temperature threshold.

[0154] For the first temperature fluctuation time window, the accumulated temperature deviation is (55-57)+(57-57)=-2. For the second window, the accumulated temperature deviation is (59-57)+(58-57)=3. For the third window, the accumulated temperature deviation is 56-57=-1.

[0155] Step S2334: inputting the temperature deviation cumulative value corresponding to each temperature fluctuation time window into a preset nonlinear influence function for mapping to obtain the window influence factor of each temperature fluctuation time window.

[0156] Assuming the preset nonlinear influence function is f(x) = x^2 + 0.5, for the first window with a cumulative temperature deviation of -2, the resulting window influence factor after mapping is (-2)^2 + 0.5 = 4.5. For the second window with a cumulative temperature deviation of 3, the resulting window influence factor after mapping is 3^2 + 0.5 = 9.5. For the third window with a cumulative temperature deviation of -1, the resulting window influence factor after mapping is (-1)^2 + 0.5 = 1.5.

[0157] Step S2335: performing time-weighted processing on the window impact factor of each temperature fluctuation time window according to the time attenuation coefficient configured in the temperature impact model to generate a weighted window impact factor sequence.

[0158] Assuming a time decay coefficient of 0.95 and a window impact factor of 4.5 for the first window, the time weight is calculated based on the ratio of the window start time to the total monitoring time. For example, if the first window starts on the second day and the total monitoring time is 90 days, the time weight is 2 / 90. The weighted window impact factor is 4.5 × 0.95^(2 / 90) ≈ 4.49. Using this method, the window impact factor of each window is time-weighted to generate a weighted window impact factor sequence.

[0159] Step S2336: cumulatively summing the weighted window influence factor sequence to generate the sample temperature correlation influence coefficient.

[0160] Add up all the values ​​in the weighted window impact factor sequence. Assuming the weighted window impact factor sequence is [4.49, 9.45, 1.48], the cumulative summation yields the sample temperature correlation impact coefficient of 4.49 + 9.45 + 1.48 = 15.42.

[0161] Step S234: determining the correlation degree of the sample mechanical vibration frequencies according to the frequency domain energy distribution of the vibration frequency sequence.

[0162] First, the vibration frequency sequence is transformed into the frequency domain. For example, a fast Fourier transform (FFT) is used to convert the time-domain vibration frequency sequence into a frequency-domain representation. For example, if the vibration frequency sequence is 48, 50, 52, 51, and 49, the FFT transform yields the energy distribution in the frequency domain. This frequency-domain energy distribution is analyzed. For example, if a specific frequency range shows a high proportion of the total energy, this proportion is used as the frequency correlation of the sample mechanical vibration, based on pre-defined rules. For example, if the energy proportion in a key frequency range is 0.6, the frequency correlation of the sample mechanical vibration is 0.6.

[0163] The sample equipment load fluctuation characteristics, sample temperature correlation influence coefficient and sample mechanical vibration frequency correlation are integrated to form the sample equipment operation status correlation characteristics, which comprehensively reflects the various factors related to the equipment operation status and lubricating oil life in the sample.

[0164] Step S240: Input the dynamic change characteristics of the sample lubricating oil properties and the sample equipment operation status correlation characteristics into the initial lubrication life prediction model, perform multi-dimensional feature fusion through the feature cross layer, and generate a sample fusion feature vector.

[0165] The dynamic change characteristics of the sample lubricant properties and the characteristics associated with the sample equipment operating status are input into the feature cross-layer of the initial lubrication life prediction model. For example, the dynamic change characteristics of the sample lubricant properties are a 50-dimensional vector, and the characteristics associated with the sample equipment operating status are a 20-dimensional vector. The feature cross-layer can perform various fusion operations on these two feature vectors.

[0166] For example, the feature crossover layer can calculate correlations between different dimensions. For example, it can calculate the dot product between a dimension in the dynamic change feature of lubricant properties and a dimension in the feature associated with the equipment operating status, obtaining a correlation value. Feature vectors are then weighted and combined based on this correlation value. For example, a new feature vector component can be generated by weighting the viscosity change dimension in the dynamic change feature of lubricant properties with the load fluctuation dimension in the feature associated with the equipment operating status. This multi-dimensional feature crossover and fusion operation ultimately generates a sample fused feature vector that contains the correlation information between lubricant properties and equipment operating status.

[0167] Step S250: calling the regression prediction layer in the initial lubrication life prediction model to perform life prediction on the sample fusion feature vector to generate a sample lubrication life prediction value.

[0168] In this embodiment, the regression prediction layer uses the sample fusion feature vector to predict the life. Assume that the regression prediction layer adopts a linear regression model, and the linear regression model contains a weight matrix W and a bias term b. The sample fusion feature vector is an n-dimensional vector X=[x1, x2,…, xn], and the sample lubrication life prediction value y is obtained by calculating y=WX+b. Each row of the weight matrix W corresponds to the weight of a feature dimension. The above weights are randomly initialized before model training, and the bias term b is a constant vector. For example, the weight matrix W is [[w11, w12,…, w1n], [w21, w22,…, w2n], …, [wm1, wm2,…, wmn]], and the bias term b is [b1, b2,…, bm]. After matrix multiplication and addition operations, the predicted value y is obtained, which is the model's prediction result for the sample lubricant life.

[0169] Step S260: Calculate the prediction loss based on the difference between the sample lubrication life prediction value and the actual lubrication life marked value, and optimize the parameters of the initial lubrication life prediction model through back propagation until the prediction loss converges, thereby generating a trained lubrication life prediction model.

[0170] In this embodiment, the prediction loss is typically calculated using a loss function, such as the mean square error (MSE) loss function. Assuming the sample lubrication life prediction value is y_pred and the actual lubrication life annotation value is y_true, the loss value L = the average of (y_pred - y_true)^2. For example, for a sample monitoring sequence with a predicted value of 55 days and an actual annotation value of 60 days, the loss value is (55 - 60)^2 = 25. The loss values ​​calculated for all sample monitoring sequences are averaged to obtain the total prediction loss.

[0171] The parameters of the initial lubrication life prediction model are then optimized using a backpropagation algorithm. This algorithm calculates the gradients of each parameter in the lubrication life prediction model based on the predicted loss, such as the gradients of the weight matrix W and the bias term b. Based on these gradients, an optimizer (such as stochastic gradient descent (SGD), Adagrad, or Adam) is used to adjust the parameters. Using stochastic gradient descent as an example, the update formula for the weight matrix W is W = W - learning_rate * gradient(W), where learning_rate is the learning rate, which controls the step size of the parameter update, and gradient(W) is the gradient of the weight matrix W. The bias term b is updated in a similar manner.

[0172] The above process of calculating the prediction loss and backpropagating the optimized parameters is repeated until the prediction loss converges. Convergence means that the prediction loss no longer decreases significantly as training progresses, and the model parameters have reached a relatively stable state, generating a trained lubrication life prediction model. This lubrication life prediction model can more accurately predict lubricant life based on the dynamic changes in the input lubricant properties and the correlation between the equipment's operating status, providing a reliable basis for decision-making for lubrication maintenance of industrial equipment.

[0173] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a machine learning-based industrial equipment lubricant life prediction system 100, which can implement the concepts of the present application, according to some embodiments of the present application. For example, a processor 120 can be used in the machine learning-based industrial equipment lubricant life prediction system 100 to perform the functions described in the present application.

[0174] The machine learning-based industrial equipment lubricant life prediction system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the machine learning-based industrial equipment lubricant life prediction method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0175] For example, the machine learning-based industrial equipment lubricant life prediction system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the machine learning-based industrial equipment lubricant life prediction system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented according to the above-mentioned program instructions. The machine learning-based industrial equipment lubricant life prediction system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0176] For ease of explanation, only one processor is described in the machine learning-based industrial equipment lubricant life prediction system 100. However, it should be noted that the machine learning-based industrial equipment lubricant life prediction system 100 in this application can also include multiple processors, so the steps performed by one processor described in this application can also be performed jointly or individually by multiple processors. For example, if the processor of the machine learning-based industrial equipment lubricant life prediction system 100 executes step A and step B, it should be understood that step A and step B can also be executed jointly by two different processors or individually in one processor. For example, the first processor executes step A and the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0177] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned method for predicting the life of lubricating oil of industrial equipment based on machine learning is implemented.

[0178] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for predicting the life of lubricating oil in industrial equipment based on machine learning, characterized in that: The method comprises: Acquire a lubricating oil state monitoring data set for industrial equipment within a preset monitoring period, the lubricating oil state monitoring data set comprising a plurality of lubricating state monitoring sequences, each lubricating state monitoring sequence comprising lubricating oil property indicator data and corresponding equipment operating state data at at least one lubricating state sampling moment; Performing dynamic feature extraction processing on the lubricating oil state monitoring data set to obtain dynamic change features of lubricating oil properties and equipment operation state correlation features of each lubricating oil state monitoring sequence; Based on a preset lubrication life prediction model, the dynamic change characteristics of the lubricating oil properties and the associated characteristics of the equipment operation status are integrated and predicted to generate a predicted value of the lubricating oil life of the lubrication status monitoring sequence; generating a lubrication maintenance decision instruction based on the lubricating oil life prediction value and a preset lubrication life threshold, and feeding back the lubrication maintenance decision instruction to the industrial equipment maintenance terminal to trigger a lubrication maintenance operation; The method of generating a lubricating oil life prediction value for the lubricating state monitoring sequence by fusing and predicting the dynamic change characteristics of the lubricating oil properties and the associated characteristics of the equipment operating state based on a preset lubricating life prediction model includes: Dividing the dynamic change characteristics of the lubricating oil properties into a viscosity change feature subset, an acid value accumulation feature subset, and an impurity fluctuation feature subset; Dividing the equipment operation state related features into a load related feature subset, a temperature related feature subset and a vibration related feature subset; Calling the feature cross layer in the lubrication life prediction model to cross-combine the viscosity change feature subset and the load correlation feature subset to generate a first cross feature vector; performing nonlinear mapping processing on the acid value cumulative feature subset and the temperature associated feature subset to generate a second cross feature vector; Performing attention weight allocation processing on the impurity fluctuation feature subset and the vibration association feature subset to generate a third cross feature vector; fusing the first cross feature vector, the second cross feature vector, and the third cross feature vector to generate a multi-dimensional lubricating oil feature set; The regression prediction layer in the lubrication life prediction model is called to perform life prediction on the multi-dimensional lubricating oil feature set to generate the lubricating oil life prediction value.

2. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 1, characterized in that: The dynamic feature extraction processing is performed on the lubricating oil state monitoring data set to obtain the dynamic change characteristics of the lubricating oil properties and the equipment operation state correlation characteristics of each lubricating state monitoring sequence, including: Performing time series segmentation processing on the lubricating oil property index data in the lubrication state monitoring sequence to obtain a plurality of lubrication property change segments; calling a preset lubrication characteristic encoder to perform dynamic trend encoding processing on the multiple lubrication property change segments to generate dynamic change characteristics of lubricating oil properties for the lubrication state monitoring sequence, wherein the dynamic change characteristics of lubricating oil properties include a lubricating oil viscosity change rate, an acid value cumulative gradient, and an impurity concentration fluctuation trend; The equipment operating status data in the lubrication status monitoring sequence is subjected to operating mode matching processing to obtain equipment operating status correlation characteristics, wherein the equipment operating status correlation characteristics include at least one of the following: equipment load fluctuation characteristics, temperature correlation influence coefficient, and mechanical vibration frequency correlation.

3. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 2, characterized in that: The calling of a preset lubrication characteristic encoder to perform dynamic trend encoding processing on the multiple lubrication property change segments to generate dynamic change characteristics of lubricating oil properties of the lubrication state monitoring sequence includes: For each lubrication property change segment, extracting an initial property indicator sequence and corresponding monitoring time interval data in the lubrication property change segment; Performing local fluctuation analysis on the initial attribute indicator sequence based on a sliding time window to generate multiple local fluctuation feature subsequences; Calling a convolutional neural network layer in a lubrication feature encoder to perform convolution processing on the local fluctuation feature subsequence to obtain a primary lubrication feature vector; Obtaining a weight adjustment coefficient corresponding to the monitoring time interval data, wherein the weight adjustment coefficient is in a nonlinear inverse proportional relationship with the duration of the monitoring time interval data; Multiplying the weight adjustment coefficient by the primary lubrication feature vector element by element to obtain a time-weighted lubrication feature vector; Calling a bidirectional long short-term memory network layer in a lubrication feature encoder to perform temporal dependency modeling on the time-weighted lubrication feature vector to generate a dynamic trend encoding result of the lubrication attribute change segment; The dynamic trend coding results of all lubrication property change segments are aggregated to generate the dynamic change characteristics of the lubricating oil properties of the lubrication state monitoring sequence.

4. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 3, characterized in that: The calling of the bidirectional long short-term memory network layer in the lubrication feature encoder to perform temporal dependency modeling on the time-weighted lubrication feature vector to generate a dynamic trend encoding result of the lubrication attribute change segment includes: Inputting the time-weighted lubrication feature vector into the forward processing branch of the bidirectional long short-term memory network layer, performing forward sequence processing on the time-weighted lubrication feature vector in chronological order to generate a forward hidden state sequence; Synchronously inputting the time-weighted lubrication feature vector into the backward processing branch of the bidirectional long short-term memory network layer, performing backward sequence processing on the time-weighted lubrication feature vector in reverse time order, and generating a backward hidden state sequence; Performing cross-sequence feature alignment on the forward hidden state sequence and the backward hidden state sequence to obtain a concatenated feature vector of the forward hidden state and the backward hidden state at each time step; Performing nonlinear activation and feature dimensionality reduction processing on the concatenated feature vector to generate a time-dependent feature of the time-weighted lubrication feature vector at a corresponding time step; Aggregating the temporal dependency features of all time steps to generate a dynamic trend encoding result of the lubrication property change segment.

5. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 2, characterized in that: The performing of operating mode matching processing on the equipment operating status data in the lubrication status monitoring sequence to obtain equipment operating status correlation features includes: extracting a load fluctuation sequence, a temperature change sequence, and a vibration frequency sequence from the equipment operation status data; Performing peak detection and valley alignment processing on the load fluctuation sequence to generate a device load fluctuation feature; Matching the temperature change sequence with a preset temperature influence model to calculate a temperature correlation influence coefficient, wherein the temperature correlation influence coefficient is used to characterize the cumulative influence of temperature changes on lubricating oil property indicators; The mechanical vibration frequency correlation is determined according to the frequency domain energy distribution of the vibration frequency sequence, and the mechanical vibration frequency correlation is used to quantify the accelerated attenuation effect of the mechanical vibration on the life of the lubricating oil.

6. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 1, characterized in that: The calling of the feature cross layer in the lubrication life prediction model to cross-combine the viscosity change feature subset and the load correlation feature subset to generate a first cross feature vector includes: extracting a viscosity gradient sequence from the viscosity change feature subset and a load fluctuation amplitude sequence from the load association feature subset; Calculating a dynamic correlation coefficient matrix between the viscosity gradient sequence and the load fluctuation amplitude sequence; Performing weighted correction on the viscosity gradient sequence based on the dynamic correlation coefficient matrix to obtain a corrected viscosity gradient characteristic vector; Performing feature concatenation on the corrected viscosity gradient eigenvector and the load fluctuation amplitude sequence to generate a first cross eigenvector; Furthermore, performing nonlinear mapping processing on the acid value cumulative feature subset and the temperature-related feature subset to generate a second cross feature vector includes: Acquire the acid value accumulation rate data in the acid value accumulation feature subset and the temperature influence coefficient in the temperature correlation feature subset; Inputting the acid value accumulation rate data into a preset exponential mapping function to obtain a nonlinear acid value accumulation characteristic; Multiplying the nonlinear acid value cumulative characteristic by the temperature influence coefficient element by element to obtain a temperature-weighted acid value characteristic vector; Calling a fully connected layer in a lubrication life prediction model to perform dimensionality reduction processing on the temperature-weighted acid value feature vector to generate a second cross feature vector; Furthermore, performing attention weight allocation processing on the impurity fluctuation feature subset and the vibration association feature subset to generate a third cross feature vector includes: extracting an impurity concentration fluctuation sequence from the impurity fluctuation feature subset and a vibration frequency correlation degree from the vibration correlation feature subset; Invoking the attention mechanism layer in the lubrication life prediction model to calculate the attention weight distribution between the impurity concentration fluctuation sequence and the vibration frequency correlation; Dynamically weighting the impurity concentration fluctuation sequence according to the attention weight distribution to generate a weighted impurity fluctuation feature; The weighted impurity fluctuation feature and the vibration frequency correlation are subjected to feature fusion to generate a third cross feature vector.

7. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 1, characterized in that: The generating of the lubrication maintenance decision instruction according to the lubricating oil life prediction value and the preset lubrication life threshold comprises: When the predicted lubricating oil life value is less than a first lubricating oil life threshold, generating a first lubricating maintenance decision instruction, wherein the first lubricating maintenance decision instruction is used to instruct to immediately perform a lubricating oil replacement operation; When the predicted value of the lubricating oil life is between a first lubricating oil life threshold and a second lubricating oil life threshold, generating a second lubricating maintenance decision instruction, the second lubricating maintenance decision instruction is used to trigger an operation of increasing the frequency of real-time monitoring of the lubricating oil state; When the predicted value of the lubricating oil life is greater than a second lubricating oil life threshold, a third lubricating maintenance decision instruction is generated, and the third lubricating maintenance decision instruction is used to maintain the current lubricating maintenance strategy and output remaining life prompt information.

8. The method for predicting the life of lubricating oil for industrial equipment based on machine learning according to claim 1, characterized in that: The training process of the lubrication life prediction model includes: Acquire a sample monitoring data set, the sample monitoring data set comprising a plurality of sample monitoring sequences, each sample monitoring sequence consisting of lubricating oil property index data, equipment operating status data, and corresponding actual lubrication life annotation values ​​arranged in chronological order; Performing dynamic trend coding processing on the lubricating oil property index data in each sample monitoring sequence to generate dynamic change characteristics of the sample lubricating oil properties; Performing operating mode matching processing on the equipment operating status data in each sample monitoring sequence to generate sample equipment operating status correlation features; Input the dynamic change characteristics of the sample lubricating oil properties and the sample equipment operation status correlation characteristics into the initial lubrication life prediction model, perform multi-dimensional feature fusion through the feature cross layer, and generate a sample fusion feature vector; Calling the regression prediction layer in the initial lubrication life prediction model to perform life prediction on the sample fusion feature vector to generate a sample lubrication life prediction value; The prediction loss is calculated based on the difference between the sample lubrication life prediction value and the actual lubrication life marked value, and the parameters of the initial lubrication life prediction model are optimized by back propagation until the prediction loss converges to generate a trained lubrication life prediction model.

9. A machine learning-based industrial equipment lubricant life prediction system, characterized in that: The industrial equipment lubricant oil life prediction system based on machine learning includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the industrial equipment lubricant oil life prediction method based on machine learning as described in any one of claims 1 to 8 above.

Images