Filter cartridge life prediction method, electronic device, and medium

Abstract

The present application relates to the technical field of strip mining machine, and particularly relates to a filter core life prediction method, an electronic device and a medium. The method constructs state information of the engineering machinery hydraulic circuit in a first preset time period into a plurality of sample features (each sample feature represents an operating state in a single preset unit time), and trains an original model to obtain a life prediction model, so that the model can learn the degradation law and time sequence characteristics of the filter core under different working conditions from historical operation samples. In actual application, real-time state information in a second preset time period is collected and input into the life prediction model, and the remaining service life of the filter core after the second preset time period is directly output.

Description

Technical Field

[0001] This invention relates to the field of open-pit mining machine technology, specifically to a filter cartridge life prediction method, electronic device, and medium. Background Technology

[0002] Among the many types of construction machinery, many (such as open-pit mining equipment) operate under high loads for extended periods in harsh environments with high dust levels, strong winds and sandstorms, and large temperature differences between day and night. The filter elements (i.e., return oil filters) in the hydraulic system's oil return lines are prone to clogging due to their continuous task of trapping particles and moisture, thus limiting their service life. Currently, filter replacement is generally done through "periodic replacement" or "single differential pressure threshold alarm to prompt replacement," which fails to consider the complexities of the work environment and diverse operating conditions, easily leading to premature or delayed replacement. Therefore, intelligently, scientifically, and accurately estimating the remaining lifespan of the filter elements is crucial for helping technicians determine when to replace them. Summary of the Invention

[0003] This invention provides a filter cartridge life prediction method, electronic device, and medium, aiming to intelligently, scientifically, and accurately predict the remaining life of the filter cartridge.

[0004] In a first aspect, embodiments of this application provide a method for predicting filter element lifespan, wherein the filter element is located in the return oil line of the hydraulic circuit of engineering machinery, and the method includes:

[0005] The state information of the hydraulic circuit within a first preset time period is obtained, wherein the state information within the first preset time period is used to characterize the operating state of the hydraulic circuit within the first preset time period.

[0006] Multiple sample features are constructed using the state information within the first preset time period, wherein each sample feature is used to characterize the operating state of the hydraulic circuit within a single preset unit of time.

[0007] The original model is trained using the features of the multiple samples to obtain a lifespan prediction model.

[0008] The state information of the hydraulic circuit within a second preset time period is obtained, and the state information within the second preset time period is input into the life prediction model to obtain the remaining service life of the filter element after the second preset time period output by the life prediction model. The state information within the second preset time period is used to characterize the operating state of the hydraulic circuit within the second preset time period.

[0009] In some implementations, the status information within the first preset time period includes the pressure difference between the upstream and downstream pressures of the filter element within the first preset time period and the oil temperature of the return oil pipeline within the first preset time period.

[0010] The construction of multiple sample features using state information within the first preset time period includes:

[0011] Preprocess the status information within the first preset time period;

[0012] The state information within the first preset time period after preprocessing is segmented according to the preset unit duration to obtain the multiple sample features;

[0013] The preprocessing includes:

[0014] For each time point within the first preset time period, the oil viscosity at that time point is determined based on the oil temperature at that time point.

[0015] The oil viscosity is normalized based on the pressure difference at the specified time point.

[0016] In some embodiments, the life prediction model includes a working condition classification sub-model, a load spectrum prediction sub-model, and a backbone prediction sub-model. The model output of the working condition classification sub-model includes the working condition category of the construction machinery at each time point within the second preset time period. The model output of the load spectrum prediction sub-model includes the load spectrum of the construction machinery at multiple time points after the second preset time period. The backbone prediction sub-model is used to determine the remaining service life of the filter element after the second preset time period based on the state information within the second preset time period, the model output of the working condition classification sub-model, and the model output of the load spectrum prediction sub-model.

[0017] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life.

[0018] The model training includes multiple iteration cycles, and the model corresponding to each iteration cycle is the target model of that iteration cycle.

[0019] The step of training the original model using the multiple sample features includes:

[0020] For each iteration cycle, the loss value of the iteration cycle is determined using the target loss function, and the model parameters of the target model for the iteration cycle are updated iteratively based on the loss value.

[0021] Wherein, the loss value of the target loss function is the sum of the first loss value, the second loss value, the third loss value and the fourth loss value, the first loss value is the product of the first weight and the loss value of the first loss function, the first loss function is used to determine the loss value of the first original sub-model in each iteration cycle, and the first original sub-model corresponds to the working condition classification sub-model;

[0022] The second loss value is the product of the second weight and the loss value of the second loss function. The second loss function is used to determine the loss value of the second original sub-model in each iteration cycle. The second original sub-model corresponds to the load spectrum prediction sub-model.

[0023] The third loss value is the product of the third weight and the loss value of the third loss function. The third loss function is used to determine the loss value of the third original sub-model corresponding to the specific value of the remaining lifetime in each iteration cycle. The third original sub-model corresponds to the backbone prediction sub-model.

[0024] The fourth loss value is the product of the fourth weight and the loss value of the fourth loss function, which is used to determine the loss value of the third original sub-model corresponding to the Weibull distribution parameters of the specific value of the remaining lifetime in each iteration cycle.

[0025] In some implementations, the first loss function is based on cross-entropy loss, the second loss function is based on quantile loss or mean squared error loss, the third loss function is based on Huber loss, and the fourth loss function includes a first sub-loss function and a second sub-loss function;

[0026] The first sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the first sample, and the second sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the second sample. The first sub-loss function is the negative log-likelihood of the probability density function of the Weibull distribution, and the second sub-loss function is the negative log-likelihood of the survival function of the Weibull distribution.

[0027] The first sample is the sample feature corresponding to a preset unit time period during which the filter element has failed among the plurality of sample features;

[0028] The second sample is the sample feature corresponding to a preset unit time during which the filter element has not failed, among the multiple sample features.

[0029] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life.

[0030] After obtaining the remaining service life of the filter element after the second preset time period as output by the service life prediction model, the method further includes:

[0031] Based on the Weibull distribution parameters of the specific remaining lifetime value, the predicted remaining lifetime values ​​corresponding to multiple preset quantiles are determined.

[0032] In some embodiments, after determining the remaining lifetime prediction values ​​corresponding to a plurality of preset quantiles, the method further includes:

[0033] Among the remaining lifetime prediction values ​​corresponding to the plurality of preset quantiles, determine the remaining lifetime prediction value corresponding to the first quantile and the remaining lifetime prediction value corresponding to the second quantile, wherein the first quantile is greater than the second quantile;

[0034] Based on the remaining life prediction value corresponding to the first quantile, the remaining life prediction value corresponding to the second quantile, and the current time point, a recommended replacement time period is determined, which is used to indicate the time period for replacing the filter element.

[0035] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life.

[0036] After obtaining the remaining service life of the filter element after the second preset time period as output by the service life prediction model, the method further includes:

[0037] The pressure difference between the upstream and downstream pressures of the filter element is obtained, and the obtained pressure difference between the upstream and downstream pressures of the filter element is determined as the current pressure difference.

[0038] Based on the Weibull distribution parameters of the specific value of the remaining lifespan, determine whether the probability that the remaining lifespan of the filter element is less than or equal to a preset lifespan threshold is greater than or equal to a preset probability threshold.

[0039] If the probability that the remaining lifespan of the filter element is less than or equal to the preset lifespan threshold is greater than or equal to the preset probability threshold, or if the current pressure difference meets the first preset condition, or if the current pressure difference meets the second preset condition, a filter element replacement prompt will be output.

[0040] Wherein, the first preset condition is that the current pressure difference is greater than or equal to the first difference value, and the first difference value is the difference obtained by subtracting the preset buffer value from the preset pressure difference threshold;

[0041] The second preset condition is that the current pressure difference is greater than or equal to the opening threshold of the bypass valve of the return oil pipeline.

[0042] Secondly, embodiments of this application provide an electronic device, the device including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0043] Memory, used to store computer programs;

[0044] When a processor executes a program stored in a memory, it implements the steps of the filter life prediction method provided in the first aspect of the embodiments of this application.

[0045] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the filter life prediction method provided in the first aspect of embodiments of this application.

[0046] In this embodiment, the state information of the hydraulic circuit of the construction machinery within a first preset time period is constructed into several sample features (each sample feature represents the operating state within a single preset unit of time). This is used to train the original model to obtain a life prediction model. The model can learn the degradation patterns and temporal characteristics of the filter element under different operating conditions from historical operating samples. In practical applications, real-time state information within a second preset time period is collected and input into the life prediction model, directly outputting the remaining service life of the filter element after the second preset time period. Thus, by training the model based on historical sample features, the model is given the ability to identify filter element degradation trends and estimate remaining life under complex operating conditions. By using the real-time state of the second preset time period as input to predict the remaining service life of the filter element after the second preset time period, intelligent, scientific, and accurate remaining life prediction is achieved. This helps technicians make intelligent replacement decisions and avoids blindly premature or delayed replacement. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic flowchart of the filter life prediction method provided in the embodiments of this application;

[0049] Figure 2 This is another flowchart illustrating the filter life prediction method provided in this application embodiment;

[0050] Figure 3This is a schematic diagram of the filter life prediction device provided in the embodiments of this application;

[0051] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0053] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0054] The filter life prediction method, device, electronic equipment, and medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0055] Those skilled in the art will understand that the filter life prediction method provided in the embodiments of this application can be executed by a single or distributed electronic device (the electronic device can be arranged in engineering machinery, or can be connected to the engineering machinery by communication or electricity). There can be one or more processors. In the case of multiple processors, the multiple processors can be connected by electricity or communication and work together as modules with different functions to execute the filter life prediction method provided in the embodiments of this application.

[0056] Figure 1 This is a flowchart illustrating the filter life prediction method provided in this application embodiment, as shown below. Figure 1 As shown, the first aspect of this application provides a method for predicting filter element life, wherein the filter element is located in the return oil line of the hydraulic circuit of engineering machinery, and includes the following steps S100-S400:

[0057] Step S100: Obtain the status information of the hydraulic circuit within a first preset time period, wherein the status information within the first preset time period is used to characterize the operating status of the hydraulic circuit within the first preset time period.

[0058] In this step, the processor aims to acquire historical operational evidence for model learning. The processor can record the status information in real time in a time series using field sensors (such as upstream / downstream pressure of the filter element, instantaneous flow rate of the circuit, oil temperature, bypass valve status, pump speed, etc.) as the status information within the first preset time period. It ensures that this information can reflect the typical working mode and abnormal events of the hydraulic circuit within this time period, providing raw materials for subsequent sample construction.

[0059] For example, the processor can set the first preset time period as the operation record of the past 30 days, with a sampling frequency of one record every 10 seconds, thereby obtaining approximately 30 × 24 × 3600 / 10 = 259,200 original time samples as the status information within the first preset time period. To facilitate subsequent processing, the processor can record and save the timestamp of each sample and the instantaneous readings of the aforementioned channels to form a complete status information sequence that can represent the distribution of the hydraulic circuit's operating status and sudden working conditions over the past 30 days.

[0060] Step S200: Construct multiple sample features using the state information within the first preset time period, wherein each sample feature is used to characterize the operating state of the hydraulic circuit within a single preset unit of time.

[0061] In this step, the state information within the first preset time period is used to construct multiple sample features. Each sample feature is used to characterize the operating state of the hydraulic circuit within a single preset unit of time. Specifically, the processor divides the long-term sequence into several segments according to a preset unit of time (e.g., 1 hour or 24 hours), and calculates representative time-series or statistical features for each segment, such as the average normalized differential pressure, differential pressure standard deviation, cumulative flow, bypass opening ratio, and average oil temperature within that segment, thereby mapping the high-frequency raw signal into a feature vector that is easy for model input.

[0062] For example, continuing with the above scenario of 30 days and 10 seconds of sampling, if the preset unit duration is 1 hour, then there are 360 ​​sampling points per hour, and 30 days can construct 30 × 24 = 720 sample features; for the sample features of the 100th hour, the average differential pressure is 0.12 MPa, the standard deviation of differential pressure is 0.02 MPa, the cumulative flow is 7,200 L, and the bypass opening percentage is 0.02 (i.e. 2%). These vectors are used as a sample feature to represent the operating status of that hour.

[0063] Step S300: Use multiple sample features to train the original model to obtain the life prediction model.

[0064] In this step, the processor uses multiple sample features to train the original model to obtain a lifespan prediction model. Specifically, the constructed sample features need to be paired with corresponding labels (such as the actual remaining lifespan or core replacement time obtained from maintenance records, and censoring labels for samples still in service) to form training and validation sets. These samples are then fed into the original model for parameter learning until a trained lifespan prediction model is obtained.

[0065] For example, the processor divides the 720 sample features obtained in the previous step into a training set of 576 (80%) and a validation set of 144 (20%) based on time or device distribution. If there are 120 historically recorded actual replacement events, they are used as event labels, and the rest are right-censored samples. The model is trained with 32 samples per batch and 100 epochs per batch. The training objective can be to minimize the censored likelihood term and auxiliary regression term. After training, the life prediction model will learn to map the input sample features to the distribution parameters or point estimates of the filter cartridge's remaining life, thereby enabling it to perform inferences on new observations.

[0066] Step S400: Obtain the status information of the hydraulic circuit within the second preset time period, and input the status information within the second preset time period into the life prediction model to obtain the remaining service life of the filter element after the second preset time period output by the life prediction model. The status information within the second preset time period is used to characterize the operating status of the hydraulic circuit within the second preset time period.

[0067] In this step, the trained life prediction model is used for online inference. The processor first acquires the state information of the hydraulic circuit within a second preset time period, then inputs this state information into the life prediction model to obtain the remaining lifespan of the filter element after the second preset time period, as output by the model. Specifically, the second preset time period is generally a short historical period (e.g., the last 24 hours) prior to the application time (the current time point). Its state information also needs to be constructed into a sample feature vector in the same way as during training and input into the model. The model can return the distribution parameters of the remaining lifespan (e.g., k and λ of a Weibull distribution) or a direct quantile lifespan estimate.

[0068] For example, when a piece of construction machinery is running in real time, it collects data from the most recent 24 hours and constructs 24 sample feature vectors by hour. After inputting the data into a trained life prediction model, the model outputs Weibull parameters k=1.8 and λ=150 hours. Based on this distribution, the median life can be calculated to be approximately 120 hours. Therefore, the system will provide an estimate of the median remaining life of the filter element for approximately 120 hours from the current time, and can generate replacement suggestions or warnings accordingly, thereby realizing intelligent replacement decisions based on the actual working conditions of a second preset time period.

[0069] Through steps S100-S400, the processor constructs several sample features (each sample feature representing the operating state within a single preset unit of time) from the state information of the hydraulic circuit of the engineering machinery within a first preset time period. This feature is then used to train the original model to obtain a life prediction model. The model learns the degradation patterns and temporal characteristics of the filter element under different operating conditions from historical operating samples. In practical applications, real-time state information within a second preset time period is collected and input into the life prediction model, directly outputting the remaining service life of the filter element after the second preset time period. Thus, by training the model based on historical sample features, the model gains the ability to identify filter element degradation trends and estimate remaining life under complex operating conditions. By using the real-time state of the second preset time period as input to predict the remaining service life of the filter element after the second preset time period, intelligent, scientific, and accurate remaining life prediction is achieved. This helps technicians make intelligent replacement decisions and avoids blindly premature or delayed replacement.

[0070] Figure 2 This is another flowchart illustrating the filter life prediction method provided in this application embodiment; please refer to it as well. Figures 1-2 In some implementations, the status information within the first preset time period includes the pressure difference between the upstream and downstream pressures of the filter element within the first preset time period and the oil temperature of the return oil line within the first preset time period.

[0071] Multiple sample features are constructed using state information within a first preset time period, including:

[0072] Preprocess the status information within the first preset time period;

[0073] The state information within the first preset time period after preprocessing is segmented according to a preset unit duration to obtain multiple sample features;

[0074] Preprocessing includes:

[0075] For each time point within the first preset time period, the oil viscosity at that time point is determined based on the oil temperature at that time point.

[0076] The oil viscosity was normalized based on the pressure difference at a given time point.

[0077] In this embodiment, the processor first directly calculates the instantaneous pressure difference across the filter element based on the original timing measurements of the hydraulic circuit:

[0078]

[0079] in, For discrete-time indexing, and They are time points (Upstream and downstream pressure of the filter element). For pressure difference;

[0080] The processor then uses this pressure difference, along with the oil temperature in the return line at the same time and other information, as the status information for the first preset time period.

[0081] The processor can then use filters such as Savitzky-Golay in the preprocessing stage to first denoise and smooth the original pressure difference sequence:

[0082]

[0083] in, For the smoothed pressure difference, The width is half the width of the window. For the order of the polynomial, This represents the total number of points in the window.

[0084] The processor then calculates the oil temperature at each time point after smoothing and noise reduction. The dynamic viscosity at that moment was calculated using the calibrated oil viscosity model. The viscosity model uses a reference viscosity. Reference temperature And with the viscosity coefficient E as a parameter, the formulaic relationship is expressed as follows:

[0085]

[0086] In obtaining Then, the smoothed pressure difference at each moment is normalized according to the viscosity ratio, specifically using the formula:

[0087]

[0088] Calculate the normalized pressure difference The meaning of this formula is to convert the pressure difference under high temperature or low temperature conditions to the reference state corresponding to the reference viscosity, thereby reducing the impact of temperature fluctuations on the pressure difference reading as a degradation indicator.

[0089] After the processor completes time-level normalization, it then segments and statistically analyzes the normalized pressure difference and concurrent oil temperature signals within the first preset time period according to preset unit durations. It constructs a sample feature vector from the time series quantities or statistical quantities (such as mean, standard deviation, cumulative flow, bypass opening ratio, etc.) within each preset unit duration. For example:

[0090]

[0091] in, K represents a multidimensional feature sequence within a preset unit time period, where K is the preset unit time period. For loop flow, Oil temperature Hydraulic pump speed This refers to the valve position in the hydraulic circuit. This refers to the hydraulic pump power and the open / closed status of the bypass valve. ;

[0092] These sample feature vectors together constitute the dataset used to train the original model. During training, the original model uses these sample features as input and combines them with the corresponding lifetime / censoring labels to learn parameters. After training, the lifetime prediction model can receive the state features obtained by the same preprocessing and unit segmentation within the second preset time period in the application stage and output the remaining lifetime of the filter element or the corresponding lifetime distribution parameters after the second preset time period.

[0093] Thus, by normalizing the differential pressure based on the temperature-viscosity physical model and constructing sample features based on unit time, and using these features to train the life prediction model, the interference of ambient temperature fluctuations on the differential pressure signal can be effectively reduced, and the accuracy of identifying the filter element degradation trend under complex working conditions can be improved. This enables more scientific and timely filter element replacement decisions, reducing premature or delayed replacements caused by blind periodic replacements or single threshold alarms.

[0094] In some implementations, the life prediction model includes a working condition classification sub-model, a load spectrum prediction sub-model, and a backbone prediction sub-model. The model output of the working condition classification sub-model includes the working condition category of the construction machinery at each time point within a second preset time period. The model output of the load spectrum prediction sub-model includes the load spectrum of the construction machinery at multiple time points after the second preset time period. The backbone prediction sub-model is used to determine the remaining service life of the filter element after the second preset time period based on the state information within the second preset time period, the model output of the working condition classification sub-model, and the model output of the load spectrum prediction sub-model.

[0095] In this embodiment, the life prediction model is constructed as a combination of three functional sub-modules: a working condition classification sub-model, a load spectrum prediction sub-model, and a backbone prediction sub-model. The system uses the state information within a second preset time period as the basic input. First, the working condition classification sub-model outputs the working condition category (i.e., provides labels such as "loaded / walking / standby" for each time point within the second preset time period). The load spectrum prediction sub-model predicts the load spectrum vector (e.g., a sequence of flow rate, temperature, or other load-related quantities for future times) for several times after the second preset time period. The backbone prediction sub-model takes the outputs from these two sub-models and the state information within the second preset time period as conditional inputs, and after time-series feature fusion, outputs the remaining lifespan of the filter element after the second preset time period (which can be represented as a point estimate or a probability distribution parameter).

[0096] Regarding the fusion mechanism, the backbone prediction sub-model can employ a self-attention type temporal network to achieve interactive weighting between historical time series, load condition embedding, and future load conditions. The attention calculation can be expressed by the formula:

[0097]

[0098] Where Q (query), K (key), and V (value) represent the query, key, and value tensors obtained from the mapping of the input sequence, respectively. Let be the dimension of the key vector. This is used to normalize similarity into attention weights;

[0099] The formula means that the similarity between the query and each candidate information is first calculated (i.e., ... Then scale to stabilize the value (divide by). The similarity is normalized and then weighted and summed into a vector V, resulting in a fusion representation that "weights attention" to historical moments and conditional information. The main prediction sub-model further regresses or parameterizes the remaining lifespan of the filter cartridge based on this fusion representation.

[0100] For the probabilistic output of lifetime, the backbone prediction sub-model can use the Weibull distribution parameterization to express the uncertainty of remaining lifetime. Overall, during model training, a multi-task joint optimization strategy can be used to train both the load classification sub-model and the load spectrum sub-model. The backbone prediction sub-model is also optimized using Weibull likelihood (negative log-likelihood of density function for observed fault samples and negative log-likelihood of survival function for censored samples) and necessary auxiliary regression loss. This allows the load, future load, and lifetime output to mutually promote each other and converge together in terms of representation.

[0101] Thus, by integrating the conditional information provided by the operating condition classification sub-model and the load spectrum prediction sub-model with the attention mechanism of the backbone prediction sub-model, a probabilistic estimate of the remaining life of the filter element can be given under the condition of considering the uncertainty of historical operating conditions and future loads. This improves the accuracy of identifying the degradation trend of the filter element under complex operating conditions and provides a quantifiable confidence basis for replacement decisions.

[0102] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to that specific value.

[0103] Model training includes multiple iteration cycles, and the model corresponding to each iteration cycle is the target model for that iteration cycle.

[0104] The original model is trained using multiple sample features, including:

[0105] For each iteration cycle, the loss value of the iteration cycle is determined using the target loss function, and the model parameters of the target model for the iteration cycle are updated iteratively based on the loss value.

[0106] The target loss function has a loss value of the sum of the first loss value, the second loss value, the third loss value, and the fourth loss value. The first loss value is the product of the first weight and the loss value of the first loss function. The first loss function is used to determine the loss value of the first original sub-model in each iteration cycle. The first original sub-model corresponds to the working condition classification sub-model.

[0107] The second loss value is the product of the second weight and the loss value of the second loss function. The second loss function is used to determine the loss value of the second original sub-model in each iteration cycle. The second original sub-model corresponds to the load spectrum prediction sub-model.

[0108] The third loss value is the product of the third weight and the loss value of the third loss function. The third loss function is used to determine the loss value of the third original sub-model corresponding to the specific value of the remaining lifetime in each iteration cycle. The third original sub-model corresponds to the backbone prediction sub-model.

[0109] The fourth loss value is the product of the fourth weight and the loss value of the fourth loss function. The fourth loss function is used to determine the loss value of the third original sub-model corresponding to the Weibull distribution parameters of the specific remaining lifetime value in each iteration cycle.

[0110] In some implementations, the first loss function is based on cross-entropy loss, the second loss function is based on quantile loss or mean squared error loss, the third loss function is based on Huber loss, and the fourth loss function includes the first sub-loss function and the second sub-loss function.

[0111] The first sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the first sample, and the second sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the second sample. The first sub-loss function is the negative log-likelihood of the probability density function of the Weibull distribution, and the second sub-loss function is the negative log-likelihood of the survival function of the Weibull distribution.

[0112] The first sample is the sample feature corresponding to the preset unit time when the filter element has failed among multiple sample features;

[0113] The second sample is the sample feature corresponding to a preset unit time during which the filter element has not failed, among multiple sample features.

[0114] Combining the two implementation methods above, the processor characterizes the remaining lifetime using remaining lifetime information, which includes both the specific value of the remaining lifetime (point estimate) and the Weibull distribution parameters describing the uncertainty of the specific value of the remaining lifetime.

[0115] The training of the original model is organized by the processor into several iteration cycles (e.g., several epochs). The model corresponding to each iteration cycle is the target model. In each iteration cycle, the performance of the current target model is evaluated by the target loss function and the model parameters are updated accordingly.

[0116] Specifically, the target loss function consists of a weighted sum of four parts: the first loss value is the loss value of the first weight multiplied by the first loss function, where the first loss function is used to measure the classification error corresponding to the working condition classification sub-model (which can be a Transformer-Encoder), and can be in the form of cross-entropy.

[0117]

[0118] in, For at any time No. The true one-hot probability (target probability) of similar working conditions. This represents the predicted probability of the corresponding class at that time point, as output by the model. Iterate through all work condition categories;

[0119] The second loss value is the product of the second weight and the second loss function. The second loss function is used to measure the load spectrum prediction sub-model (the load spectrum prediction sub-model can be a Transformer, used to predict the load spectrum vector sequence for the next H time steps). , Indicates flow rate or other continuous load. Indicates temperature. The sequence prediction error (representing other components related to the load) can be trained using pinball loss or mean squared error loss as the training objective, and its form is as follows:

[0120] Mean squared error loss:

[0121] Quantile loss:

[0122] in, This represents the true load spectrum vector; The predicted load spectrum vector, For the number of moments in the foreseeable future; quantiles; The residuals are from the quantile regression.

[0123] For the first The predicted residual of the step, For the desired quantile (e.g., 0.5 represents the median), this asymmetric linear penalty can be used to directly learn the prediction results at different confidence levels;

[0124] The third loss value is the product of the third weight and the third loss function. The third loss function measures the regression error of the backbone prediction sub-model with respect to the specific value of remaining lifetime, and a robust Huber loss can be used.

[0125]

[0126] in, This represents the actual remaining lifespan of the filter cartridge. The specific value of remaining lifetime output by the main prediction sub-model;

[0127] The fourth loss value is the product of the fourth weight and the fourth loss function. The fourth loss function measures the fitting error of the backbone prediction sub-model to the probability distribution parameters (Weibull distribution parameters) of the specific remaining lifetime value. This fourth loss function consists of two sub-loss functions (the first sub-loss function and the second sub-loss function):

[0128] For the sample of the filter cartridge that has failed (the first sample), the negative log-likelihood of the probability density of the Weibull distribution is applied:

[0129]

[0130] in, The parameters of the Weibull distribution in the remaining lifetime information. Let be the probability density function of the Weibull distribution, with parameters... For shape parameters, For scale parameters, This represents the actual remaining lifespan.

[0131] The negative log-likelihood of the Weibull survival function was applied to the non-failed filter sample (the second sample):

[0132]

[0133] Among them, the survival function This represents the probability of surviving beyond time r.

[0134] During training, each iteration first calculates the first to fourth losses (corresponding to load condition classification, load spectrum prediction, lifetime point estimation, and lifetime distribution fitting, respectively) for a batch of samples using the current target model, and then calculates the losses according to the weighted coefficients. (First weight) (Second weight) (Third weight) (Fourth weight) combination is the overall objective loss:

[0135]

[0136] in, The first loss value, This is the second loss value. This is the third loss value. This is the fourth loss value;

[0137] The processor updates the model parameters based on the overall loss through backpropagation and optimization algorithms (such as gradient descent or its variants), repeating this process for several iterations until convergence, thereby obtaining a joint lifetime prediction model that simultaneously possesses good capabilities in identifying operating conditions, predicting load spectrum, and estimating probabilistic remaining lifetime.

[0138] In this way, by using multi-task joint loss to train working condition identification, load spectrum prediction and probabilistic lifetime estimation in a coordinated manner, and by appropriately processing failed samples and censored samples with Weibull density and survival likelihood respectively during training, it is possible to output accurate and uncertain remaining lifetime information under complex working conditions and incomplete labeling conditions, thereby improving the scientificity and reliability of filter replacement decisions.

[0139] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to that specific value.

[0140] After obtaining the remaining service life of the filter element after a second preset time period from the life prediction model output, the method further includes:

[0141] Based on the Weibull distribution parameters of the specific remaining lifetime value, the predicted remaining lifetime value corresponding to multiple preset quantiles is determined.

[0142] In this embodiment, the remaining service life is characterized by remaining service life information, which includes the specific value of the remaining service life and the Weibull distribution parameters corresponding to the specific value of the remaining service life. After obtaining the remaining service life of the filter element after the second preset time period output by the service life prediction model, the processor determines the remaining service life prediction values ​​corresponding to multiple preset quantiles based on the Weibull distribution parameters of the specific value of the remaining service life. The mathematical expression of this can be:

[0143]

[0144] in, Indicates the corresponding preset quantile. The predicted remaining life of the filter element (i.e., the time point at this quantile level). ∈ (0, 1) is a preset quantile, representing the cumulative failure probability (e.g., =0.5 represents the median. =0.9 represents the 90th percentile).

[0145] Formula (14) can be understood as the cumulative distribution function of the Weibull distribution. Solve for the quantile function: using a given preset quantile, i.e., probability... Request The point in time when it is established, thus giving the output a probabilistic meaning (Weibull parameter). Mapped into directly interpretable and usable time quantities That is, the predicted remaining lifespan.

[0146] For example, suppose the Weibull parameters output by the lifetime prediction model after a certain inference are shape parameter k=1.8 and scale parameter λ=150 (unit: hours). Now calculate the remaining lifetime prediction values ​​corresponding to three commonly used quantiles (i.e., three preset quantiles) (taking τ=0.1, 0.5, 0.9):

[0147] First, substitute the formula step by step and calculate:

[0148] For τ=0.1, first calculate Then, raising this value to the power of 1 / k, the exponent 1 / k = 1 / 1.8 ≈ 0.55555556, so... Finally, multiply by λ to get =150×0.2869≈43.0 hours;

[0149] For τ=0.5, first calculate Taking the power, we get (0.6931471806)1 / 1.8≈exp(0.55555556×ln0.6931471806)≈exp0.55555556×(−0.36651292))≈exp(−0.20361718)≈0.8155. Multiplying by the scale parameter, we get... =150×0.8155≈122.3 hours (i.e., median remaining lifetime is approximately 122.3 hours);

[0150] For τ=0.9, first calculate Raising the power, we get (2.302585093)1 / 1.8≈exp(0.55555556×ln2.302585093)≈exp(0.55555556×0.834032445)≈exp(0.46335136)≈1.5897. Multiplying by the scale parameter, we get... =150 × 1.5897 ≈ 238.5 hours;

[0151] The intuitive meaning can be derived from this: under the example conditions, the model gives a 10th percentile (conservative lower limit) of approximately 43.0 hours, a median (50th percentile) of approximately 122.3 hours, and a 90th percentile (optimistic upper limit) of approximately 238.5 hours; based on this, technicians can choose to change the time point or generate "suggested time period changes" under different risk tolerance levels.

[0152] In this way, by mapping the Weibull distribution parameters of the model output to the remaining lifetime prediction values ​​corresponding to multiple preset quantiles, the processor can simultaneously provide lifetime estimates at different confidence levels (quantiles), such as conservative, central, and optimistic, providing quantifiable time basis for planned maintenance and early warning decisions based on risk preferences and actual operation and maintenance constraints.

[0153] In some implementations, after determining the remaining lifetime predictions corresponding to multiple preset quantiles, the method further includes:

[0154] Among the remaining lifetime prediction values ​​corresponding to multiple preset quantiles, determine the remaining lifetime prediction value corresponding to the first quantile and the remaining lifetime prediction value corresponding to the second quantile, where the first quantile is greater than the second quantile.

[0155] Based on the remaining life prediction value corresponding to the first quantile, the remaining life prediction value corresponding to the second quantile, and the current time point, a recommended replacement time period is determined. The recommended replacement time period is used to indicate the time period for replacing the filter element.

[0156] In this embodiment, based on the remaining lifetime prediction values ​​corresponding to multiple preset quantiles, a first quantile is first determined among the remaining lifetime prediction values ​​corresponding to multiple preset quantiles. Corresponding remaining life prediction value With the second quantile Corresponding remaining life prediction value (The first quantile is greater than the second quantile, A conservative lower limit could be 0.9. (This can be 0.5, i.e., the center upper limit); to provide a suggested time period, the processor uses the current time point... Based on this, the remaining lifetime predictions corresponding to these two quantiles are mapped to the endpoints of a time window, suggesting that the time period be changed to start from the earlier endpoint. to the later endpoint ,Right now In this way, the processor can both provide maintenance personnel with an early-to-late timeframe for planned maintenance and make the risk preferences at different confidence levels explicit.

[0157] In some implementations, the remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to that specific value.

[0158] After obtaining the remaining service life of the filter element after a second preset time period from the life prediction model output, the method further includes:

[0159] Obtain the pressure difference between the upstream and downstream pressures of the filter element, and determine the obtained pressure difference between the upstream and downstream pressures of the filter element as the current pressure difference;

[0160] Based on the Weibull distribution parameters of the remaining lifespan value, determine whether the probability that the remaining lifespan of the filter element is less than or equal to the preset lifespan threshold is greater than or equal to the preset probability threshold.

[0161] If the probability that the remaining life of the filter element is less than or equal to the preset life threshold is greater than or equal to the preset probability threshold, or if the current pressure difference meets the first preset condition, or if the current pressure difference meets the second preset condition, output a filter element replacement prompt.

[0162] The first preset condition is that the current pressure difference is greater than or equal to the first difference value, which is the difference obtained by subtracting the preset buffer value from the preset pressure difference threshold.

[0163] The second preset condition is that the current pressure difference is greater than or equal to the opening threshold of the bypass valve in the return oil pipeline.

[0164] In this embodiment, after obtaining the probabilistic output of the life prediction model for the remaining lifespan of the filter element (i.e., the Weibull distribution parameters of the specific value of the remaining lifespan), the processor first acquires and determines the current physical quantity—that is, the difference between the upstream and downstream pressures of the filter element measured in real time—as the current pressure difference. Then, based on the Weibull distribution parameters k and λ output by the life prediction model, the filter element's lifespan threshold at a given early warning time is calculated. The probability of internal failure (i.e., the probability that the remaining lifespan of the filter element is less than or equal to the preset lifespan threshold) is calculated using the following expression:

[0165]

[0166] The probability Characterizing the filter element's future performance under current operating conditions The confidence level that failure will occur within a certain time frame;

[0167] At the same time, the processor also reads the preset pressure difference threshold from the device parameters. Buffer value and the opening threshold of the bypass valve in the return oil line ;

[0168] The processor's decision-making logic is executed according to three parallel rules:

[0169] when (where π∈[0,1] is a preset probability threshold) or the current pressure difference meets the first preset condition. When this happens, the processor outputs the first type of filter replacement prompt (warning / recommendation to replace);

[0170] When the current pressure difference meets the second preset condition When the processor outputs the second type of filter replacement prompt, namely the forced replacement prompt, it enters the forced replacement strategy (bypass protection is the highest priority). Therefore, this decision not only uses probabilistic lifetime information to quantify the risk, but also provides direct safety constraints with physical thresholds.

[0171] In this way, by combining probabilistic remaining life assessment with real-time differential pressure thresholds to trigger early warnings and mandatory replacements, maintenance decisions can be made intelligently and proactively based on a quantitative assessment of future failure risks, and can also be taken swiftly and forcefully when physical safety thresholds are reached, thus balancing preventative maintenance with operational safety.

[0172] Please participate Figure 3 This is a schematic diagram of the filter element life prediction device provided in the embodiments of this application. A second aspect of this application provides a filter element life prediction device 10, wherein the filter element is located in the return oil line of the hydraulic circuit of engineering machinery. The device 10 includes:

[0173] The acquisition module 11 is used to acquire the status information of the hydraulic circuit within a first preset time period, wherein the status information within the first preset time period is used to characterize the operating status of the hydraulic circuit within the first preset time period.

[0174] The construction module 12 is used to construct multiple sample features using the state information within a first preset time period, wherein each sample feature is used to characterize the operating state of the hydraulic circuit within a single preset unit of time.

[0175] Training module 13 is used to train the original model using multiple sample features to obtain a lifespan prediction model;

[0176] Application module 14 is used to acquire the status information of the hydraulic circuit within a second preset time period, and input the status information within the second preset time period into the life prediction model to obtain the remaining service life of the filter element after the second preset time period output by the life prediction model. The status information within the second preset time period is used to characterize the operating status of the hydraulic circuit within the second preset time period.

[0177] The filter life prediction device 10 provided in the second aspect of this application can realize the various processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0178] Please see Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. A third aspect of this application provides an electronic device 400, including a processor 410 and a memory 420. The memory 420 stores machine-executable instructions that can be executed by the processor 410. The processor 410 can execute the machine-executable instructions to implement the above-mentioned filter life method.

[0179] A fourth aspect of this application provides a machine-readable storage medium storing instructions that, when executed by a processor, cause the processor to implement the above-described filter lifespan method.

[0180] In one embodiment of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the filter life method according to the above embodiments.

[0181] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0182] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0183] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0184] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0185] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0186] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0187] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0188] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. A method for predicting filter cartridge lifespan, characterized in that, The filter element is located in the return oil line of the hydraulic circuit of the construction machinery, and the method includes: The state information of the hydraulic circuit within a first preset time period is obtained, wherein the state information within the first preset time period is used to characterize the operating state of the hydraulic circuit within the first preset time period. Multiple sample features are constructed using the state information within the first preset time period, wherein each sample feature is used to characterize the operating state of the hydraulic circuit within a single preset unit of time. The original model is trained using the features of the multiple samples to obtain a lifespan prediction model. The state information of the hydraulic circuit within a second preset time period is obtained, and the state information within the second preset time period is input into the life prediction model to obtain the remaining service life of the filter element after the second preset time period output by the life prediction model. The state information within the second preset time period is used to characterize the operating state of the hydraulic circuit within the second preset time period. The life prediction model includes a working condition classification sub-model, a load spectrum prediction sub-model, and a main prediction sub-model. The model output of the working condition classification sub-model includes the working condition category of the construction machinery at each time point within the second preset time period. The model output of the load spectrum prediction sub-model includes the load spectrum of the construction machinery at multiple time points after the second preset time period. The main prediction sub-model is used to determine the remaining service life of the filter element after the second preset time period based on the state information within the second preset time period, the model output of the working condition classification sub-model, and the model output of the load spectrum prediction sub-model.

2. The method according to claim 1, characterized in that, The status information within the first preset time period includes the pressure difference between the upstream and downstream pressures of the filter element within the first preset time period and the oil temperature of the return oil pipeline within the first preset time period. The construction of multiple sample features using state information within the first preset time period includes: Preprocess the status information within the first preset time period; The state information within the first preset time period after preprocessing is segmented according to the preset unit duration to obtain the multiple sample features; The preprocessing includes: For each time point within the first preset time period, the oil viscosity at that time point is determined based on the oil temperature at that time point. The oil viscosity is normalized based on the pressure difference at the specified time point.

3. The method according to claim 1, characterized in that, The remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life. The model training includes multiple iteration cycles, and the model corresponding to each iteration cycle is the target model of that iteration cycle. The step of training the original model using the multiple sample features includes: For each iteration cycle, the loss value of the iteration cycle is determined using the target loss function, and the model parameters of the target model for the iteration cycle are updated iteratively based on the loss value. Wherein, the loss value of the target loss function is the sum of the first loss value, the second loss value, the third loss value and the fourth loss value, the first loss value is the product of the first weight and the loss value of the first loss function, the first loss function is used to determine the loss value of the first original sub-model in each iteration cycle, and the first original sub-model corresponds to the working condition classification sub-model; The second loss value is the product of the second weight and the loss value of the second loss function. The second loss function is used to determine the loss value of the second original sub-model in each iteration cycle. The second original sub-model corresponds to the load spectrum prediction sub-model. The third loss value is the product of the third weight and the loss value of the third loss function. The third loss function is used to determine the loss value of the third original sub-model corresponding to the specific value of the remaining lifetime in each iteration cycle. The third original sub-model corresponds to the backbone prediction sub-model. The fourth loss value is the product of the fourth weight and the loss value of the fourth loss function, which is used to determine the loss value of the third original sub-model corresponding to the Weibull distribution parameters of the specific value of the remaining lifetime in each iteration cycle.

4. The method according to claim 3, characterized in that, The first loss function is based on cross-entropy loss, the second loss function is based on quantile loss or mean squared error loss, the third loss function is based on Huber loss, and the fourth loss function includes a first sub-loss function and a second sub-loss function; The first sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the first sample, and the second sub-loss function is used to determine the loss value of the Weibull distribution parameter corresponding to the second sample. The first sub-loss function is the negative log-likelihood of the probability density function of the Weibull distribution, and the second sub-loss function is the negative log-likelihood of the survival function of the Weibull distribution. The first sample is the sample feature corresponding to a preset unit time period during which the filter element has failed among the plurality of sample features; The second sample is the sample feature corresponding to a preset unit time during which the filter element has not failed, among the multiple sample features.

5. The method according to claim 1, characterized in that, The remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life. After obtaining the remaining service life of the filter element after the second preset time period as output by the service life prediction model, the method further includes: Based on the Weibull distribution parameters of the specific remaining lifetime value, the predicted remaining lifetime values ​​corresponding to multiple preset quantiles are determined.

6. The method according to claim 5, characterized in that, After determining the remaining life prediction values ​​corresponding to multiple preset quantiles, the method further includes: Among the remaining lifetime prediction values ​​corresponding to the plurality of preset quantiles, determine the remaining lifetime prediction value corresponding to the first quantile and the remaining lifetime prediction value corresponding to the second quantile, wherein the first quantile is greater than the second quantile; Based on the remaining life prediction value corresponding to the first quantile, the remaining life prediction value corresponding to the second quantile, and the current time point, a recommended replacement time period is determined, which is used to indicate the time period for replacing the filter element.

7. The method according to claim 1, characterized in that, The remaining useful life is characterized by remaining useful life information, which includes a specific value of the remaining useful life and a Weibull distribution parameter relating to the specific value of the remaining useful life. After obtaining the remaining service life of the filter element after the second preset time period as output by the service life prediction model, the method further includes: The pressure difference between the upstream and downstream pressures of the filter element is obtained, and the obtained pressure difference between the upstream and downstream pressures of the filter element is determined as the current pressure difference. Based on the Weibull distribution parameters of the specific value of the remaining lifespan, determine whether the probability that the remaining lifespan of the filter element is less than or equal to a preset lifespan threshold is greater than or equal to a preset probability threshold. If the probability that the remaining lifespan of the filter element is less than or equal to the preset lifespan threshold is greater than or equal to the preset probability threshold, or if the current pressure difference meets the first preset condition, or if the current pressure difference meets the second preset condition, a filter element replacement prompt will be output. Wherein, the first preset condition is that the current pressure difference is greater than or equal to the first difference value, and the first difference value is the difference obtained by subtracting the preset buffer value from the preset pressure difference threshold; The second preset condition is that the current pressure difference is greater than or equal to the opening threshold of the bypass valve of the return oil pipeline.

8. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in a memory, it implements the steps of the filter cartridge life prediction method according to any one of claims 1-7.

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

Images