Self-cleaning filtering method, system and equipment of metal-free friction equipment

Abstract

The invention discloses a self-cleaning filtering method, system and equipment of metal-free friction equipment, and relates to the technical field of filtering equipment.The method comprises the steps that a modular equipment filter element comprising a main filter element and a standby filter element is designed, rapid insertion and extraction switching is achieved, and the filter element is integrated with an RFID tag to read historical cleaning filtering data; constructing a filter element state collaborative prediction channel based on the data; filtering the multi-dimensional data flow by combining equipment acquired by a multi-parameter sensor network to obtain a filter element state prediction parameter; and then target parameters are determined through a graded filter element self-cleaning strategy, and filter element retentate cleaning and filtering are executed by an equipment self-cleaning mechanism. The technical problems that in the cleaning and running process of traditional filtering equipment, metal foreign matter is likely to be generated, a filter element is inconvenient to replace, a cleaning strategy is not accurate enough, and the requirement of a high-precision filtering scene is difficult to meet are solved, and the purposes of avoiding generation of the metal foreign matter in the filtering process, rapidly replacing the filter element and precisely controlling self-cleaning are achieved. The filtering efficiency and the applicability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of filtering equipment, in particular to a self-cleaning filtering method, system and equipment of a metal-friction-free equipment. BACKGROUND

[0002] In the fields of industrial production and water treatment, filtering and separation technology is a key link to realize material purification and improve product quality. In the prior art, traditional filtering equipment and manual cleaning methods are often used to complete separation operations. These methods have played a certain role in closed systems or low-frequency filtering scenarios. However, with the increasing requirements for filtering precision and metal contamination in the chemical industry, lithium battery industry and other industries, the traditional technology has exposed limitations in application: traditional equipment relies on metal component friction to complete cleaning, which is easy to produce metal foreign matter pollution; and filter core replacement is cumbersome, and cleaning timing is judged by experience, resulting in low separation efficiency and high cost, which is difficult to meet the separation needs of high-precision filtering scenarios. SUMMARY

[0003] The present application provides a self-cleaning filtering method, system and equipment of a metal-friction-free equipment, which is used to solve the technical problems that the traditional filtering equipment is easy to produce metal foreign matter during cleaning and running, the filter core is inconvenient to replace, the cleaning strategy is not accurate enough, and it is difficult to meet the needs of high-precision filtering scenarios.

[0004] In a first aspect, the present application provides a self-cleaning filtering method of a metal-friction-free equipment, which comprises: designing a modular equipment filter core, the modular equipment filter core comprising a main filter core and a plurality of standby filter cores, wherein the main filter core and the plurality of standby filter cores are combined through a buckle type quick installation structure and a multi-channel reversing valve for quick plug-in switching, and each equipment filter core is integrated with an RFID tag, and equipment historical cleaning and filtering data sets are read through the RFID tag; performing associated feature extraction and state prediction training based on the equipment historical cleaning and filtering data sets, constructing a filter core state cooperative prediction channel; deploying a multi-parameter sensor network to collect equipment filtering multidimensional data streams, performing prediction analysis on the equipment filtering multidimensional data streams based on the filter core state cooperative prediction channel to obtain filter core state prediction parameters; constructing a hierarchical filter core self-cleaning strategy, performing strategy analysis on the filter core state prediction parameters based on the hierarchical filter core self-cleaning strategy, determining target filter core self-cleaning strategy parameters, and performing filter core retentate cleaning and filtering through an equipment self-cleaning mechanism according to the target filter core self-cleaning strategy parameters.

[0005] In a second aspect of the present application, a self-cleaning filtration system for a metal-free friction device is provided, comprising: a modular device filter core construction module for designing a modular device filter core, the modular device filter core comprising a main filter core and a plurality of backup filter cores, wherein the main filter core and the plurality of backup filter cores are quickly plugged and switched through a buckle type quick installation structure and a multi-channel reversing valve combination, and each device filter core is integrated with an RFID tag, and device historical cleaning and filtration data sets are read through the RFID tag; a filter core state cooperative prediction channel construction module for performing associated feature extraction and state prediction training based on the device historical cleaning and filtration data sets, and constructing a filter core state cooperative prediction channel; a filter core state prediction parameter acquisition module for deploying a multi-parameter sensor network to collect device filtration multi-dimensional data streams, and performing prediction analysis on the device filtration multi-dimensional data streams based on the filter core state cooperative prediction channel to obtain filter core state prediction parameters; and a filter core retentate cleaning and filtration execution module for constructing a hierarchical filter core self-cleaning strategy, performing strategy analysis on the filter core state prediction parameters based on the hierarchical filter core self-cleaning strategy, determining target filter core self-cleaning strategy parameters, and executing filter core retentate cleaning and filtration through a device self-cleaning mechanism according to the target filter core self-cleaning strategy parameters.

[0006] In a third aspect of the present application, an electronic device is provided, comprising: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement a self-cleaning filtration method for a metal-free friction device.

[0007] One or more technical solutions provided in the present application have at least the following technical effects or advantages: The present application designs a modular filter core with an RFID tag, quickly switches through a buckle type quick installation structure and a multi-channel reversing valve, constructs a filter core state cooperative prediction channel based on historical data, performs prediction analysis in combination with multi-dimensional data streams collected by a multi-parameter sensor network, determines target parameters through a hierarchical self-cleaning strategy, and executes cleaning by a self-cleaning mechanism, thereby avoiding the generation of metal foreign matter, achieving efficient filtration, and making the filtration and self-cleaning effect of the metal-free friction device more precise and reliable, and achieving the technical effects of avoiding the generation of metal foreign matter in the filtration process, precise control of filter core quick replacement and self-cleaning, and improving filtration efficiency and applicability. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.

[0009] Figure 1This is a schematic flowchart of the self-cleaning filtration method for a metal-free friction device provided in the embodiments of this application.

[0010] Figure 2 This is a schematic diagram of the structure of the self-cleaning filtration system of the metal-free friction device provided in the embodiments of this application.

[0011] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Figure reference numerals: Modular equipment filter element construction module 1, filter element state collaborative prediction channel construction module 2, filter element state prediction parameter acquisition module 3, filter element retained material cleaning and filtration execution module 4, input device 301, processor 302, memory 303, output device 304. Detailed Implementation

[0013] This application provides a self-cleaning filtration method, system, and device for metal frictionless equipment, which solves the technical problems of traditional filtration equipment that easily generates metal foreign objects during cleaning and operation, is inconvenient to replace filter elements, has insufficiently precise cleaning strategies, and is difficult to meet the needs of high-precision filtration scenarios.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown, a self-cleaning filtration method for a metal-free frictionless device, wherein the method includes: Step A100: design a modular device filter element, the modular device filter element comprising a main filter element and a plurality of backup filter elements, wherein the main filter element and the plurality of backup filter elements are switched quickly by a buckle type quick installation structure and a multi-channel reversing valve combination, and each device filter element is integrated with an RFID tag, and a device historical cleaning and filtering data set is read through the RFID tag.

[0017] In the embodiments of the present application, the RFID tag is a radio frequency identification tag, which is only used as a data storage carrier to realize the tracing and calling of the historical operation data of the filter element.

[0018] Specifically, when designing the modular device filter element, first, the combined structure of the main filter element and the plurality of backup filter elements is determined to meet the continuous filtering requirements. The main filter element serves as the main filtering component, and the backup filter element is used for temporary replacement when the main filter element needs to be cleaned or replaced, so as to ensure that the filtering process is not interrupted.

[0019] In order to realize the quick switching of the filter element, the buckle type quick installation structure is used to connect the main filter element and the backup filter element, and the multi-channel reversing valve is used to control the fluid passage. When the main filter element needs to be switched, the buckle type quick installation structure can realize the quick plugging and unplugging of the filter element, and the multi-channel reversing valve can synchronously switch the fluid flow direction to the backup filter element. The time required for the whole switching process is much lower than that of the traditional bolt connection method, which greatly improves the continuous operation capability of the device.

[0020] Meanwhile, the RFID tag is integrated inside each filter element. The tag can record the filtering time, the accumulated amount of trapped substances, the cleaning frequency and other device historical cleaning and filtering data sets in real time. The storage capacity of a single tag can reach more than 1000, which ensures the complete tracing of the data. When the filter element is switched or the state is detected, the RFID tag information can be read through the reader, so that the whole life cycle data of the filter element can be quickly obtained, which provides a basis for subsequent state analysis.

[0021] Further, the RFID tag does not directly collect the device historical cleaning and filtering data set. Its core function is to store data rather than actively collect. The actual collection process is completed by the whole non-metallic friction self-cleaning and filtering device including the modular device filter element, i.e. the matching sensing system and control system of the non-metallic friction device: when the filter element is running, the timing module on the non-metallic friction device records the filtering time, the pressure or flow sensor indirectly calculates the accumulated amount of trapped substances by monitoring the pressure difference change, and the action counter of the cleaning mechanism records the cleaning frequency. These data are transmitted to the control system of the non-metallic friction device in real time. The control system writes the filtering time, the accumulated amount of trapped substances, the cleaning frequency and other information into the RFID tag integrated inside the filter element for storage through the reader. When the data needs to be called, the device historical cleaning and filtering data set stored in the tag is read through the reader.

[0022] By designing a multi-filter combination, a quick installation and switching structure, and an integrated data recording function, the rapid replacement of filter cartridges and the effective collection of historical data are realized, laying a foundation for the efficient operation and intelligent management of the filtering equipment.

[0023] Step A200: Correlation feature extraction and state prediction training based on the device historical cleaning filtration dataset, to construct a filter cartridge state collaborative prediction channel.

[0024] Optionally, when constructing the filter cartridge state collaborative prediction channel, the device historical cleaning filtration dataset needs to be cleaned according to the filter cartridge data application standard to obtain a usable device cleaning filtration dataset, a filter cartridge state label library containing filter cartridge use, clogging degree, and adsorption performance state is constructed, and the usable device cleaning filtration dataset is labeled according to the correlation feature extraction and labeling to obtain a device cleaning filtration label feature sample set. The sample set is then subjected to correlation feature extraction and migration training optimization, and the specific steps are described in detail in A210-A240.

[0025] Step A300: Deploy a multi-parameter sensor network to collect device filtration multi-dimensional data streams, and perform prediction analysis on the device filtration multi-dimensional data streams based on the filter cartridge state collaborative prediction channel to obtain filter cartridge state prediction parameters.

[0026] In an embodiment of the present application, first, when deploying a multi-parameter sensor network, according to the key influencing factors of the filter cartridge filtration process, a variety of sensors including differential pressure sensors, flow sensors, temperature sensors, and retained material concentration sensors are installed at key positions such as the inlet and outlet of the filter cartridge of the metal-free friction equipment, the filtration cavity, and the fluid pipeline, forming a monitoring network covering the entire filtration process. These sensors capture multi-dimensional parameters in real time according to the preset sampling frequency, such as collecting differential pressure data every 3 seconds and flow data every 5 seconds, to ensure that the real-time working state of the filter cartridge can be fully reflected.

[0027] Next, the original data collected by the multi-parameter sensor network will be aggregated to the data processing unit of the metal-free friction equipment through wired or wireless transmission for preliminary preprocessing. This process includes real-time elimination of abnormal values, such as values exceeding 15% of the normal range due to sensor transient fluctuations; data standardization, which maps parameters of different magnitudes to the 0-1 interval; and time alignment, which ensures that the collected data of different sensors are consistent in time dimension, finally forming a structured device filtration multi-dimensional data stream, providing a high-quality data basis for subsequent analysis.

[0028] Then, the pre-processed equipment filter multi-dimensional data stream input is input to the constructed filter state collaborative prediction channel. The channel can automatically identify the filter state signal contained in the equipment filter multi-dimensional data stream by integrating the feature correlation model formed by training the historical data, for example, by analyzing the rising slope of the differential pressure over time to judge the development trend of the blockage, combining the flow decay rate to evaluate the change of the adsorption performance, and comprehensively judging the potential influence of the temperature on the filter material.

[0029] After analysis and calculation by the filter state collaborative prediction channel, the final output is specific filter state prediction parameters, including quantitative indexes such as the current use time ratio of the filter, the blockage degree percentage, the adsorption efficiency coefficient, and qualitative state descriptions such as mild blockage and high adsorption efficiency, which fully present the real-time state of the filter.

[0030] By deploying a multi-parameter sensor network to collect comprehensive data, pre-processing to ensure data quality, and analyzing and correlating features with the help of the filter state collaborative prediction channel, accurate filter state prediction parameters are finally obtained, providing a scientific and real-time basis for the development of subsequent self-cleaning strategies.

[0031] Step A400: Construct a hierarchical filter self-cleaning strategy, perform strategy analysis on the filter state prediction parameters based on the hierarchical filter self-cleaning strategy, determine target filter self-cleaning strategy parameters, and execute filter retention cleaning by the equipment self-cleaning mechanism according to the target filter self-cleaning strategy parameters.

[0032] In the embodiments of the present application, the equipment self-cleaning mechanism specifically includes an elastomer wiper and a pneumatic pulse backflushing device, wherein the motion trajectory of the elastomer wiper is parallel to the axial direction of the modular equipment filter, and the modular equipment filter is arranged at the nozzle outlet of the pneumatic pulse backflushing device.

[0033] Specifically, first, determine the target strategy based on the hierarchical strategy matching filter state prediction parameters, obtain the parameter threshold value through analysis, construct the cleaning effect target function and iterate optimization within the threshold value, determine the target filter self-cleaning strategy parameters, and the specific steps are described in detail in A410-A440.

[0034] Next, after the equipment self-cleaning mechanism receives the target filter self-cleaning strategy parameters, first start the control system of the non-metallic friction equipment, convert the instructions of the target filter self-cleaning strategy parameters into specific execution signals, and transmit them to the elastomer wiper and the pneumatic pulse backflushing device. Among them, the elastomer wiper makes reciprocating motion along the axial direction of the modular equipment filter under the fluid power drive according to the motion frequency and contact pressure set in the parameters, and strips the retention adhered to the outer surface of the filter through flexible contact, avoiding the generation of metal foreign matter by rigid friction.

[0035] At the same time, the pneumatic pulse backflushing device sprays high-pressure airflow into the inside of the filter element through the nozzle according to the pressure value, pulse interval and duration in the target parameters. Since the filter element is arranged at the outlet of the nozzle, the high-speed airflow can directly impact the filter element channel, blow out the deep-trapped substances that are not removed by the scraper, form synergy with surface cleaning, and improve the overall cleaning effect.

[0036] During the cleaning process, the multi-parameter sensor network deployed on the metal-free friction equipment monitors the filter element differential pressure, flow and other data in real time, and compares them with the preset cleaning effect threshold, such as the recovery of the differential pressure to 90% of the initial value. The specific cleaning effect threshold is set by the person skilled in the art according to the actual project and equipment requirements. If the cleaning does not meet the standard after one cleaning, the cleaning process will be automatically repeated according to the parameters until the real-time monitoring data meet the requirements of the cleaning effect threshold.

[0037] After the cleaning is completed, the device self-cleaning mechanism is automatically reset, the elastomeric scraper returns to the initial position, the pneumatic pulse device stops working, and the parameters and effects of this cleaning are recorded. The RFID tag integrated with the filter element is updated to the device historical cleaning filtration data set, providing data support for subsequent state prediction.

[0038] Through the cooperative operation of the elastomeric scraper and the pneumatic pulse backflushing device according to the target parameters by the device self-cleaning mechanism, combined with real-time monitoring and feedback adjustment, efficient removal of filter element trapped substances is achieved, the filtration performance is restored, and metal foreign matter is avoided, ensuring the stable operation of the metal-free friction equipment.

[0039] Further, step A200 in the method provided by the embodiment of the application comprises: A210: performing abnormal data cleaning on the device historical cleaning filtration data set according to filter element data application standards, to obtain a usable device cleaning filtration data set.

[0040] A220: constructing a filter element state label library, wherein the filter element state label library comprises a filter element use state, a clogging degree state and an adsorption performance state.

[0041] A230: performing associated feature extraction and feature labeling on the usable device cleaning filtration data set according to the filter element state label library, to obtain a device cleaning filtration label feature sample set.

[0042] A240: respectively performing associated feature extraction and migration training optimization on the device cleaning filtration label feature sample set, to construct a filter element state cooperative prediction channel.

[0043] Specifically, first, the device historical cleaning filtration data set is cleaned of abnormal data according to the filter data application standard. The construction of the filter data application standard can be based on the design parameters of the device, the performance indicators of the filter, and the normal data characteristics of long-term operation: those skilled in the art can combine the design threshold of the filtration device, such as the maximum allowable pressure difference, the rated filtration time, the rated service life of the filter, the number of tolerable cleanings, and other inherent parameters, and then determine the reasonable fluctuation range of indicators such as filtration pressure difference and retention accumulation rate by statistical analysis of historical normal operation data to form standardized data screening rules.

[0044] The device historical cleaning filtration data set may contain abnormal values due to sensor fluctuations and recording errors. The determination of abnormal values in the device historical cleaning filtration data set is mainly based on the threshold range and logical reasonableness in the filter data application standard: for example, when the filtration pressure difference of a certain data instantaneously exceeds 30% of the device design threshold, or the cleaning number appears negative, the filtration time contradicts the actual operation cycle, etc., it will be determined as an abnormal value; values deviating from the normal data distribution range can also be identified by statistical methods, such as data exceeding 3 times the standard deviation of the mean, and then these abnormal data due to sensor fluctuations and recording errors are removed. After cleaning, assuming that there are 920 groups of data that meet the standard among the original 1000 groups of historical data, a usable device cleaning filtration data set is formed, providing a reliable basis for subsequent analysis.

[0045] Subsequently, a filter state label library is constructed, which clearly contains three core labels of filter usage state, clogging degree state, and adsorption performance state, and the specific data is shown in Table 1.

[0046] Table 1: Core label division and state table

[0047] Based on the constructed filter state label library, the available equipment cleaning filtration data set is associated with feature extraction and feature labeling. First, the key features related to the filter state are extracted from the available equipment cleaning filtration data set, including cumulative filtration time, unit time retention weight, filtration flow rate change rate, inlet and outlet pressure difference change amount, and historical cleaning times. These features are directly related to the filter use state, the clogging degree, and the adsorption performance. Then, according to the definition standards of the three states in the filter state label library, the feature values of each data are compared one by one. If the cumulative filtration time of a certain data is 300 hours, the filter use state label of normal use is matched. If the pressure difference rise amplitude is 25%, the clogging degree state label of moderate clogging is matched. If the current retention efficiency calculated by the retention weight and the initial filtration efficiency is 85%, the adsorption performance state label of medium efficiency is matched. Through this accurate comparison of feature values and label thresholds, the matching of three types of state labels is completed for each data, and finally the equipment cleaning filtration label feature sample set with complete labels is formed.

[0048] For example, a set of data showing filtration time of 200 hours, pressure difference rise of 15%, and retention efficiency of 90% is labeled as filter use state: normal use, clogging degree state: mild clogging, and adsorption performance state: high efficiency, and finally forms an equipment cleaning filtration label feature sample set containing 920 labeled data.

[0049] Then, the filter state collaborative prediction channel is constructed, the filter prediction multi-task framework is constructed according to the filter state label library, the source domain multi-task prediction channel set is extracted and analyzed, the multi-dimensional filter state prediction channel is obtained by classification transfer training of the equipment cleaning filtration label feature sample set, and the online learning mechanism is introduced for optimization and update. The specific steps are described in detail in A241-A244.

[0050] Through data cleaning to ensure the quality of basic data, constructing label library to clarify analysis dimensions, generating label sample set to provide training materials, and constructing filter state collaborative prediction channel through transfer training and optimization, the accurate collaborative prediction of multi-dimensional filter state is realized, which provides a reliable basis for the subsequent development of self-cleaning strategy.

[0051] Further, the method provided in the embodiment of the application comprises the following steps A240: A241: Constructing a filter prediction multi-task framework according to the filter state label library.

[0052] A242: Extracting and analyzing each prediction task in the filter prediction multi-task framework to obtain a source domain multi-task prediction channel set.

[0053] A243: using the source domain multi-task prediction channel set respectively to perform classification migration training on the equipment cleaning filter label feature sample set, and obtaining a multi-dimensional filter state prediction channel.

[0054] A244: introducing an online learning mechanism to verify the performance of the multi-dimensional filter state prediction channel and iteratively learn and update, and constructing the filter state collaborative prediction channel.

[0055] Optionally, when constructing the filter prediction multi-task framework according to the filter state label library, first, the three core labels of filter use state, clogging degree state, and adsorption performance state contained in the filter state label library are determined, and the prediction target of the filter prediction multi-task framework is determined based on this, that is, the precise prediction of the three types of states is realized at the same time. Assuming that the filter state label library covers 1000 pieces of filter state annotation data under different working conditions, the data proportion of the filter use state of newly enabled, normal use, and aging is 20%, 60%, and 20% respectively, the clogging degree data from no clogging to severe clogging is distributed in a gradient, and the adsorption performance data proportion of high, medium, and low efficiency is about 5:3:2.

[0056] Based on the above data distribution characteristics, the input feature dimension and output layer structure of the filter prediction multi-task framework are designed. First, 12 key input features are selected by combining the influencing factors of filter use state, clogging degree, and adsorption performance, including cumulative filtration time, unit time pressure difference change rate, cumulative weight of retained material, filtration flow fluctuation rate, initial filtration efficiency, cleaning frequency, material loss rate, temperature influence coefficient, fluid viscosity, pollutant concentration, filter pore size change amount, and historical adsorption efficiency decay rate, to form a 12-dimensional feature vector of the input layer, so as to fully capture the feature differences of various states under different working conditions.

[0057] The output layer adopts a multi-branch structure, corresponding to three types of prediction tasks: a 3-classification branch for use state, including newly enabled, normal use, and aging, with an output dimension of 3, matching the proportion distribution of 20%, 60%, and 20% in the filter use state data; a 4-classification branch for clogging degree, including no clogging, mild clogging, moderate clogging, and severe clogging, with an output dimension of 4, adapting to the gradient distribution of the category division; a 3-classification branch for adsorption performance, including high, medium, and low efficiency, with an output dimension of 3, corresponding to the proportion relationship of 5:3:2. Each branch adopts a softmax activation function, and class weights are introduced in the loss function, such as giving higher weights to the aging and low efficiency categories with low proportion, balancing the data distribution difference, and ensuring that the framework can accurately cover and distinguish various states.

[0058] Then, the source domain dataset containing multi-task filtering state data of different types of filter elements is extracted and selected for each prediction task in the filter element prediction multi-task framework, and the LSTM network is trained to obtain a set of multi-type filter element state prediction channels. The model parameters are extracted and optimized to obtain a set of source domain multi-task prediction channels. The specific steps are described in detail in A242-1-A242-3.

[0059] After that, the source domain multi-task prediction channel set is adaptively associated and the migration learning strategy is preset according to the filter element prediction multi-task target. The strategy and the source domain channel set are used to train the sample set to generate three branch prediction channels, and then the channels are integrated and merged to obtain a multi-dimensional filter element state prediction channel. The specific steps are described in detail in A243-1-A243-3.

[0060] Next, an online learning mechanism is introduced. First, the iteration trigger condition is set. Every time 100 real-time filtering data are added, a model update is started. The real-time filtering data includes 12 key input features such as differential pressure change, flow fluctuation, and trapped weight, as well as the corresponding actual state labels. At the same time, a sliding window mechanism is used to retain the last 3000 effective data as a dynamic training set, avoiding the influence of historical redundant data on the adaptability of the multi-dimensional filter element state prediction channel.

[0061] Next, the newly collected real-time data is used to verify the performance of the multi-dimensional filter element state prediction channel. The predicted state (such as clogging degree, adsorption performance, etc.) output by the multi-dimensional filter element state prediction channel is compared with the actual detected state label, and the prediction accuracy, recall rate, and overall loss value of each category are calculated to obtain the verification result. For example, in the initial verification, it is found that the prediction accuracy of the multi-dimensional filter element state prediction channel for moderate clogging state is only 78%, which is lower than the 85%-90% of other categories, and the overall loss value is 0.12, which is higher than the preset threshold value 0.08, indicating that there is room for optimization.

[0062] Based on the above verification results, parameter iteration optimization is performed. For the state category with higher error, its weight is increased in the loss function, and the dynamic training set is used to perform incremental update on the shared feature layer and each branch prediction layer parameter of the multi-dimensional filter element state prediction channel. After each iteration, the performance is re-evaluated. If the overall accuracy of continuous 3 iterations is improved by less than 1%, the current round of optimization is stopped. After 5 iterations, the prediction accuracy of the moderate clogging state in the example is improved to 86%, the overall loss value is reduced to 0.06, and the stability of the multi-dimensional filter element state prediction channel is significantly improved.

[0063] The above verification and optimization process is continuously repeated, and as data accumulates, after processing 5000 real-time data, the multi-dimensional filter state prediction channel accurately predicts the comprehensive prediction accuracy of the filter usage state, the clogging degree, and the adsorption performance to more than 90%, and the prediction error under different working conditions (such as fluid viscosity change and pollutant concentration fluctuation) is controlled within ± 3%, and finally a filter state collaborative prediction channel that can dynamically adapt to changes in filter operating state is constructed.

[0064] By constructing a prediction framework covering multiple states, migrating training to adapt to new scenarios, and online learning to dynamically optimize the model, precise collaborative prediction of filter usage, clogging, and adsorption performance is achieved, providing a scientific basis for subsequent self-cleaning strategy development.

[0065] Further, the method provided in the embodiment of the application comprises the following step A242: A242-1: Extractively analyze each prediction task in the filter prediction multi-task framework, and select a source domain filter state data set, wherein the source domain filter state data set comprises a multi-task filter state data set of different types of filters.

[0066] A242-2: Use an LSTM network structure to respectively perform identification prediction training on the source domain filter state data set, and obtain a multi-type filter state prediction channel set.

[0067] A242-3: Extract model parameter information of the multi-type filter state prediction channel set for aggregated learning optimization, and obtain the source domain multi-task prediction channel set.

[0068] In the embodiment of the application, the LSTM network structure is a long short-term memory network.

[0069] Specifically, first, it is clear that the filter prediction multi-task framework includes three prediction tasks of filter usage state, clogging degree state, and adsorption performance state, and then the input features and output targets of each task are disassembled, the input features include filtering duration, pressure difference change rate, and interception efficiency. Based on the disassembly result, a source domain filter state data set is selected from a historical database storing a collection of past filtering operation data of various filters, which covers a multi-task filter state data set of PP cotton, activated carbon, ceramic and other different types of filters. Each data record contains 15 feature parameters and corresponding state labels, which include the aforementioned 12 key input features and more subdivided dimensions or auxiliary features, such as environmental factors and equipment operating parameters, to ensure that the data covers the filter state under different materials and different working conditions.

[0070] Then, when the model parameter information of the multi-type filter core state prediction channel set is extracted for aggregated learning optimization, the core parameters such as the weight matrix and the bias vector of each channel are extracted first, and then a weighted aggregation strategy is adopted, and weights are allocated according to the application frequency of different types of filter cores in the target scene, such as a PP cotton filter core accounting for 40% and a ceramic filter core accounting for 30%. Through parameter sharing and conflict resolution, the parameters of multiple independent channels are integrated into a unified model parameter set, and the prediction accuracy of the aggregated model for mixed-type filter core data can be improved to 90% during testing, and finally a source domain multi-task prediction channel set is obtained.

[0071] Then, when the model parameter information of the multi-type filter core state prediction channel set is extracted for aggregated learning optimization, the core parameters such as the weight matrix and the bias vector of each channel are extracted first, and then a weighted aggregation strategy is adopted, and weights are allocated according to the application frequency of different types of filter cores in the target scene, such as a PP cotton filter core accounting for 40% and a ceramic filter core accounting for 30%. Through parameter sharing and conflict resolution, the parameters of multiple independent channels are integrated into a unified model parameter set, and the prediction accuracy of the aggregated model for mixed-type filter core data can be improved to 90% during testing, and finally a source domain multi-task prediction channel set is obtained.

[0072] Through analyzing the task to select the source domain data, training the single-type prediction channel with the LSTM network, and aggregating and optimizing the parameters, the source domain multi-task prediction channel set that can adapt to multiple types of filter cores is obtained, which provides an efficient base model for subsequent transfer training.

[0073] Further, step A243 in the method provided by the embodiment of the application includes: A243-1: performing adaptively associated analysis on the source domain multi-task prediction channel set according to a filter core prediction multi-task target, and presetting a transfer learning strategy.

[0074] A243-2: performing classification transfer training on the equipment cleaning filter label feature sample set based on the source domain multi-task prediction channel set respectively by using the transfer learning strategy, to generate a filter core use state branch prediction channel, a blockage degree state branch prediction channel and an adsorption performance state branch prediction channel.

[0075] A243-3: integrating and merging the filter core use state branch prediction channel, the blockage degree state branch prediction channel and the adsorption performance state branch prediction channel, to obtain a multi-dimensional filter core state prediction channel.

[0076] In the embodiments of the present application, the filter core prediction multi-task target is to simultaneously predict the use state, the clogging degree state and the adsorption performance state of the filter core, which corresponds to the three core states covered in the filter core state label library.

[0077] Specifically, when performing adaptive correlation analysis on the source domain multi-task prediction channel set according to the filter core prediction multi-task target, it is first determined that the target is to simultaneously predict the use state, the clogging degree state and the adsorption performance state of the filter core, and then the matching degree of the input features, the output dimensions of each model in the source domain multi-task prediction channel set and the target task are compared one by one. The shared features and the difference features are identified. Based on this correlation analysis, the preset parameter migration learning strategy, the shared feature parameters in the source domain multi-task prediction channel set are retained, and only the parameters corresponding to the difference features are reinitialized, laying a foundation for subsequent migration training.

[0078] Then, using the preset migration learning strategy, the device cleaning and filtering label feature sample set is classified and migrated based on the source domain multi-task prediction channel set. The device cleaning and filtering label feature sample set is divided into three subsets according to the prediction task: a filter core use state subset, a clogging degree state subset and an adsorption performance state subset. For each subset, the pre-trained model corresponding to the task in the source domain multi-task prediction channel set is called to fine-tune, for example, in the clogging degree state training, the mild clogging (10%-20% pressure difference rising amplitude) data in the target sample set is iterated for 15 rounds, so that the recognition accuracy of the model for this category is improved. Finally, three independent branch prediction channels are generated, i.e., a filter core use state branch prediction channel, a clogging degree state branch prediction channel and an adsorption performance state branch prediction channel, which correspond to the prediction of the three states respectively.

[0079] Finally, the three branch prediction channels are integrated and merged by combining feature splicing and weighted voting: the intermediate feature vectors of each channel are spliced into a unified feature matrix, and then the weights are allocated according to the prediction confidence of each channel, the clogging degree state branch prediction channel weight is 40%, the filter core use state branch prediction channel weight is 30%, and the adsorption performance state branch prediction channel weight is 30%. The comprehensive prediction accuracy of the integrated multi-dimensional filter core state prediction channel for the mixed sample is improved compared with the single channel, and the recognition accuracy for the edge state (such as aging + moderate clogging) is improved more significantly.

[0080] Through adaptive analysis to determine the migration strategy, classification training to generate branch channels and integrated merging to optimize the prediction performance, the precise collaborative prediction of the multi-dimensional state of the filter core is realized, and comprehensive data support is provided for the dynamic adjustment of the subsequent self-cleaning strategy.

[0081] Further, step A400 in the method provided in the embodiments of the present application comprises: A410: Strategy matching is performed on the filter state prediction parameters based on the hierarchical filter self-cleaning strategy to determine a target filter self-cleaning strategy.

[0082] A420: Strategy analysis is performed on the filter state prediction parameters using the target filter self-cleaning strategy to obtain filter self-cleaning strategy parameter thresholds.

[0083] A430: A filter cleaning effect target function is constructed according to filter cleaning requirements.

[0084] A440: Iterative optimization is performed within the filter self-cleaning strategy parameter thresholds based on the filter cleaning effect target function to determine target filter self-cleaning strategy parameters.

[0085] Specifically, first, according to the core state dimensions in the filter state label library, the filter usage state, the clogging degree state, and the adsorption performance state are divided into cleaning levels to form a multi-dimensional linked hierarchical filter self-cleaning strategy. For example, according to the clogging degree state, it is divided into three levels: mild (pressure difference increase of 10%-20%), moderate (20%-40%), and severe (>40%), combined with the filter usage state (newly activated, normal use, and aging) and the adsorption performance state (high efficiency, medium efficiency, and low efficiency) to develop differentiated strategies: a normal use filter with mild clogging and medium adsorption efficiency is cleaned with an elastomer scraper; an aging filter with moderate clogging and low adsorption efficiency is cleaned with an elastomer scraper + pressure 0.3 MPa pneumatic pulse backwashing; and a severely clogged filter is cleaned with high-strength pneumatic pulse backwashing + continuous scraping 3 times at a pressure of 0.5 MPa to ensure that the strategy is accurately matched with the actual state of the filter.

[0086] Next, the filter state prediction parameters are matched with the strategy based on the hierarchical filter self-cleaning strategy, and the real-time prediction parameters are compared with the level thresholds in the hierarchical system one by one. For example, the real-time prediction parameters are clogging degree 25%, adsorption performance medium efficiency, and cumulative use 300 hours. When the prediction parameters show clogging degree 25% (moderate) + adsorption performance medium efficiency + normal use, the corresponding moderate combined cleaning strategy is automatically matched, i.e., elastomer scraper + 0.3 MPa backwashing. If the real-time prediction parameters are clogging degree 45% (severe) + adsorption performance low efficiency + aging, the severe emergency cleaning strategy is matched, i.e., 0.5 MPa backwashing + continuous scraping 3 times, thereby determining the target filter self-cleaning strategy and realizing the dynamic correspondence between the state and the strategy.

[0087] Then, the state prediction parameters are analyzed by adopting the target filter self-cleaning strategy to convert the abstract strategy into specific executable filter self-cleaning strategy parameter thresholds. For example, for the moderate combined cleaning strategy, the motion frequency of the elastic wiper is 3 times per minute, the wiper contact pressure is 0.1 MPa, the pressure range of the pneumatic pulse backwash is 0.25-0.35 MPa, and the flushing duration is 10 seconds per time. These filter self-cleaning strategy parameter thresholds are obtained by the skilled person in the art based on historical cleaning data statistics.

[0088] Then, a filter cleaning effect target function is constructed according to the filter cleaning requirement, which needs to comprehensively consider the core indicators: the recovery rate of the interception efficiency after cleaning (target ≥ 90%) as the primary target, supplemented by the cleaning duration (target ≤ 30 seconds), the gas consumption (target ≤ 0.5 m³), the filter loss rate (target ≤ 0.1% per time), and other constraint conditions. Through weight distribution, the efficiency recovery rate is 0.5, the duration is 0.2, the gas consumption is 0.2, and the loss is 0.1 to construct a multi-objective optimization filter cleaning effect target function, ensuring the balance of cleaning effect, economy, and environmental protection.

[0089] Finally, N intervals are divided within the filter self-cleaning strategy parameter thresholds, and each interval is evaluated based on the cleaning effect target function to determine the target filter self-cleaning strategy parameter through comparison and iterative optimization. The specific steps are described in detail in A441-A443.

[0090] By constructing a hierarchical strategy to adapt to different filter states, accurately matching and analyzing parameters, and multi-objective optimization, the target filter self-cleaning strategy parameter with cleaning effect and economy is finally determined, realizing efficient and intelligent self-cleaning of the filtration equipment.

[0091] Further, step A440 in the method provided by the embodiments of the present application includes: A441: N filter self-cleaning strategy parameter intervals are divided within the filter self-cleaning strategy parameter thresholds.

[0092] A442: Random parameter selection and evaluation are performed within the N filter self-cleaning strategy parameter intervals based on the filter cleaning effect target function to obtain a set of N strategy parameter interval cleaning effects.

[0093] A443: The filter self-cleaning strategy parameter thresholds are compared and selected and iteratively approximated and optimized according to the set of N strategy parameter interval cleaning effects to determine the target filter self-cleaning strategy parameter.

[0094] In one embodiment, when dividing N filter core self-cleaning strategy parameter intervals within the filter core self-cleaning strategy parameter threshold, the interval granularity needs to be determined according to the physical meaning and adjustment accuracy of the parameter. For example, if the parameter threshold is the pneumatic pulse backwash pressure 0.2-0.5 MPa, set N=6, and divide it into 0.2-0.25 MPa, 0.25-0.3 MPa, 0.3-0.35 MPa, 0.35-0.4 MPa, 0.4-0.45 MPa, and 0.45-0.5 MPa six intervals according to the equal interval principle, each interval covers an adjustment range of 0.05 MPa, which not only ensures that the number of intervals is moderate for evaluation, but also reflects the difference in the effect of parameter changes on cleaning.

[0095] Then, based on the filter core cleaning effect objective function, 3-5 specific parameter values are randomly selected from each interval for evaluation, and the cleaning effect is calculated by substituting the values into the objective function. Taking the above pneumatic pulse backwash pressure parameter as an example, select 0.31 MPa, 0.33 MPa, and 0.34 MPa in the 0.3-0.35 MPa interval, simulate the cleaning process respectively, and assume that the corresponding retention efficiency recovery rate (89%, 92%, 90%) and gas consumption (0.42 m³, 0.45 m³, 0.43 m³) are obtained. Then, through the weight distribution of the filter core cleaning effect objective function: efficiency recovery rate 0.5, time length 0.2, gas consumption 0.2, and loss 0.1, the comprehensive score is calculated to form the cleaning effect set of the interval. Similarly, the evaluation of the other five intervals is completed, and finally the cleaning effect set of the N=6 strategy parameter intervals is obtained.

[0096] Finally, according to the N strategy parameter interval cleaning effect set, the interval comparison and selection and iterative approximation optimization of the filter core self-cleaning strategy parameter threshold are performed. First, compare the average scores of each interval, and select 1-2 intervals with the highest scores, for example, the 0.3-0.35 MPa interval has a score of 0.88, and the 0.35-0.4 MPa interval has a score of 0.86. Further subdivide the selected intervals, for example, subdivide the 0.3-0.35 MPa interval into 0.3-0.32 MPa, 0.32-0.34 MPa, and 0.34-0.35 MPa three subintervals, and repeat the random selection and evaluation steps. After three iterations, 0.325 MPa is finally determined as the optimal parameter in the 0.32-0.33 MPa subinterval, thereby determining the target filter core self-cleaning strategy parameter, which is significantly better than the average level of the initial interval.

[0097] By dividing the parameter intervals, the evaluation range is focused, the random selection and evaluation capture the characteristics of the interval, and the multiple iterations approximate the optimal value, finally determining the target self-cleaning strategy parameter that optimizes the cleaning effect of the filter core, improving the precision and economy of the cleaning.

[0098] Further, the step A400 in the method provided by the embodiment of the present application comprises: A450: the self-cleaning mechanism of the device specifically comprises an elastomer blade and a pneumatic pulse backwashing device, wherein the motion trajectory of the elastomer blade is parallel to the axial direction of the modular device filter element, and the modular device filter element is arranged at the nozzle outlet of the pneumatic pulse backwashing device.

[0099] Optionally, the elastomer blade in the self-cleaning mechanism of the device is made of a high-molecular elastomer material, is arranged in parallel with the modular device filter element, and has a motion trajectory extending along the axial direction of the filter element. When the cleaning program is triggered, the elastomer blade reciprocates along the surface of the filter element at a stable speed under the driving of fluid power, and the retained substances on the outer surface of the filter element are stripped by flexible contact, so that the rigid friction between the traditional metal blade and the filter element is avoided, and the possibility of generation of metal foreign matters is eliminated in structure.

[0100] The pneumatic pulse backwashing device is composed of a high-pressure gas source, a control valve and a directional nozzle, one end of the modular device filter element is opposite to the nozzle outlet, and a reasonable distance is maintained between the two to ensure the washing effect. During washing, the device releases a high-pressure gas flow at a set frequency, the gas flow forms a high-speed jet through the nozzle, and directly impacts the internal passage of the filter element to blow out the deep retained substances that are not removed by the blade in the reverse direction, thereby complementing the surface cleaning of the elastomer blade and improving the overall cleaning efficiency.

[0101] The cooperative work of the elastomer blade and the pneumatic pulse backwashing device is linked through the control system of the metal-friction-free device: the elastomer blade first performs surface cleaning to remove most of the visible retained substances; then the pneumatic pulse device is started to process the fine residues by using the impact force of the gas flow; after the cleaning is completed, the system detects the differential pressure of the filter element through a sensor to confirm whether the cleaning effect meets the standard, and if not, the above process is repeated.

[0102] The core design goal of the metal-friction-free device is to generate no metal foreign matters, and the self-cleaning mechanism is a key component for achieving this goal. The elastomer blade and the pneumatic pulse backwashing device used in the self-cleaning mechanism are both free of metal contact components, which avoids metal friction during cleaning, ensures the safe application of the metal-friction-free device in fields sensitive to metal foreign matters such as chemical industry and lithium battery product filtration, and guarantees the stability of the filtration performance through the efficient cooperative cleaning mode.

[0103] By specifying the composition, working mode and cooperative mechanism of the self-cleaning mechanism, it is described how the self-cleaning mechanism as a core component of the metal-friction-free device can achieve efficient self-cleaning while avoiding metal friction, and ensure that the device has environmental protection, safety and filtration efficiency.

[0104] In summary, the self-cleaning filtration method of the metal-friction-free device provided by the embodiment of the present application has the following technical effects: This application designs modular filter cartridges, collects historical cleaning and filtration datasets, and real-time multi-dimensional filtration data streams. Through features extraction and state prediction training, it obtains filter cartridge state prediction parameters, calculates information such as filter cartridge usage status, clogging level, and adsorption performance, and adjusts these parameters based on the analysis and iterative optimization results of the graded filter cartridge self-cleaning strategy. This accurately determines the target filter cartridge self-cleaning strategy parameters, and the cleaning is performed by the self-cleaning mechanism of the metal-free frictionless device. This results in a filtration performance that is both highly efficient and environmentally friendly, suitable for applications such as chemical and lithium battery product filtration, and generates no metal foreign matter. It achieves the technical effects of avoiding the generation of metal foreign matter during filtration, enabling rapid filter cartridge replacement and precise self-cleaning control, and improving filtration efficiency and applicability.

[0105] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a self-cleaning filtration system for a metal-free frictionless device, the system comprising: Modular equipment filter element construction module 1 is used to design modular equipment filter elements. The modular equipment filter element includes a main filter element and multiple spare filter elements. The main filter element and multiple spare filter elements are quickly plugged in and switched through a snap-fit ​​quick-installation structure and a multi-channel reversing valve combination. Each equipment filter element integrates an RFID tag, and the historical cleaning and filtration dataset of the equipment is read through the RFID tag.

[0106] The filter element status collaborative prediction channel construction module 2 is used to construct the filter element status collaborative prediction channel by extracting associated features and training status prediction based on the historical cleaning and filtration dataset of the equipment.

[0107] The filter element state prediction parameter acquisition module 3 is used to deploy a multi-parameter sensor network acquisition device to filter multi-dimensional data streams, and to perform predictive analysis on the multi-dimensional data streams filtered by the device based on the filter element state collaborative prediction channel to obtain filter element state prediction parameters.

[0108] The filter cartridge retention cleaning and filtration execution module 4 is used to construct a graded filter cartridge self-cleaning strategy, analyze the filter cartridge state prediction parameters based on the graded filter cartridge self-cleaning strategy, determine the target filter cartridge self-cleaning strategy parameters, and perform filter cartridge retention cleaning and filtration according to the target filter cartridge self-cleaning strategy parameters through the equipment self-cleaning mechanism.

[0109] Furthermore, the filter element state collaborative prediction channel construction module 2 is used to perform the following steps: According to the filter element data application standard, the device historical cleaning filtration data set is subjected to abnormal data cleaning to obtain an available device cleaning filtration data set; a filter element state label library is constructed, the filter element state label library including a filter element use state, a blockage degree state and an adsorption performance state; according to the filter element state label library, associated feature extraction and feature labeling are performed on the available device cleaning filtration data set to obtain a device cleaning filtration label feature sample set; the device cleaning filtration label feature sample set is subjected to associated feature extraction and migration training optimization respectively to construct a filter element state collaborative prediction channel.

[0110] Further, the filter element state collaborative prediction channel construction module 2 is configured to perform the following steps: According to the filter element state label library, a filter element prediction multi-task framework is constructed; each prediction task in the filter element prediction multi-task framework is subjected to extraction analysis to obtain a source domain multi-task prediction channel set; the source domain multi-task prediction channel set is used to perform classification migration training on the device cleaning filtration label feature sample set respectively to obtain a multi-dimensional filter element state prediction channel; an online learning mechanism is introduced to verify the performance optimization and iterative learning update of the multi-dimensional filter element state prediction channel to construct the filter element state collaborative prediction channel.

[0111] Further, the filter element state collaborative prediction channel construction module 2 is configured to perform the following steps: Each prediction task in the filter element prediction multi-task framework is subjected to extraction analysis, and a source domain filter element filtration state data set is selected, the source domain filter element filtration state data set including a multi-task filtration state data set of different types of filter elements; an LSTM network structure is used to perform identification prediction training on the source domain filter element filtration state data set respectively to obtain a multi-type filter element state prediction channel set; model parameter information of the multi-type filter element state prediction channel set is extracted for aggregated learning optimization to obtain the source domain multi-task prediction channel set.

[0112] Further, the filter element state collaborative prediction channel construction module 2 is configured to perform the following steps: According to the filter element prediction multi-task target, the source domain multi-task prediction channel set is subjected to adaptively associated analysis, and a migration learning strategy is preset; the migration learning strategy is used to perform classification migration training on the device cleaning filtration label feature sample set based on the source domain multi-task prediction channel set respectively to generate a filter element use state branch prediction channel, a blockage degree state branch prediction channel and an adsorption performance state branch prediction channel; the filter element use state branch prediction channel, the blockage degree state branch prediction channel and the adsorption performance state branch prediction channel are integrated and merged to obtain a multi-dimensional filter element state prediction channel.

[0113] Further, the filter element retentate cleaning filtration execution module 4 is configured to perform the following steps: Based on the graded filter element self-cleaning strategy, the filter element state prediction parameters are matched to determine the target filter element self-cleaning strategy; the target filter element self-cleaning strategy is then used to analyze the filter element state prediction parameters to obtain the filter element self-cleaning strategy parameter thresholds; according to the filter element cleaning and filtration requirements, a filter element cleaning effect objective function is constructed; based on the filter element cleaning effect objective function, iterative optimization is performed within the filter element self-cleaning strategy parameter thresholds to determine the target filter element self-cleaning strategy parameters.

[0114] Furthermore, the filter cartridge retention cleaning and filtration execution module 4 is used to perform the following steps: N filter element self-cleaning strategy parameter intervals are obtained within the threshold range of the filter element self-cleaning strategy parameters; based on the objective function of the filter element cleaning effect, parameters are randomly selected and evaluated within the N filter element self-cleaning strategy parameter intervals to obtain a set of cleaning effects for the N strategy parameter intervals; the filter element self-cleaning strategy parameter thresholds are compared and selected and iteratively approximated and optimized according to the set of cleaning effects for the N strategy parameter intervals to determine the target filter element self-cleaning strategy parameters.

[0115] Furthermore, the filter cartridge retention cleaning and filtration execution module 4 is used to perform the following steps: The self-cleaning mechanism of the equipment specifically includes an elastomer scraper and a pneumatic pulse backwashing device, wherein the movement trajectory of the elastomer scraper is parallel to the axial direction of the modular equipment filter element, and the modular equipment filter element is located at the nozzle outlet of the pneumatic pulse backwashing device.

[0116] Example 3, as Figure 3 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an electronic device, the electronic device comprising: The memory 303 is used to store executable instructions; the processor 302 is used to execute the executable instructions stored in the memory 303 to realize the self-cleaning filtration method of the metal frictionless device.

[0117] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, showing a block diagram of an exemplary electronic device suitable for implementing the embodiments of the present invention. Figure 3The electronic device shown is merely an example and should not bring any limitation to the function and use range of the embodiments of the present application. The electronic device is in the form of a general computing device, and its components can include, but are not limited to, an input device 301, a processor 302, a memory 303, and an output device 304. The processor 302 can be one or more; the memory 303 can include a computer readable medium and at least one program product having a set of (at least one) program modules configured to perform the functions of the embodiments of the present application.

[0118] The memory 303 shown in the embodiments of the present application can employ any combination of one or more computer readable media; the computer readable storage media can be, but is not limited to, an infrared ray, a semiconductor system, a device, or an apparatus, or any combination of the above, for storing software programs, computer executable programs, and modules, such as the program instructions / modules corresponding to the self-cleaning filtration method of the non-metallic friction device in the embodiments of the present application. The processor 302 performs various function applications and data processing of the computer device by running the software programs, instructions, and modules stored in the memory 303, i.e., implements the self-cleaning filtration method of the non-metallic friction device.

[0119] The above description of the disclosed embodiments enables those skilled in the art to carry out or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0120] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A self-cleaning filtration method for metal-free tribological equipment, characterized in that, The method comprises: Designing a modular device filter element, the modular device filter element comprising a main filter element and a plurality of backup filter elements, wherein the main filter element and the plurality of backup filter elements are quickly plugged and switched through a combination of a snap-on quick-mount structure and a multi-channel reversing valve, and each device filter element is integrated with an RFID tag, and device historical cleaning and filtering data sets are read through the RFID tag; Based on the device historical cleaning and filtering data sets, associated feature extraction and state prediction training are performed to construct a filter element state collaborative prediction channel; A multi-parameter sensor network is deployed to collect device filtering multi-dimensional data streams, and the device filtering multi-dimensional data streams are predicted and analyzed based on the filter element state collaborative prediction channel to obtain filter element state prediction parameters; A hierarchical filter element self-cleaning strategy is constructed, the filter element state prediction parameters are analyzed based on the hierarchical filter element self-cleaning strategy, target filter element self-cleaning strategy parameters are determined, and filter element retention cleaning and filtering are performed according to the target filter element self-cleaning strategy parameters through a device self-cleaning mechanism; The determination of the target filter element self-cleaning strategy parameters comprises: The filter element state prediction parameters are matched based on the hierarchical filter element self-cleaning strategy to determine a target filter element self-cleaning strategy; The filter element state prediction parameters are analyzed based on the target filter element self-cleaning strategy to obtain filter element self-cleaning strategy parameter thresholds; According to filter cleaning and filtering requirements, a filter cleaning effect objective function is constructed; Based on the filter cleaning effect objective function, iterative optimization is performed within the filter element self-cleaning strategy parameter thresholds to determine target filter element self-cleaning strategy parameters.

2. The self-cleaning filtration method of metal-free tribological equipment according to claim 1, characterized in that, The construction of the filter element state collaborative prediction channel comprises: According to filter data application standards, abnormal data cleaning is performed on the device historical cleaning and filtering data sets to obtain available device cleaning and filtering data sets; A filter element state label library is constructed, the filter element state label library comprising filter element usage state, clogging degree state, and adsorption performance state; According to the filter element state label library, associated feature extraction and feature labeling are performed on the available device cleaning and filtering data sets to obtain a device cleaning and filtering label feature sample set; Respectively, associated feature extraction and transfer training optimization are performed on the device cleaning and filtering label feature sample set to construct a filter element state collaborative prediction channel.

3. The self-cleaning filtration method of metal-free tribological equipment according to claim 2, characterized in that, The construction of the filter element state collaborative prediction channel through the respective associated feature extraction and transfer training optimization of the device cleaning and filtering label feature sample set comprises: According to the filter element state label library, a filter element prediction multi-task framework is constructed; Each prediction task in the filter element prediction multi-task framework is extracted and analyzed to obtain a source domain multi-task prediction channel set; The source domain multi-task prediction channel set is used to respectively perform classification transfer training on the device cleaning and filtering label feature sample set to obtain a multi-dimensional filter element state prediction channel; An online learning mechanism is introduced to verify the performance of the multi-dimensional filter element state prediction channel, optimize and iteratively update the performance, and construct the filter element state collaborative prediction channel.

4. The self-cleaning filtration method of metal-free tribological equipment according to claim 3, characterized in that, The obtaining of the source domain multi-task prediction channel set comprises: The prediction tasks in the filter core prediction multi-task framework are parsed by extraction, a source domain filter core filtering state data set is selected, and the source domain filter core filtering state data set includes a multi-task filtering state data set of different types of filter cores; An LSTM network structure is used to perform identification prediction training on the source domain filter core filtering state data set, and a multi-type filter core state prediction channel set is obtained; Model parameter information of the multi-type filter core state prediction channel set is extracted for aggregated learning optimization, and a source domain multi-task prediction channel set is obtained.

5. The self-cleaning filtration method of metal-free tribological equipment according to claim 3, characterized in that, The multi-dimensional filter core state prediction channel includes: According to the filter core prediction multi-task target, the source domain multi-task prediction channel set is adaptively associated and analyzed, and a preset transfer learning strategy is set; Using the transfer learning strategy, the source domain multi-task prediction channel set is used to perform classification transfer training on the equipment cleaning and filtering label feature sample set respectively, to generate a filter core use state branch prediction channel, a clogging degree state branch prediction channel, and an adsorption performance state branch prediction channel; The filter core use state branch prediction channel, the clogging degree state branch prediction channel, and the adsorption performance state branch prediction channel are integrated and merged to obtain a multi-dimensional filter core state prediction channel.

6. The self-cleaning filtration method of metal-free tribological equipment according to claim 1, characterized in that, The target filter core self-cleaning strategy parameter is determined by performing iterative optimization within the filter core self-cleaning strategy parameter threshold based on the filter core cleaning effect objective function, including: N filter core self-cleaning strategy parameter intervals are divided within the filter core self-cleaning strategy parameter threshold; Parameter random selection evaluation is performed within the N filter core self-cleaning strategy parameter intervals based on the filter core cleaning effect objective function, to obtain a N strategy parameter interval cleaning effect set; The target filter core self-cleaning strategy parameter is determined by interval comparison selection and iterative approximation optimization of the filter core self-cleaning strategy parameter threshold according to the N strategy parameter interval cleaning effect set.

7. The self-cleaning filtration method of metal-free tribological equipment according to claim 1, characterized in that, The device self-cleaning mechanism specifically includes an elastomer wiper and a pneumatic pulse backwashing device, wherein the motion trajectory of the elastomer wiper is axially parallel to the modular device filter core, and the modular device filter core is arranged at the nozzle outlet of the pneumatic pulse backwashing device.

8. A self-cleaning filtration system for metal-free tribological equipment, characterized in that, A self-cleaning filtering method for implementing the metal-friction-free device of any one of claims 1-7, the system comprising: A modular device filter core construction module for designing a modular device filter core, the modular device filter core comprising a main filter core and a plurality of backup filter cores, wherein the main filter core and the plurality of backup filter cores are quickly plugged and switched through a buckle type quick installation structure and a multi-channel reversing valve combination, and each device filter core is integrated with an RFID tag to read a device historical cleaning and filtering data set through the RFID tag; A filter core state collaborative prediction channel construction module for performing associated feature extraction and state prediction training based on the device historical cleaning and filtering data set to construct a filter core state collaborative prediction channel; A filter core state prediction parameter acquisition module for deploying a multi-parameter sensor network to collect device filtering multi-dimensional data streams, and performing prediction analysis on the device filtering multi-dimensional data streams based on the filter core state collaborative prediction channel to obtain filter core state prediction parameters; The filter element retention cleaning filtration execution module is configured to construct a hierarchical filter self-cleaning strategy, perform strategy analysis on the filter state prediction parameters based on the hierarchical filter self-cleaning strategy, determine target filter self-cleaning strategy parameters, and perform filter element retention cleaning filtration through a device self-cleaning mechanism according to the target filter self-cleaning strategy parameters.

9. An electronic device, comprising: The electronic device comprises: a memory for storing executable instructions; a processor for executing the executable instructions stored in the memory to implement the self-cleaning filtration method of the metal-friction-free device according to any one of claims 1 to 7.

Images