Sand mill cleaning system based on intelligent control and control method thereof

Abstract

The application provides a sand mill cleaning system based on intelligent control and a control method thereof, and belongs to the technical field of wet ultrafine grinding / dispersion equipment. The method comprises the following steps: performing time domain decomposition and frequency domain filtering processing on signals collected during the operation of the sand mill, performing feature extraction and abnormal value elimination on the filtered signals, and generating multi-source sensing feature data; performing multi-dimensional correlation labeling processing on the multi-source sensing feature data, and generating a multi-dimensional failure representation feature set; through cross discrimination of three signals of spindle load ripple, differential pressure high-frequency pulsation energy ratio and cylinder wall acoustic emission impact rate, the three failure states of bead bed compaction, rigid wedge and viscoelastic paste can be distinguished when the separation interface pressure difference has not obviously climbed, and the defect that the prior art can only trigger mechanical poking brush when the pressure difference exceeds the limit and cannot identify the failure mechanism is fundamentally overcome.

Description

Technical Field

[0001] This invention proposes a sand mill cleaning system and its control method based on intelligent control, belonging to the technical field of wet ultrafine grinding / dispersion equipment. Background Technology

[0002] As the core equipment for wet ultrafine grinding, the smoothness of the separator at the discharge end of the sand mill directly affects the grinding efficiency and the stability of the product particle size distribution. In existing technologies, differential pressure monitoring or flow rate attenuation is commonly used as the basis for judging blockage in the separator screen or annular gap, and a fixed-cycle mechanical shoveling action or high-pressure flushing procedure is triggered after a preset threshold is reached.

[0003] However, the material processed at the discharge end of a sand mill is a two-phase flow system consisting of grinding media and slurry. The particle size range of grinding media such as zirconia or ceramic beads is typically between 0.1 and 2 millimeters, and the consequences of them passing through the screen mesh are far more severe than in conventional filtration scenarios. In actual operation, the decrease in the flow capacity of the separation interface is not caused by uniform material deposition, but rather by the interplay and competition among three failure modes: bead bed compaction and forward movement, rigid wedging, and viscoelastic caking. In the compacted state, the beads as a whole creep towards the discharge end to form a dense layer with low porosity. In the rigid wedging state, individual large beads or broken bead fragments embed into the screen openings under the action of turbulence and pressure gradient, accompanied by high-frequency impact pulsation. In the viscoelastic caking state, high solids content or resin-based slurry forms a viscoelastic deposit at the openings, and the regeneration rate is higher than the removal frequency. The correct handling methods for the above three failure modes are completely different: bead bed compaction requires pressure reduction combined with resuspension shearing; rigid wedging requires instantaneous shear wave or tangential backflush peeling; and viscoelastic caking requires viscosity reduction or pulsating pressure gradient desorption. Existing technologies simplify the complex working conditions into a crude framework where mechanical action is triggered when the differential pressure exceeds the limit. This not only fails to distinguish the failure mechanism, but the mechanical brushing can also easily scratch the precision separation surface. High-pressure rinsing also cannot fundamentally solve the problem of rapid regeneration of the viscoelastic layer. Therefore, there is an urgent need for an intelligent control method that can identify the failure state of the separation interface and apply targeted parameter modulation. Summary of the Invention

[0004] This invention provides a sand mill cleaning system and its control method based on intelligent control, to solve the problems mentioned in the background art above: The present invention proposes a control method for a sand mill cleaning system based on intelligent control, the method comprising: S1. Perform time-domain decomposition and frequency-domain filtering on the signals collected during the operation of the sand mill. The collected signals include the drive motor current signal, the separation interface differential pressure signal, and the cylinder wall acoustic emission signal. Perform feature extraction and outlier removal on the filtered signals to generate multi-source sensor feature data. Perform multi-dimensional correlation and annotation processing on the multi-source sensor feature data to extract the spindle load ripple amplitude and harmonic ratio features, the separation interface differential pressure high-frequency pulsation energy ratio features, and the cylinder wall acoustic emission impact event rate features to generate a multi-dimensional failure characterization feature set. S2. Competitive failure mode identification analysis is performed on the multidimensional failure characterization feature set. Multi-source feature fusion discrimination technology is used to classify the failure state of the separation interface by mechanism. The failure state is mapped to the preset three-state competitive classification system of bead bed compaction state, rigid wedging state and viscoelastic paste state to generate failure mechanism classification label data. The failure mechanism classification label data is extended by working condition correlation to supplement the implicit parameter modulation requirements under the current separation interface state and generate extended modulation requirement data. S3. Perform control instruction conversion processing on the extended modulation requirement data, convert the failure mechanism classification results into a set of machine-executable parameter modulation instructions, and generate structured modulation instruction data; perform execution strategy matching analysis on the structured modulation instruction data, select the modulation type that matches the current failure mechanism from the preset modulation strategy library, and generate modulation type matching data. S4. Perform parameter boundary delineation processing on the modulation type matching data. Combine the adjustable range of the main shaft speed of the current sand mill, the adjustable range of the feed pump flow rate, and the safety threshold of the pressure difference at the separation interface to limit the parameter space of each matching modulation strategy and generate parameter range limit data for each modulation strategy. Sort the parameter range limit data of each modulation strategy by execution priority and generate modulation strategy call priority sequence data. S5. Perform resource pre-allocation processing on the modulation strategy call priority sequence data, allocate independent execution channels and timing windows for each modulation strategy, and generate modulation resource pre-allocation data; prepare for parallel calls on the modulation resource pre-allocation data and synchronously configure the separate interface status backtracking monitoring channel to generate multi-modulation strategy parallel execution ready signals and effect verification backtracking data.

[0005] The present invention proposes a sand mill cleaning system based on intelligent control, wherein the system comprises: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0006] The beneficial effects of this invention are as follows: By cross-discriminating three signals—spindle load ripple, differential pressure high-frequency pulsating energy ratio, and cylinder wall acoustic emission impact rate—it can distinguish between three failure states: bead bed compaction, rigid wedging, and viscoelastic agglomeration—before the pressure difference at the separation interface has significantly increased. This fundamentally overcomes the shortcomings of existing technologies that only trigger mechanical brushing based on pressure difference exceeding limits without identifying the failure mechanism. For the bead bed compaction state, the system uses pressure reduction and speed pulse modulation to resuspend the bead bed, avoiding the problem that mechanical brushing would further push the beads towards the screen. For the rigid wedging state, instantaneous shear waves combined with directional backwashing are used to peel off the embedded beads, eliminating the risk of scratching the screen apertures by vigorous brushing. For the viscoelastic agglomeration state, the viscoelastic deposit layer is loosened by pulsating pressure gradient and micro-vibration desorption, solving the problem that high-pressure washing cannot suppress the rapid regeneration of the viscoelastic layer. This method transforms the cleaning action from a crude intervention involving external mechanical attachments to a short-term parameter modulation of the grinding process itself. This protects the integrity of the precision separation surface, maintains the continuity of the grinding operation, and significantly reduces the frequency of unplanned downtime and the cost of screen replacement. Attached Figure Description

[0007] Figure 1 This is a diagram illustrating the steps of the method described in this invention; Figure 2 This is a detailed flowchart of step S4 in this invention. Detailed Implementation

[0008] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0009] One embodiment of the present invention, such as Figure 1 As shown, a control method for a sand mill cleaning system based on intelligent control is described, the method comprising: S1. Perform time-domain decomposition and frequency-domain filtering on the signals collected during the operation of the sand mill. The collected signals include the drive motor current signal, the separation interface differential pressure signal, and the cylinder wall acoustic emission signal. Perform feature extraction and outlier removal on the filtered signals to generate multi-source sensor feature data. Perform multi-dimensional correlation and annotation processing on the multi-source sensor feature data to extract the spindle load ripple amplitude and harmonic ratio features, the separation interface differential pressure high-frequency pulsation energy ratio features, and the cylinder wall acoustic emission impact event rate features to generate a multi-dimensional failure characterization feature set. S2. Competitive failure mode identification analysis is performed on the multidimensional failure characterization feature set. Multi-source feature fusion discrimination technology is used to classify the failure state of the separation interface by mechanism. The failure state is mapped to the preset three-state competitive classification system of bead bed compaction state, rigid wedging state and viscoelastic paste state to generate failure mechanism classification label data. The failure mechanism classification label data is extended by working condition correlation to supplement the implicit parameter modulation requirements under the current separation interface state and generate extended modulation requirement data. S3. Perform control instruction conversion processing on the extended modulation requirement data, convert the failure mechanism classification results into a set of machine-executable parameter modulation instructions, and generate structured modulation instruction data; perform execution strategy matching analysis on the structured modulation instruction data, select the modulation type that matches the current failure mechanism from the preset modulation strategy library, and generate modulation type matching data. S4. Perform parameter boundary delineation processing on the modulation type matching data. Combine the adjustable range of the main shaft speed of the current sand mill, the adjustable range of the feed pump flow rate, and the safety threshold of the pressure difference at the separation interface to limit the parameter space of each matching modulation strategy and generate parameter range limit data for each modulation strategy. Sort the parameter range limit data of each modulation strategy by execution priority and generate modulation strategy call priority sequence data. S5. Perform resource pre-allocation processing on the modulation strategy call priority sequence data, allocate independent execution channels and timing windows for each modulation strategy, and generate modulation resource pre-allocation data; prepare for parallel calls on the modulation resource pre-allocation data and synchronously configure the separate interface status backtracking monitoring channel to generate multi-modulation strategy parallel execution ready signals and effect verification backtracking data.

[0010] The working principle and effects of the above technical solution are as follows: By jointly analyzing multi-source sensor signals and accurately identifying three-state failures, the accuracy of determining the failure of the sand mill's separation interface can be improved, the adaptability of cleaning modulation commands to equipment failure conditions can be enhanced, and the frequency of ineffective cleaning operations can be reduced. By matching dedicated modulation cleaning strategies in a graded manner, the equipment operating losses caused by blindly adjusting sand mill parameters can be reduced, and the probability of slurry agglomeration and media jamming recurrence can be reduced. By constraining the entire domain of hardware parameter boundaries and prioritizing strategies, spindle overload and interface pressure difference exceeding limits caused by parameter exceeding threshold adjustments can be avoided, and the control operation exceeding the equipment's bearing capacity can be avoided, which would lead to accelerated wear of the cavity structure. By allocating differentiated channel timing resources in combination with status backtracking monitoring, multiple sets of cleaning strategies can be executed smoothly in parallel, and the cleaning and rectification effects can be verified in real time, reducing the workload of manual on-site adjustment, enhancing the adaptability and operational stability of sand mill cleaning control, and ensuring continuous and stable grinding operations.

[0011] In one embodiment of the present invention, S1 includes: S11. Synchronously collect multiple types of raw operating sensor signals during the continuous operation of the sand mill. The raw signals include the real-time current waveform signal of the drive motor, the dynamic differential pressure continuous sampling signal of the separation interface inside the cylinder, and the distributed acoustic emission pulse signal of the outer wall of the grinding cylinder. All raw signals are uniformly converted into standardized digital sampling sequences with equal time steps to generate a multi-channel raw sampling sequence set for the sand mill. S12. Perform time-domain moving average noise reduction and segmented frequency-domain band-stop filtering operations on the multi-channel original sampling sequence set of the sand mill in sequence to filter out three types of interference components: power grid frequency harmonics, fixed frequency noise from mechanical foundation vibration, and random pulse noise from the feed fluid, and output a multi-channel clean filtered sampling dataset. S13. Traverse the multi-channel clean filter sampling dataset and perform three-dimensional feature mining of time-domain extreme value features, frequency-domain energy distribution features, and pulse event statistical features for each channel. Simultaneously remove global abnormal sampling points caused by sampling jumps and instantaneous start-up and shutdown of equipment, extract stable and effective sensing feature components, and generate standardized multi-source sensing feature data. S14. Using the three cleaning-related dimensions of grinding media accumulation state, grinding slurry adhesion degree, and separation ring blockage degree as annotation benchmarks, cross-signal dimension coupling and association annotation is completed for all feature components in the standardized multi-source sensor feature data. The combined features of the periodic ripple amplitude of the spindle load and the proportion of each harmonic energy, the ratio of high-frequency pulse segment energy to low-frequency steady-state segment energy of the differential pressure at the separation interface, and the frequency of instantaneous impact events of acoustic emission from the cylinder wall per unit time are extracted and superimposed to complete the set integration and generate a complete multi-dimensional failure characterization feature set.

[0012] The working principle and effects of the above technical solution are as follows: By synchronously acquiring multiple types of sensor signals and standardizing the timing process, the synchronization and regularity of the sand mill operating condition signal acquisition can be improved, avoiding the problem of feature matching mismatch caused by different signal sampling time differences. Through dual-layer noise reduction filtering, various environmental and equipment noise interferences can be filtered out, reducing the interference of invalid noise data on operating condition analysis and improving the purity of the sensor data itself. Through three-dimensional feature mining combined with abnormal sampling point removal, the deviation of failure features caused by instantaneous operating condition disturbances can be reduced, enhancing the stability of extracted sensor features. By extracting exclusive failure features through multi-dimensional operating condition coupling annotation, it can not only conform to the actual occurrence mechanism of the three types of faults at the separation interface, but also refine the differentiated characteristics of faults, reducing the probability of misjudgment in subsequent failure mode identification and ensuring the reliability of the data base for the entire process of subsequent fault analysis, cleanup, and control.

[0013] In one embodiment of the present invention, S2 includes: S21. Input the complete multidimensional failure characterization feature set into the competitive failure identification operation unit, build a multidimensional feature coupling discrimination operation model, calculate the independent discrimination weight of single-class sensing features and the comprehensive discrimination weight of multi-class feature fusion respectively, complete the layer-by-layer decomposition of the underlying physical mechanism of different failure manifestations of the separation interface, and output the preliminary discrimination result of interface failure. S22. The preliminary judgment results of interface failure are assigned to the pre-built three-state competitive classification system, and matched with the media accumulation failure category corresponding to bead bed compaction, the hard particle jamming failure category corresponding to rigid wedge, and the slurry adhesion and agglomeration failure category corresponding to viscoelastic paste. Each set of feature samples is matched with a unique classification label, and failure mechanism classification label data with sample index is generated. S23. Retrieve the static and dynamic fluctuation parameters of the current real-time operating conditions of the sand mill. The static parameters include the total amount of media filling and the initial solid content of the slurry. The dynamic parameters include the instantaneous feed flow rate and the real-time spindle speed. Bind the two types of operating condition parameters to the failure mechanism classification label data sample by sample. S24. Based on the completed working conditions and failure corresponding sample groups, deduce the three types of adjustment constraints required for the cleaning operation under different failure states: the change range of adjustment parameters, the duration of adjustment, and the start and stop sequence of adjustment. Supplement the data into the corresponding sample entries of the tag data to generate extended modulation requirement data covering working conditions and adjustment constraints.

[0014] The working principle and effects of the above technical solution are as follows: By using feature coupling and weighted discrimination to disassemble the underlying failure mechanism layer by layer, the accuracy of failure judgment at the sand mill separation interface can be improved, reducing the judgment bias caused by single sensor feature judgment. The failure category classification is completed through a three-state competitive classification system, which can refine the type boundaries of different interface faults and reduce the probability of confusion and misclassification of different failure conditions. By binding the dynamic and static operating parameters of the entire domain with the failure labels from the same source, it can closely match the actual grinding conditions on site, supplement the missing operating condition correlation information in failure judgment, and enhance the on-site adaptability of failure classification results. By deducing and cleaning up control constraints and expanding data content through operating condition samples, it can not only adapt to specific control standards for various failures, but also save the workload of manual judgment of control parameters, avoiding the problem of poor cleaning and rectification effects due to the disconnect between subsequent control commands and on-site operating conditions.

[0015] In one embodiment of the present invention, step S21 includes: The complete multidimensional failure characterization feature set is retrieved and imported into the competing failure identification operation unit. Three types of original feature data are collected: spindle load ripple features, differential pressure high-frequency pulsation features, and acoustic emission impact event features. Feature time sequence alignment and feature magnitude normalization are completed to generate a normalized failure feature dataset. A feature-layered coupling calculation was performed on the normalized failure feature dataset to build a multi-dimensional feature coupling discrimination operation model adapted to the separation interface working condition. The model was divided into an inner layer single feature operation branch and an outer layer multi-feature fusion operation branch to generate model branch architecture data. The model performs independent discrimination operations on the three types of original features one by one through the inner layer operation branches, iteratively calculates the independent discrimination weight of each sensing feature, and generates a single feature independent weight dataset; the model superimposes the global feature linkage relationship through the outer layer operation branches, calculates the multi-feature linkage fusion comprehensive discrimination weight, and generates a multi-feature fusion weight dataset. The system performs coupled judgment operations by linking single-feature independent weight datasets and multi-feature fusion weight datasets, and decomposes the physical mechanisms of failure at the bottom layer of three types of separation interfaces: media extrusion and compaction, hard particle embedding, and slurry cross-linking and adhesion. It also decomposes the abnormal equipment operation performance corresponding to each mechanism and generates layered decomposition data of failure mechanisms. By integrating failure mechanism data and breaking it down into layers, a comprehensive judgment of the failure status across the entire domain is completed. The judgment results of failure anomalies in each sampling period are summarized, invalid judgment information caused by floating-point errors in model calculations is removed, and preliminary judgment results of interface failures are output.

[0016] The working principle and effects of the above technical solution are as follows: By performing time-series alignment and magnitude normalization on the three types of original failure features, the calculation standards for multiple types of sensor features can be unified, improving the adaptability of feature coupling calculation and avoiding the problem of calculation result distortion caused by inconsistent feature scales. By building a two-layer branched coupling discrimination model and separating independent and fusion calculation processes, the feature weight calculation dimension can be refined, improving the fit of failure weight assignment. By deconstructing the underlying physical mechanism of interface failure in layers, the operational anomalies of different faults can be distinguished, reducing the misjudgment of failure mechanisms under similar operating conditions. By eliminating invalid judgment information derived from model floating-point errors, the interference of algorithm calculation errors on the judgment results can be reduced, enhancing the credibility of the preliminary failure judgment results. It can take into account both accurate judgment of single-type features and comprehensive judgment of multi-feature linkage, improving the identification ability of hidden failures and composite failures of the separated interface, and consolidating the data foundation for backend failure classification.

[0017] In one embodiment of the present invention, S3 includes: S31. Decompose the adjustment constraints carried by each sample in the extended modulation requirement data, and convert the adjustment constraints corresponding to the three types of failures, namely, bead bed compaction, rigid wedging, and viscoelastic caking, into digital instruction fields that can be parsed by the controller. The fields are further subdivided into speed adjustment fields, feed flow adjustment fields, and backwashing medium push pressure adjustment fields. All fields are integrated to form ordered instruction entries and generate structured modulation instruction data. S32. Build a hierarchical modulation strategy library that stores different cleaning and adjustment logics. The entries in the library are divided into four types of logic: single parameter fine-tuning, multi-parameter coordinated linkage, short-time pulse impact cleaning, and continuous steady-state unblocking cleaning. Each strategy stores a unique feature matching threshold range. S33. Extract the failure feature threshold range contained in the instruction field inside the structured modulation instruction data one by one, and perform interval overlap comparison calculation with the preset matching threshold range of each entry in the hierarchical modulation strategy library. Select the single modulation strategy with the highest overlap value as the current sample adaptation strategy and generate one-to-one corresponding modulation type matching data. S34. Statistically match all modulation strategy entries in the modulation type matching data, merge and summarize the adjustment requirements of multiple samples with the same strategy, remove duplicate matching sample entries, and output a unified modulation matching summary dataset after integration without duplicate entries.

[0018] The working principle and effects of the above technical solution are as follows: By converting different failure constraints into standardized digital instruction fields, the standard for identifying equipment controller instructions can be unified, improving the readability of operating condition control instructions and avoiding the problem of non-standard instructions being unrecognizable by the equipment, leading to stagnation in cleaning and control. By dividing the sand mill into four categories of differentiated cleaning and control strategies, it can cover all scenarios of faults from mild blockage to severe agglomeration, enhancing the ability of strategies to adapt to all failure conditions. By comparing and selecting suitable control strategies through threshold interval overlap, the matching accuracy between failure conditions and cleaning strategies can be improved, reducing ineffective control operations caused by strategy mismatch. By merging similar sample requirements and eliminating duplicate adaptation entries, the backend control computation volume can be simplified, reducing redundant computing power consumption of the controller. This ensures accurate invocation of single-fault dedicated strategies and reduces the time spent on matching strategies for the entire machine, improving the response speed of sand mill fault cleaning and control.

[0019] In one embodiment of the present invention, S33 includes: The system iterates through all ordered instruction entries in the structured modulation instruction data, removes redundant operation fields, and extracts the three core failure feature threshold ranges embedded in each entry: failure level threshold, parameter adjustment amplitude threshold, and operation duration threshold, to generate single-entry feature threshold extraction data. Retrieve the inherent matching threshold ranges of all policy entries stored in the hierarchical modulation policy library, uniformly calibrate the threshold sampling accuracy and numerical magnitude, complete the global policy threshold standardization process, and generate a standardized policy threshold dataset in the library. A two-way interval coupling comparison operation is performed between the single-item feature threshold extraction data and the standardized in-library strategy threshold dataset to calculate the overlap ratio of each threshold interval and generate global threshold overlap calculation data. Filter the modulation strategies corresponding to the peak values ​​in the global threshold overlap calculation data, eliminate redundant alternative strategies caused by the overlap of thresholds of multiple strategies, lock the exclusive cleanup modulation strategies for a single sample, and generate initial sample strategy selection data. Each instruction entry is bound to the corresponding modulation strategy encoding information, completing the one-to-one coupling and archiving of instructions and strategies, organizing the full-domain coupled archive information, and generating one-to-one corresponding modulation type matching data.

[0020] The working principle and effects of the above technical solution are as follows: By stripping redundant fields from modulation instructions and extracting three types of core threshold intervals, the volume of data for strategy comparison calculations can be simplified, improving the efficiency of subsequent threshold comparison calculations and avoiding the problem of strategy matching delay caused by redundant fields occupying controller computing power. By unifying the precision level of policy thresholds within the library, comparison bias caused by differences in threshold measurement standards can be eliminated, enhancing the fairness of bidirectional threshold coupling comparison results. By calculating the overlap ratio of intervals to complete the initial strategy selection, suitable solutions can be accurately selected to meet fault control requirements, reducing the probability of misselection of strategies under similar operating conditions. By eliminating redundant alternative strategies from the same source, it can be ensured that a single fault is matched only with a dedicated control solution, reducing the risk of conflicts between multiple parallel strategies. It can both take into account the differentiated matching requirements of different failure conditions and standardize the instruction strategy archiving logic, improving the uniqueness and execution stability of sand mill fault cleaning strategy matching.

[0021] In one embodiment of the present invention, S34 includes: Traverse the entire domain sample coupling entries of modulation type matching data, collect three types of related information bound to a single sample: modulation strategy code, adaptation adjustment parameters, and failure condition attributes, and integrate them to form strategy adaptation ledger data with sample traceability numbers. Cluster the policy adaptation ledger data according to the modulation policy code, collect all related samples belonging to the same modulation policy, integrate the differential adjustment constraint information of similar samples, and generate aggregated demand data for the same policy. Verify the internal working condition adaptation boundaries, parameter adjustment amplitudes, and execution timing node overlap of the aggregated data with the same strategy. Filter out redundant adaptation sample entries with highly overlapping parameters and identical working conditions to generate simplified clustering strategy data. The data format and field arrangement rules of the clustering strategy data are standardized and streamlined. The annotation information of the full-domain working condition adaptation for each strategy is supplemented. All standardized independent strategy entries are integrated, and a unified modulation matching summary dataset with no duplicate entries is output.

[0022] The working principle and effects of the above technical solution are as follows: By aggregating strategy codes, adjusting parameters, and generating traceability ledger data, the traceability information of the entire strategy matching process can be improved, enhancing the convenience of subsequent strategy review and maintenance, and avoiding the problem of not being able to trace historical control schemes after a fault recurrence. By integrating similar strategy control requirements through coding clustering, scattered and repetitive control requests can be merged, reducing the frequency of the controller repeatedly calling the same cleaning strategy. By verifying the overlap of operating conditions and parameters and filtering out redundant samples from the same source, the data storage volume of the whole machine control can be compressed, reducing the local storage resource occupation of the sand mill. By unifying the data format and supplementing the operating condition adaptation annotations, the data interface barrier between strategy data and backend parameter limitation links can be broken down, which can both retain the space for fine-tuning of different operating conditions and avoid parameter command conflicts caused by repeated strategy calls, improving the smoothness of the whole machine cleaning control process.

[0023] One embodiment of the present invention, such as Figure 2 As shown, S4 includes: S41. Retrieve the factory-calibrated hardware operating limit parameters of the sand mill equipment. The hardware operating limit parameters include the upper and lower fluctuation limit range of the spindle speed, the maximum and minimum adjustable range of the continuous output flow of the feed pump, and the safe threshold range of the differential pressure at the separation interface for long-term stable operation. S42. Read the adjustment instruction field corresponding to each strategy in the unified modulation matching summary dataset, perform interval intersection operation between the parameter to be adjusted in the field and the hardware operation limit parameter, remove the over-limit parameter range that exceeds the device hardware carrying capacity, retain the effective adjustment parameter range that the device can operate safely, and generate independent parameter range limit data for a single modulation strategy. S43. Compare the failure hazard level values ​​corresponding to different modulation strategies. Bead bed compaction corresponds to low hazard level, rigid wedging corresponds to medium hazard level, and viscoelastic paste corresponds to high hazard level. The higher the hazard level value, the earlier the strategy is executed. The synchronous superposition parameter adjustment time index is used for secondary correction, and the short-term effective strategy is appropriately moved forward. S44. Perform linear sorting on all single modulation strategy independent parameter range-limited data according to the corrected order, bind each data to a unique order number, and generate modulation strategy call priority sequence data with ordered numbers.

[0024] The working principle and effects of the above technical solution are as follows: By retrieving the original factory hardware limit parameters for the entire range, the safe operating boundaries of each component of the sand mill can be anchored, improving the compliance of parameter control. By eliminating over-limit control ranges through parameter interval intersection calculations, the excessive parameter control amplitude can be avoided, preventing problems such as spindle overspeed, feed flow overload, and excessive interface pressure difference that lead to wear and tear on equipment parts. By combining the failure hazard level with the control time for dual judgment and execution priority, high-risk agglomeration failures can be prioritized, balancing the urgency of the failure with the efficiency of the cleaning operation, and reducing the duration of severe failures continuously eroding the grinding chamber. By unifying the priority numbering and standardizing the strategy calling order, the parallel calling logic of multiple strategies can be streamlined, reducing the probability of interference and conflict between multiple sets of cleaning instructions. This ensures that the equipment completes the cleaning operation within the safe parameter range throughout the process, optimizes the fault handling sequence, and improves the fault handling efficiency in complex and complex failure scenarios of the sand mill.

[0025] In one embodiment of the present invention, S42 includes: Read each modulation strategy entry in the unified modulation matching summary dataset line by line, parse the three types of original adjustment instruction fields embedded in the entry: spindle control parameters, feed flow control parameters, and medium push pressure control parameters, extract the original parameter floating range of each field, and generate the original control parameter range data of the strategy. By benchmarking the full-domain hardware operation limit parameters retrieved by S41, we split three types of hardware parameters into dedicated safe and controllable ranges, unify the numerical accuracy and sampling measurement standards of the original control parameter range and the hardware limit range, and generate standardized dual-group range comparison data. Perform bidirectional interval intersection calculations on the standardized two-group interval comparison data parameter by parameter, divide the interval into two categories: parameter compliant overlapping intervals and parameter exceeding the standard extension intervals, and generate full-domain parameter interval division result data; Remove all out-of-range extension intervals in the result data of the global parameter interval division, retain the compliant overlapping intervals that are compatible with the hardware carrying capacity of the device, simultaneously mark the failure condition type of the interval adaptation, and generate a single modulation strategy independent parameter range limit data.

[0026] The working principle and effects of the above technical solution are as follows: By extracting the original floating ranges of the three types of control parameters, the adjustment requirements inherent in each cleaning strategy can be fully preserved, improving the information completeness of the parameter verification process and avoiding the problem of interruption in the verification process due to the lack of local control parameters. By unifying the numerical precision and measurement standards of the two sets of intervals, the calculation deviation caused by differences in numerical scales can be eliminated, enhancing the reliability of the interval intersection calculation results. By dividing the compliant and excessive intervals through bidirectional intersection calculation, the safe adjustable range and the prohibited operating range of the equipment can be clearly distinguished, reducing the workload of manually checking the parameter boundaries one by one. By completely eliminating the excessive extension range and marking the suitable working conditions, the control commands that damage the equipment can be intercepted from the source, avoiding the main shaft, feed pump, and separation interface from being subjected to excessive loads for a long time, which would cause accelerated component wear. It can not only fully preserve the adjustment range of the adapted fault, but also maintain the bottom line of safe operation of the equipment hardware throughout the process, extending the service life of the core components of the sand mill.

[0027] In one embodiment of the present invention, step S5 includes: S51. Divide the internal parallel computing hardware resources of the sand mill controller. The hardware resources include three types of resource units: independent parameter adjustment computing channel, real-time sensor data feedback monitoring channel, and cleaning action execution output timing window. Calculate the total number of strategies in the modulation strategy call priority sequence data. S52. According to the priority allocation logic in the sequence from high to low, allocate a set of independent parameter adjustment operation channels and execution timing windows that do not occupy each other for each modulation strategy, record the resource unit number and timing start and end nodes occupied by each strategy, and generate complete modulation resource pre-allocation data. S53. Based on the start and end nodes of the timing window of each strategy in the pre-allocation data of modulation resources, set up multi-channel synchronous trigger signals to synchronously open the dedicated backtracking monitoring channel of the separation interface status. The monitoring channel continuously collects two types of verification sensor data: differential pressure of the separation interface and acoustic emission of medium impact throughout the cleaning operation. S54. Integrate multi-channel synchronous trigger signals into a unified device identification ready signal, synchronously collect the verification sensor data collected throughout the monitoring channel and store it in time-series segments to form a status backtracking sample, merge the ready signal and the time-series segmented backtracking sample, and generate a multi-modulation strategy parallel execution ready signal and effect verification backtracking data.

[0028] The working principle and effects of the above technical solution are as follows: By pre-splitting multiple hardware resource units within the controller and calculating the total number of strategies to be executed, the upper limit of the overall computing power can be determined in advance, improving the rationality of resource allocation planning and avoiding the problem of computing power overload and lag after multiple sets of cleaning tasks are started simultaneously. By allocating independent computing channels and timing windows according to priority, instruction conflicts caused by different control strategies competing for hardware resources can be eliminated, reducing the probability of parallel control process failure. By synchronously collecting interface status data through a dedicated monitoring channel, the entire process of cleaning operation changes can be completely retained, enhancing the data support for reviewing the effectiveness of fault rectification. By integrating trigger signals and segmented stored monitoring samples, start commands and verification materials can be output simultaneously, enabling multiple cleaning strategies to run synchronously without interference, and quickly determining whether the cleaning operation has met the expected standards based on complete timing records, reducing the frequency of manual on-site inspection and verification.

[0029] In one embodiment of the present invention, S52 includes: Read the full-domain sequential encoding information of the modulation strategy call priority sequence data, verify the parameter computing power requirements and action execution time requirements of each sequential modulation strategy, and generate single-strategy resource adaptation requirement data; Retrieve the idle independent parameter adjustment operation channels and blank execution timing window resource ledgers of S51 partitioning, check the channel occupancy markers and timing overlap markers, filter reusable idle hardware resource units, and generate available hardware resource survey data. Following the priority arrangement rule from high to low, the single strategy resource adaptation requirement data and available hardware resource survey data are matched sequentially. Each priority modulation strategy is matched with a dedicated independent parameter adjustment calculation channel. The start and end times of each strategy execution are staggered to avoid time sequence overlap and crowding, and preliminary strategy resource matching data is generated. Collect three types of resource binding information for each strategy matching channel: exclusive code, time series start node, and time series end node. Perform full-domain compliance verification on the resource binding information, investigate potential risks of minor resource crowding, correct and fine-tune the time series layout points, and generate strategy resource binding archive data. Integrate all verified policy resource binding archive data, standardize data field formats, supplement resource allocation traceability markers, and generate complete modulation resource pre-allocation data.

[0030] The working principle and effects of the above technical solution are as follows: By verifying the computing power and time requirements of each control strategy to form matching demand data, the resource consumption scale of each cleanup task can be accurately grasped, improving the fit of hardware resource allocation and avoiding the problem of resource waste caused by small-load tasks occupying a large amount of idle resources. By identifying idle channels and timing windows in advance, unoccupied hardware units can be quickly screened, shortening the overall resource matching time. By staggering the execution time of each task according to the fault handling priority, the situation of multiple tasks overlapping and competing for channels can be completely eliminated, reducing the probability of control command errors and failures. By verifying resource binding information across the entire domain and correcting the timing arrangement, the hidden dangers of minor resource crowding can be eliminated, enhancing the stability of the parallel control process. It can both prioritize the cleanup tasks corresponding to severe failures to seize hardware resources and completely retain the traceability information of the entire resource allocation process, facilitating the later tracing of resource scheduling anomalies.

[0031] One embodiment of the present invention provides a sand mill cleaning system based on intelligent control, the system comprising: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0032] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A control method for a sand mill cleaning system based on intelligent control, characterized in that, The method includes: S1. Perform time-domain decomposition and frequency-domain filtering on the signals collected during the operation of the sand mill. Extract features and remove outliers from the filtered signals to generate multi-source sensor feature data. Perform multi-dimensional association and annotation processing on the multi-source sensor feature data to generate a multi-dimensional failure characterization feature set. S2. Competitive failure mode identification analysis is performed on the multi-dimensional failure characterization feature set. Multi-source feature fusion discrimination technology is used to classify the failure state of the separation interface by mechanism. The failure state is mapped to the preset three-state competitive classification system of bead bed compaction state, rigid wedging state and viscoelastic paste state to generate failure mechanism classification label data. The failure mechanism classification label data is extended by working condition correlation to generate extended modulation requirement data. S3. Perform control command conversion processing on the extended modulation requirement data to generate structured modulation command data; perform execution strategy matching analysis on the structured modulation command data, select the modulation type that matches the current failure mechanism from the preset modulation strategy library, and generate modulation type matching data. S4. Perform parameter boundary delineation processing on the modulation type matching data. Combine the adjustable range of the main shaft speed of the current sand mill, the adjustable range of the feed pump flow rate, and the safety threshold of the pressure difference at the separation interface to limit the parameter space of each matching modulation strategy and generate parameter range limit data for each modulation strategy. Sort the parameter range limit data of each modulation strategy by execution priority and generate modulation strategy call priority sequence data. S5. Perform resource pre-allocation processing on the modulation strategy call priority sequence data to generate modulation resource pre-allocation data; prepare for parallel call of the modulation resource pre-allocation data and synchronously configure the separation interface status backtracking monitoring channel to generate multi-modulation strategy parallel execution ready signal and effect verification backtracking data.

2. The control method for the sand mill cleaning system based on intelligent control according to claim 1, characterized in that, S1 includes: S11. Synchronously collect raw signals from multiple operating sensors during the continuous operation phase of the sand mill, and convert all raw signals into standardized digital sampling sequences with equal time steps to generate a multi-channel raw sampling sequence set for the sand mill. S12. Perform time-domain moving average noise reduction and piecewise frequency-domain band-stop filtering operations sequentially on the multi-channel original sampling sequence set of the sand mill to output a multi-channel clean filtered sampling dataset. S13. Traverse the multi-channel clean filter sampling dataset and perform three-dimensional feature mining of time-domain extreme value features, frequency-domain energy distribution features, and pulse event statistical features for each channel. Simultaneously remove global abnormal sampling points caused by sampling jumps and instantaneous start-up and shutdown of equipment, extract stable and effective sensing feature components, and generate standardized multi-source sensing feature data. S14. Using the three types of cleanup association dimensions as the annotation benchmark, complete the cross-signal dimension coupling association annotation of all feature components in the standardized multi-source sensing feature data, and superimpose the three types of features to complete the set integration and generate a complete multi-dimensional failure characterization feature set.

3. The control method for the sand mill cleaning system based on intelligent control according to claim 1, characterized in that, The S2 includes: S21. Input the complete multidimensional failure characterization feature set into the competitive failure identification operation unit, build a multidimensional feature coupling discrimination operation model, calculate the independent discrimination weight of single-class sensing features and the comprehensive discrimination weight of multi-class feature fusion respectively, complete the layer-by-layer decomposition of the underlying physical mechanism of different failure manifestations of the separation interface, and output the preliminary discrimination result of interface failure. S22. The preliminary judgment results of interface failure are assigned to the pre-built three-state competitive classification system, and matched with the media accumulation failure category corresponding to bead bed compaction, the hard particle jamming failure category corresponding to rigid wedge, and the slurry adhesion and agglomeration failure category corresponding to viscoelastic paste. Each set of feature samples is matched with a unique classification label, and failure mechanism classification label data with sample index is generated. S23. Retrieve the static and dynamic fluctuation parameters of the current real-time operating conditions of the sand mill. The static parameters include the total amount of media filling and the initial solid content of the slurry. The dynamic parameters include the instantaneous feed flow rate and the real-time spindle speed. Bind the two types of operating condition parameters to the failure mechanism classification label data sample by sample. S24. Based on the completed working conditions and failure corresponding sample groups, deduce the three types of adjustment constraints required for the cleaning operation under different failure states: the change range of adjustment parameters, the duration of adjustment, and the start and stop sequence of adjustment. Supplement the data into the corresponding sample entries of the tag data to generate extended modulation requirement data covering working conditions and adjustment constraints.

4. The control method for the sand mill cleaning system based on intelligent control according to claim 3, characterized in that, S21 includes: The complete multidimensional failure characterization feature set is retrieved and imported into the competing failure identification operation unit. Three types of original feature data are collected: spindle load ripple features, differential pressure high-frequency pulsation features, and acoustic emission impact event features. Feature time sequence alignment and feature magnitude normalization are completed to generate a normalized failure feature dataset. A feature-layered coupling calculation was performed on the normalized failure feature dataset to build a multi-dimensional feature coupling discrimination operation model adapted to the separation interface working condition. The model was divided into an inner layer single feature operation branch and an outer layer multi-feature fusion operation branch to generate model branch architecture data. The model performs independent discrimination operations on the three types of original features one by one through the inner layer operation branches, iteratively calculates the independent discrimination weight of each sensing feature, and generates a single feature independent weight dataset; the model superimposes the global feature linkage relationship through the outer layer operation branches, calculates the multi-feature linkage fusion comprehensive discrimination weight, and generates a multi-feature fusion weight dataset. The system performs coupled judgment operations by linking single-feature independent weight datasets and multi-feature fusion weight datasets, and decomposes the physical mechanisms of failure at the bottom layer of three types of separation interfaces: media extrusion and compaction, hard particle embedding, and slurry cross-linking and adhesion. It also decomposes the abnormal equipment operation performance corresponding to each mechanism and generates layered decomposition data of failure mechanisms. By integrating failure mechanism data and breaking it down into layers, a comprehensive judgment of the failure status across the entire domain is completed. The judgment results of failure anomalies in each sampling period are summarized, invalid judgment information caused by floating-point errors in model calculations is removed, and preliminary judgment results of interface failures are output.

5. The control method for the sand mill cleaning system based on intelligent control according to claim 1, characterized in that, The S3 includes: S31. Decompose the adjustment constraints carried by each sample in the extended modulation requirement data, and convert the adjustment constraints corresponding to the three types of failures, namely, bead bed compaction, rigid wedging, and viscoelastic caking, into digital instruction fields that can be parsed by the controller. The fields are further subdivided into speed adjustment fields, feed flow adjustment fields, and backwashing medium push pressure adjustment fields. All fields are integrated to form ordered instruction entries and generate structured modulation instruction data. S32. Build a hierarchical modulation strategy library that stores different cleaning and adjustment logics. The entries in the library are divided into four types of logic: single parameter fine-tuning, multi-parameter coordinated linkage, short-time pulse impact cleaning, and continuous steady-state unblocking cleaning. Each strategy stores a unique feature matching threshold range. S33. Extract the failure feature threshold range contained in the instruction field inside the structured modulation instruction data one by one, and perform interval overlap comparison calculation with the preset matching threshold range of each entry in the hierarchical modulation strategy library. Select the single modulation strategy with the highest overlap value as the current sample adaptation strategy and generate one-to-one corresponding modulation type matching data. S34. Statistically match all modulation strategy entries in the modulation type matching data, merge and summarize the adjustment requirements of multiple samples with the same strategy, remove duplicate matching sample entries, and output a unified modulation matching summary dataset with no duplicate entries.

6. The control method for the sand mill cleaning system based on intelligent control according to claim 5, characterized in that, S33 includes: The structured modulation instruction data is traversed one by one to all ordered instruction entries, redundant operation fields are stripped out, and the three core failure feature threshold ranges embedded in each entry are extracted: failure level threshold, parameter adjustment amplitude threshold, and operation duration threshold, to generate single-entry feature threshold extraction data. Retrieve the inherent matching threshold ranges of all policy entries stored in the hierarchical modulation policy library, uniformly calibrate the threshold sampling accuracy and numerical magnitude, complete the global policy threshold standardization process, and generate a standardized policy threshold dataset in the library. A two-way interval coupling comparison operation is performed between the single-item feature threshold extraction data and the standardized in-library strategy threshold dataset to calculate the overlap ratio of each threshold interval and generate global threshold overlap calculation data. Filter the modulation strategies corresponding to the peak values ​​in the global threshold overlap calculation data, eliminate redundant alternative strategies caused by the overlap of thresholds of multiple strategies, lock the exclusive cleanup modulation strategies for a single sample, and generate initial sample strategy selection data. Each instruction entry is bound to the corresponding modulation strategy encoding information, completing the one-to-one coupling and archiving of instructions and strategies, organizing the full-domain coupled archive information, and generating one-to-one corresponding modulation type matching data.

7. The control method for the sand mill cleaning system based on intelligent control according to claim 1, characterized in that, The S4 includes: S41. Retrieve the factory-calibrated hardware operating limit parameters of the sand mill equipment. The hardware operating limit parameters include the upper and lower fluctuation limit range of the spindle speed, the maximum and minimum adjustable range of the continuous output flow of the feed pump, and the safe threshold range of the differential pressure that allows long-term stable operation at the separation interface. S42. Read the adjustment instruction field corresponding to each strategy in the unified modulation matching summary dataset, perform interval intersection operation between the parameter to be adjusted in the field and the hardware operation limit parameter, remove the over-limit parameter range that exceeds the device hardware carrying capacity, retain the effective adjustment parameter range that the device can operate safely, and generate independent parameter range limit data for a single modulation strategy. S43. Compare the failure hazard level values ​​corresponding to different modulation strategies. Bead bed compaction corresponds to low hazard level, rigid wedging corresponds to medium hazard level, and viscoelastic paste corresponds to high hazard level. The higher the hazard level value, the earlier the strategy is executed. The synchronous superposition parameter adjustment time index is used for secondary correction, and the short-term effective strategy is appropriately moved forward. S44. Perform linear sorting on all single modulation strategy independent parameter range-limited data according to the corrected order, bind each data to a unique order number, and generate modulation strategy call priority sequence data with ordered numbers.

8. The control method for the sand mill cleaning system based on intelligent control according to claim 7, characterized in that, S42 includes: Read each modulation strategy entry in the unified modulation matching summary dataset line by line, parse the three types of original adjustment instruction fields embedded in the entry: spindle control parameters, feed flow control parameters, and medium push pressure control parameters, extract the original parameter floating range of each field, and generate the original control parameter range data of the strategy. By benchmarking against the full-domain hardware operation limit parameters retrieved by S41, we split three types of hardware parameters into dedicated safe and controllable ranges, unify the numerical accuracy and sampling measurement standards of the original control parameter range and the hardware limit range, and generate standardized dual-group range comparison data. Perform bidirectional interval intersection calculations on the standardized two-group interval comparison data parameter by parameter, divide the interval into two categories: parameter compliant overlapping intervals and parameter exceeding the standard extension intervals, and generate full-domain parameter interval division result data; Remove all out-of-range extension intervals in the result data of the global parameter interval division, retain the compliant overlapping intervals that are compatible with the hardware carrying capacity of the device, simultaneously mark the failure condition type of the interval adaptation, and generate a single modulation strategy independent parameter range limit data.

9. The control method for the sand mill cleaning system based on intelligent control according to claim 1, characterized in that, The S5 includes: S51. Divide the internal parallel computing hardware resources of the sand mill controller. The hardware resources include three types of resource units: independent parameter adjustment computing channel, real-time sensor data feedback monitoring channel, and cleaning action execution output timing window. Calculate the total number of strategies in the modulation strategy call priority sequence data. S52. According to the priority allocation logic in the sequence from high to low, allocate a set of independent parameter adjustment operation channels and execution timing windows that do not occupy each other for each modulation strategy, record the resource unit number and timing start and end nodes occupied by each strategy, and generate complete modulation resource pre-allocation data. S53. Based on the start and end nodes of the timing window of each strategy in the pre-allocation data of modulation resources, set up multi-channel synchronous trigger signals to synchronously open the dedicated backtracking monitoring channel of the separation interface status. The monitoring channel continuously collects two types of verification sensor data: differential pressure of the separation interface and acoustic emission of medium impact throughout the cleaning operation. S54. Integrate multi-channel synchronous trigger signals into a unified device identification ready signal, synchronously collect the verification sensor data collected throughout the monitoring channel and store it in time-series segments to form a status backtracking sample, merge the ready signal and the time-series segmented backtracking sample, and generate a multi-modulation strategy parallel execution ready signal and effect verification backtracking data.

10. A sand mill cleaning system based on intelligent control, characterized in that, The system includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 9.

Images