Intelligent Control Method for Ball Mill Grinding Process Based on Machine Learning

By constructing an intelligent control method for the ball mill grinding process based on machine learning, the grinding load status is identified by using feeding, driving power and vibration data. This solves the problem of inaccurate grinding load identification in ball mills, achieves precise matching of feeding control, and improves the operational stability and efficiency of ball mills.

CN123102455APending Publication Date: 2026-11-17HENGYANG CHAOYANG HEAVY MASCH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611629933.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-21
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies cannot accurately identify the grinding load state and its changing trend during the ball mill grinding process, resulting in a mismatch between the feed control type and the actual grinding conditions, which affects the operating efficiency and stability of the ball mill.

Method used

By acquiring operational data such as feed mass flow rate, drive power, vibration acceleration, and discharge mass flow rate of the ball mill, a grinding response feature set is constructed. Then, a machine learning model is used to identify the grinding load state and load migration direction, and the feed control parameters are adjusted to match the actual grinding conditions.

Benefits of technology

It improves the accuracy of identifying grinding load and its direction of change, ensures that the feed adjustment matches the actual grinding conditions, reduces the problem of feed control mismatch caused by inaccurate judgment of load changes, and improves the operational stability and efficiency of the ball mill.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123102455A_ABST
    Figure CN123102455A_ABST
Patent Text Reader

Abstract

This invention relates to the field of mechanical grinding control technology, specifically to an intelligent control method for ball mill grinding processes based on machine learning. The method includes acquiring ball mill grinding operation data; constructing a grinding response feature set based on changes in feed, drive power, vibration, and discharge, as well as response timing; determining the grinding load state and load migration direction using a pre-set machine learning model, and accordingly determining feed control parameters and the expected grinding response; determining the grinding response drift based on the deviation between the actual and expected grinding response, and correcting subsequent grinding load states and load migration directions while updating the feed control parameters. This invention improves the reliability of identifying grinding load states and trends under complex feed conditions, ensures that feed adjustment matches actual grinding conditions, and reduces the impact of response drift caused by changes in the state of the grinding media and liners on long-term control accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of mechanical grinding control technology, and more specifically, to a machine learning-based intelligent control method for ball mill grinding processes. Background Technology

[0002] Ball mills are commonly used grinding equipment in metallurgical mining and mineral processing, typically for further grinding crushed ore materials. During continuous operation, operating parameters such as the feed rate need to be adjusted according to changes in the mill load to maintain a suitable grinding state.

[0003] In actual production, the particle size distribution, coarse particle ratio, and grindability of the upstream crushed material can change. Even if the feed mass flow rate remains relatively constant for a short period, the material retention and grinding load within the mill may still change. Due to the lag and nonlinear relationship between operating parameters such as drive power and changes in mill load, different mill load states may exhibit similar drive power. When adjusting the feed solely based on power or current, it is difficult to accurately determine the direction of current load change, which can easily lead to a mismatch between the feed control type and the actual mill conditions.

[0004] Furthermore, as the ball mill operates continuously, the state of the grinding media and the cylinder liner gradually changes, causing deviations in the drive power, ball mill vibration response, and discharge response under the same grinding load, which in turn affects the continuous judgment of the current grinding condition. Therefore, there is an urgent need for an intelligent control method for the ball mill grinding process based on machine learning, which can accurately identify the grinding load state and its changing trend in the continuous ball milling process, and keep the feed adjustment matching the actual grinding condition when the grinding response relationship changes. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent control method for the ball mill grinding process based on machine learning, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention aims to provide an intelligent control method for the ball mill grinding process based on machine learning, comprising the following steps: S1. Obtain grinding operation data of the ball mill in the current control cycle and historical control cycles, and construct ball mill operation sequence data; The grinding operation data includes at least the feed mass flow rate, drive power, ball mill vibration acceleration data, and discharge mass flow rate. S2. Determine the feed change, drive power change, ball mill vibration response change, and discharge change based on the ball mill running sequence data, and determine the response timing of drive power, ball mill vibration response, and discharge status relative to the feed change. Construct the grinding response feature set corresponding to the current control cycle based on each change and the corresponding response timing. S3. Input the grinding response feature set into a preset machine learning model to determine the current grinding load state of the ball mill and the corresponding load migration direction. S4. Determine the corresponding feed control parameters for the ball mill based on the grinding load state and load migration direction, and determine the expected grinding response for the next control cycle based on the feed control parameters. S5. Obtain the actual grinding response corresponding to the next control cycle, and determine the grinding response drift amount based on the deviation between the actual grinding response and the expected grinding response. S6. Based on the grinding response drift, the grinding load state and load migration direction determined by the preset machine learning model in the next control cycle are corrected and judged, and the corresponding grinding load state and load migration direction are determined based on the corrected judgment, thereby updating the corresponding feeding control parameters of the ball mill.

[0007] Furthermore, the grinding operation data is periodically statistically analyzed according to a preset control cycle, and a grinding response feature set is constructed based on the parameter changes in adjacent control cycles and the number of response lag cycles of each operating parameter relative to the effective feed change.

[0008] Furthermore, the preset machine learning model includes a grinding load state classification sub-model and a load migration direction classification sub-model, both of which adopt a random forest classification model based on historical ball mill operation samples.

[0009] Further, the feed control type and feed control change amount are determined based on the grinding load state and load migration direction, and the target feed mass flow rate corresponding to the next control cycle is determined accordingly; candidate historical control cycles are determined from historical control cycles that have the same grinding load state, load migration direction, and feed control type as the current control cycle; historical reference control cycles are determined based on the difference between the feed control change amount corresponding to each candidate historical control cycle and the feed control change amount corresponding to the current control cycle; and the expected grinding response corresponding to the next control cycle is determined based on the subsequent grinding response corresponding to the historical reference control cycle.

[0010] Furthermore, the grinding response drift is determined based on the deviation of the driving power response, the ball mill vibration response, and the discharge response between the actual grinding response and the expected grinding response. When the continuous grinding response drift meets the preset drift correction condition, the current response deviation relationship is determined based on the driving power response deviation, the ball mill vibration response deviation, and the discharge response deviation. A historical reference working condition matching the current response deviation relationship is determined from the historical control record of the target ball mill. Based on the grinding load state and load migration direction corresponding to the historical reference working condition, the grinding load state and load migration direction determined by the preset machine learning model are corrected and judged.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. In this intelligent control method for ball mill grinding process based on machine learning, by acquiring the feed mass flow rate, driving power, ball mill vibration acceleration and discharge mass flow rate, and combining the changes of each operating parameter and their response lag relationship relative to the changes in feed, a grinding response feature set is constructed. The grinding load state and load migration direction are determined by using a preset machine learning model. This method can characterize the current ball mill working condition from the dynamic response relationship of multiple parameters, improve the accuracy of identifying the grinding load and its change direction, and solve the problem that it is difficult to accurately determine the actual grinding load and its change direction based solely on power or current.

[0012] 2. In this intelligent control method for the ball mill grinding process based on machine learning, the feeding control type and feeding control change amount are determined together according to the grinding load state and the load migration direction, and the target feeding mass flow rate for the next control cycle is determined accordingly. This allows the feeding adjustment to take into account the current load range and its changing trend, improves the matching degree between the feeding control and the actual grinding conditions, and reduces the problem of mismatch between the feeding control type or adjustment range caused by inaccurate judgment of the load change direction.

[0013] 3. In this intelligent control method for the ball mill grinding process based on machine learning, the expected grinding response is determined by comparing the actual grinding response with the expected grinding response to obtain the grinding response drift. When a continuous response drift occurs, the grinding load state and load migration direction are corrected by combining the historical response deviation relationship. This enables the feeding control to adapt to the changes in the grinding response relationship during the long-term operation of the ball mill, reducing the impact of changes in the state of the grinding media and cylinder liner on the continuous judgment of the grinding condition and subsequent feeding adjustment. Attached Figure Description

[0014] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: This example provides a machine learning-based intelligent control method for the grinding process of a ball mill, which is applied to continuous ball milling in metallurgical mines or mineral processing production lines. The ball mill is used to receive ore materials processed by upstream crushing equipment and to further grind the materials by the relative movement between the grinding media and the materials to be ground inside the cylinder. During the production process, the ball mill is kept in a corresponding grinding operation state by adjusting the feed rate into the ball mill.

[0017] Specifically, in continuous production, the particle size distribution and grindability of the material after upstream crushing may change with the incoming material. Therefore, even if the mass of material entering the ball mill per unit time is basically similar, the drive power, ball mill vibration response, and discharge process may still exhibit different dynamic responses. Based on this, this embodiment does not use any single operating parameter as a direct measurement result of the actual load inside the mill. Instead, it constructs grinding response characteristics based on the change amplitude and response hysteresis relationship of multiple ball mill operating parameters before and after the change in feed, providing an input basis for the subsequent identification of grinding load status and load migration direction.

[0018] See Figure 1 As shown in the figure, the intelligent control method for the ball mill grinding process based on machine learning provided in this embodiment includes the following steps: S1. Obtain grinding operation data of the ball mill in the current control cycle and historical control cycles, and construct ball mill operation sequence data; The grinding operation data includes at least the feed mass flow rate, drive power, ball mill vibration acceleration data, and discharge mass flow rate.

[0019] It should be noted that the feed mass flow rate can be obtained from the feed-side electronic belt scale; the drive power can be obtained from the frequency converter, power measurement module, or drive control system corresponding to the main drive motor of the ball mill; the ball mill vibration acceleration data can be obtained from vibration acceleration sensors installed on the main bearing housing, end support housing, or other locations capable of stably transmitting mechanical vibration, preferably piezoelectric vibration acceleration sensors; the discharge mass flow rate can be obtained from the discharge-side electronic belt scale according to the discharge method, or determined according to the discharge slurry flow rate and corresponding slurry density. The vibration acceleration data is used to reflect the changes in mechanical vibration response during ball milling and is not used alone as a direct detection result of the grinding load inside the mill.

[0020] Furthermore, let the first... The duration of each control cycle is (Used for periodic division of continuously acquired ball mill operation data), for the feed mass flow rate, drive power, and discharge mass flow rate acquired within this control cycle, calculate the corresponding periodic average values ​​respectively: ; ; ; In the formula, Indicates the first The cycle average feed mass flow rate corresponding to each control cycle is used to characterize the average material supply level entering the ball mill within that control cycle, and the unit is t / h. Indicates the first The average drive power corresponding to each control cycle is used to characterize the average power response of the main drive system of the ball mill within that control cycle, and the unit is kW; Indicates the first The cycle average discharge mass flow rate corresponding to each control cycle is used to characterize the average mass flow rate of the material discharged from the ball mill within that control cycle, and the unit is t / h; Indicates the first Within the first control cycle, the first Each sampled value of the feed mass flow rate; Indicates the first Within the first control cycle, the first One drive power sample value; Indicates the first Within the first control cycle, the first Each discharge mass flow rate sample value; Indicates the first The number of effective feed mass flow sampling points that participate in the calculation of the cycle average feed mass flow within each control cycle; Indicates the first The number of effective drive power sampling points participating in the calculation of the cycle average drive power within each control cycle. Indicates the first The number of effective discharge mass flow sampling points used in the calculation of the cycle average discharge mass flow within each control cycle. Indicates the sequence number of the data sampling point within the corresponding control cycle; This indicates the control cycle number for the current data statistics process.

[0021] For ball mill vibration acceleration data, the root mean square value of the vibration acceleration within the control period is preferably used as the periodic vibration response value: ; In the formula, Indicates the first The periodic vibration response value corresponding to each control cycle is used to characterize the effective intensity of the mechanical vibration response of the ball mill within that control cycle, and the unit is m / s². Indicates the first Within the first control cycle, the first One effective vibration acceleration sample value is obtained from the ball mill vibration acceleration sensor; Indicates the first The number of effective vibration acceleration sampling points used in the calculation of periodic vibration response values ​​within each control cycle.

[0022] Based on this, the first Each control cycle generates a set of cycle operation data corresponding to the ball milling process: ; The current control cycle and its consecutive previous historical control cycles are corresponding to Arranged in chronological order according to the control cycles, this constitutes the ball mill's runtime sequence data: ; in, The number of historical control cycles that participated in the current grinding status analysis, which is a positive integer greater than or equal to 1; Indicates the first Each control cycle is the ball mill runtime sequence data constructed for the current control cycle; Indicates the first The cycle grinding operation data corresponding to each control cycle; This represents the cycle grinding operation data corresponding to the earliest historical control cycle; This represents the cycle grinding operation data corresponding to the previous historical control cycle. Indicates the current number The sequence number of the previous historical control cycle adjacent to each control cycle.

[0023] It should be noted that the number of historical control cycles This is used to limit the range of historical data participating in the current grinding response analysis. Its value is determined based on the target ball mill's response duration during normal continuous grinding, including the adjusted drive power, mechanical vibration, and discharge process. The time range consisting of the historical control cycle and the current control cycle can cover the historical operation process required for subsequent grinding response relationship analysis.

[0024] S2. Determine the changes in feed, drive power, ball mill vibration response, and discharge based on the ball mill operation sequence data, and determine the response timing of drive power, ball mill vibration response, and discharge status relative to the changes in feed. Construct the grinding response feature set corresponding to the current control cycle based on each change and the corresponding response timing.

[0025] It should be noted that in the continuous ball milling process of metallurgical mines, the particle size distribution, coarse particle ratio, and grindability of the ore after upstream crushing will change with each batch of incoming material. Simultaneously, the material retention in the mill will also dynamically change with the feeding and discharging processes. Therefore, even if the main drive power of the ball mill is at a similar level, the actual grinding load inside the mill may still be in different states, possibly exhibiting continued increase, stabilization, or gradual decrease. Furthermore, after a change in feed, the main drive power, ball mill mechanical vibration, and discharge status do not change synchronously but have different response lags. Therefore, relying solely on a single drive power or operating parameter is insufficient to accurately determine the current grinding load and its direction of change in the ball mill, easily leading to a mismatch between the subsequent feed control type and the actual grinding state. Based on this, this embodiment further constructs a grinding response feature set according to changes in feed, drive power, ball mill vibration response, and discharge, and their response timing relationships, and determines the current grinding load state and load migration direction accordingly.

[0026] In this embodiment, based on ball mill runtime sequence data First, determine the changes in the ball mill operating parameters between adjacent control cycles: Among them, for the current number The first control cycle is determined based on the average feed mass flow rate of the current control cycle and the previous control cycle. The feed change corresponding to each control cycle is: ; In the formula, Indicates the first The cycle average feed mass flow rate corresponding to each control cycle; Indicates the first The control cycle is relative to the first The change in feed rate for each control cycle; it should be noted that when When, it indicates that the feed mass flow rate in the current control cycle has increased compared to the previous control cycle; when When this occurs, it indicates that the feed mass flow rate in the current control cycle is lower than that in the previous control cycle.

[0027] Based on the average drive power of the current control cycle and the previous control cycle, the change in drive power is determined as follows: ; In the formula, Indicates the first The average drive power of each control cycle; Indicates the first The control cycle is relative to the first The change in drive power per control cycle.

[0028] Based on the periodic vibration response values ​​corresponding to the current control cycle and the previous control cycle, the change in the ball mill vibration response is determined as follows: ; In the formula, Indicates the first The control cycle is relative to the first The change in the vibration response of the ball mill during each control cycle; Indicates the first The periodic vibration response value corresponding to each control cycle.

[0029] Based on the average discharge mass flow rate of the current control cycle and the previous control cycle, the discharge variation is determined as follows: ; In the formula, Indicates the first The control cycle is relative to the first The change in discharge mass flow rate during each control cycle; Indicates the first The average discharge mass flow rate corresponding to each control cycle.

[0030] Furthermore, to distinguish between normal operating fluctuations during a stable continuous grinding process and dynamic responses caused by actual feed adjustments, this embodiment determines the upper limits of normal fluctuations for the absolute values ​​of feed changes, drive power changes, ball mill vibration response changes, and discharge mass flow rate changes from historical control cycles of the target ball mill during continuous feeding, continuous grinding, and continuous discharge, excluding start-up, shutdown, no-load, maintenance, significant material blockage, and significant measurement anomalies. The upper limits of normal fluctuations are determined using the maximum normal value of the corresponding absolute values ​​of the changes from the aforementioned historical normal operating data.

[0031] From the historical control cycles corresponding to the stable continuous grinding conditions, the changes in feed rate, drive power, ball mill vibration response, and discharge mass flow rate are obtained respectively, and their normal fluctuation upper limits are determined according to the normal fluctuation range of the corresponding changes; specifically, these include the normal fluctuation upper limit of the absolute value of the feed rate change of the target ball mill under stable continuous grinding conditions. The upper limit of normal fluctuation in the absolute value of the change in drive power of the target ball mill under stable continuous grinding conditions. The upper limit of normal fluctuation of the absolute value of the vibration response change of the target ball mill under stable continuous grinding conditions. And the upper limit of normal fluctuation of the absolute value of the change in discharge mass flow rate of the target ball mill under stable continuous grinding conditions. It should be noted that the stable continuous grinding condition refers to the ball mill being in a state of continuous feeding and continuous discharging, without significant feeding adjustments, and not in an abnormal continuous grinding stage such as startup, shutdown, no-load operation, maintenance, or significant material blockage.

[0032] In this embodiment, from the ball mill running sequence data The control period in which the feed change meets the preset feed change conditions is determined, and this period is taken as the target feed change period. .

[0033] Among them, the upper limit of normal fluctuation of feed under stable continuous grinding conditions. Determine the preset feed change threshold: ; in, This indicates the preset feed change threshold; This indicates the upper limit of normal fluctuation in the feed variation under stable continuous grinding conditions. This represents the margin coefficient for determining changes in feed rate.

[0034] It should be noted that in continuous ore ball milling operations, the ball mill is continuously fed via an electronic belt scale. During stable operation, the feed mass flow rate will still experience slight periodic variations due to the distribution of conveyor belt load, fluctuations in upstream material supply, and metering errors. To distinguish actual feed adjustments from the aforementioned normal feed fluctuations, the feed change judgment margin coefficient... The value range is 1.2 to 1.5, and in this embodiment, 1.3 is preferred, so that the preset feed change threshold is higher than the upper limit of normal feed fluctuation under stable continuous grinding state, while avoiding the omission of feed adjustments that are of control significance in the actual production process due to the threshold being set too high.

[0035] Furthermore, when the first Each control cycle satisfies At that time, the first The control cycle is defined as the target feed change cycle.

[0036] in, The cycle number indicating the target feed change cycle; Indicates the first The control cycle relative to its previous control cycle (the first control cycle) The change in feed rate (per cycle).

[0037] Therefore, in this embodiment, the feed change amount satisfying the preset feed change condition specifically means that the absolute value of the feed change amount reaches or exceeds the preset feed change threshold determined based on the normal feed fluctuation range under the stable continuous grinding condition of the target ball mill.

[0038] Furthermore, based on the target feed change cycle... As a baseline response period, in subsequent continuous control periods, it is determined whether changes in drive power, ball mill vibration response, and discharge mass flow rate meet their respective preset response conditions. It should be noted that the preset response conditions refer to the effective response judgment conditions set for different ball mill operating parameters, used to determine whether the changes in the corresponding operating parameters have exceeded the range of changes formed by normal fluctuations during stable grinding.

[0039] The effective response threshold of the driving power is determined based on the upper limit of normal fluctuation of the driving power under stable continuous grinding conditions: ; in, Indicates the effective response threshold of the drive power; This indicates the upper limit of normal fluctuation in the change of drive power under stable continuous grinding conditions; Indicates the decision margin coefficient for drive power response; It should be noted that, considering the normal fluctuations in the main drive power during the stable grinding phase in continuous ore ball milling, and the fact that the drive power response after changes in effective feed usually needs to be significantly higher than this normal fluctuation range, the drive power response judgment margin coefficient is... The value range is 1.2 to 1.6, and in this embodiment, 1.3 is preferred. When the value is too small, it is easy to identify the normal fluctuation of the motor load as an effective response; when the value is too large, the actual drive power response may be delayed in being identified.

[0040] For the first [period] after the target feed change cycle One control cycle, when the condition is met When the change in drive power during the control cycle satisfies the corresponding preset response condition, it is determined that the change in drive power during that control cycle meets the preset response condition; where, This indicates the control cycle number after the target feed change cycle, where a response judgment is to be made. Indicates the first The change in drive power corresponding to each control cycle.

[0041] The effective response threshold for ball mill vibration is determined based on the upper limit of normal vibration fluctuation under stable continuous grinding conditions. ; In the formula, This indicates the effective response threshold for ball mill vibration; This represents the upper limit of normal fluctuation in the vibration response variation of a ball mill under stable continuous grinding conditions. This represents the vibration response margin coefficient for the ball mill. It should be noted that, considering the significant natural vibration fluctuations caused by the mechanical collisions between the steel balls, ore, and the cylinder structure during continuous ball milling, this coefficient should be slightly higher than the power side. The value range is 1.3 to 1.8, and 1.5 is preferred in this embodiment. If the value is too small, it is easily affected by instantaneous collision vibration, and if the value is too large, it may weaken the sensitivity to changes in the actual grinding state.

[0042] when At that time, determine the first The vibration response of the ball mill in each control cycle meets the corresponding preset response conditions.

[0043] The effective response threshold for discharge mass flow rate is determined based on the upper limit of normal fluctuations in discharge under stable continuous grinding conditions:

[0044] In the formula, Indicates the effective response threshold for discharge mass flow rate; This represents the upper limit of normal fluctuation in the change of discharge mass flow rate under stable continuous grinding conditions. This represents the margin coefficient for the material discharge response; it should be noted that this takes into account the certain pulsation and metering fluctuations that occur on the discharge side during continuous discharge. The value range is 1.2 to 1.6, and in this embodiment, 1.3 is preferred. If the value is too small, normal discharge pulsation is easily identified as a valid response, and if the value is too large, the real response on the discharge side may be missed.

[0045] when At that time, determine the first The change in discharge mass flow rate in each control cycle meets the corresponding preset response conditions.

[0046] Therefore, the satisfaction of the corresponding preset response condition described in this embodiment specifically refers to the absolute value of the change in the corresponding operating parameter reaching or exceeding the effective response threshold determined based on the normal fluctuation range of the stable continuous grinding condition of the operating parameter after the target feed change occurs.

[0047] Furthermore, during the target feed change cycle In the subsequent continuous control cycle, the target response cycle for the first time that the changes in drive power, ball mill vibration response, and discharge mass flow rate meet the corresponding preset response conditions is determined.

[0048] Specifically, when the first When the drive power change in a control cycle first meets the preset drive power response condition, the number of drive power response lag cycles is determined: ; In the formula, This indicates the target response period number in which the change in drive power first meets the corresponding preset response condition. Indicates the target feed change cycle number; This indicates the number of control cycles that drive power response lag cycles take to form a recognizable response after an effective change in feed.

[0049] When the When the ball mill vibration response change in a control cycle first meets the corresponding preset response condition: ; In the formula, This indicates the target response period number in which the change in the ball mill's vibration response first meets the corresponding preset response condition. This indicates the number of hysteresis cycles in the ball mill's vibration response, used to characterize the number of control cycles required for the ball mill's mechanical vibration to form a identifiable response after an effective change in the feed.

[0050] When the When the change in the discharge mass flow rate of a control cycle first meets the corresponding preset response condition: ; in, This indicates the target response period number in which the change in discharge mass flow rate first meets the corresponding preset response condition. This indicates the number of discharge response lag cycles, used to characterize the number of control cycles required for the discharge mass flow rate to form an identifiable response after an effective change in the feed.

[0051] It should be noted that, , and These terms are used to describe the differences in response delay exhibited by the ball mill drive power, mechanical vibration response, and discharge process after a single effective change in feed.

[0052] Furthermore, when there is no control cycle in the current ball mill running sequence data that meets the preset feed change conditions, the response lag cycle number determined by the most recent effective feed change during the previous normal continuous grinding process of the target ball mill is adopted; when a certain operating parameter does not meet the corresponding preset response conditions within the preset response retrieval range, the response lag cycle number corresponding to the operating parameter is not updated using this feed change, and the corresponding response lag cycle number determined during the previous effective response process continues to be adopted.

[0053] In this embodiment, the response timing of the driving power, ball mill vibration response, and discharge state relative to changes in feed is specifically determined by the number of driving power response lag cycles. Number of hysteresis cycles in the vibration response of the ball mill and the number of material discharge response lag cycles The characteristics are as follows: the number of response lag cycles reflects the number of control cycles that the corresponding operating parameters experience from the target feed change cycle to the first effective response.

[0054] Furthermore, based on the changes in feed rate, drive power, ball mill vibration response, and discharge mass flow rate corresponding to the current control cycle, as well as the number of lag cycles for drive power response, ball mill vibration response, and discharge response, the first... Grinding response feature set corresponding to each control cycle: ; In the formula, Indicates the first The grinding response feature set corresponding to each control cycle.

[0055] It should be noted that, in this embodiment, the changes in feed rate, drive power, ball mill vibration response, and discharge mass flow rate are used to describe the actual operational changes in the current ball milling process relative to the previous control cycle; the number of drive power response lag cycles, ball mill vibration response lag cycles, and discharge response lag cycles are used to describe the dynamic response delays of different operating parameters of the ball mill after the actual feed rate has changed effectively.

[0056] S3. Input the grinding response feature set into a preset machine learning model to determine the current grinding load state of the ball mill and the corresponding load migration direction.

[0057] Specifically, the preset machine learning model is used to identify the grinding load state and its changing trend in the current ball milling process based on the changes in feed, drive power response, mechanical vibration response, discharge response, and the lag relationship of each response relative to the changes in feed during the continuous grinding process of the ball mill.

[0058] In this embodiment, the preset machine learning model includes a grinding load state classification sub-model and a load migration direction classification sub-model. Both the grinding load state classification sub-model and the load migration direction classification sub-model adopt a random forest classification model pre-established based on historical ball mill operation samples.

[0059] Furthermore, the random forest classification model is an existing classification model that classifies input features using multiple decision trees and determines the final classification result based on the classification results of the multiple decision trees. This embodiment does not change the basic classification principle of the random forest model itself, but uses the grinding response features formed during the continuous grinding process of the ball mill as the model input to determine the grinding load state and load migration direction corresponding to the ball milling conditions.

[0060] It should be noted that the grinding load state classification sub-model and the load migration direction classification sub-model are both pre-established based on historical ball mill operation data generated during the normal continuous production process of the target ball mill.

[0061] For the Each historical control cycle yields the corresponding historical grinding response feature set: ; in, Indicates the first Historical grinding response feature set corresponding to each historical control cycle; Indicates the first The feed change corresponding to each historical control cycle; Indicates the first The change in drive power corresponding to each historical control cycle; Indicates the first The change in ball mill vibration response corresponding to each historical control cycle; Indicates the first Changes in discharge mass flow rate corresponding to each historical control cycle; This indicates the number of drive power response lag cycles for the corresponding historical operating phase; This indicates the number of hysteresis cycles in the vibration response of the ball mill during the corresponding historical operating phase. This indicates the number of material discharge response lag cycles corresponding to the historical operation phase.

[0062] The historical ball mill operation samples were selected from production operation records formed when the ball mill was in a continuous feeding, continuous grinding and continuous discharging state, and data corresponding to start-up, shutdown, no-load, maintenance, obvious material blockage and obvious measurement abnormalities were excluded.

[0063] This ensures that the historical samples used in the model establishment phase correspond to the actual operating conditions of the ball mill during the subsequent normal continuous grinding phase.

[0064] At the same time, for each historical grinding response feature set, the corresponding historical grinding load state label is determined.

[0065] In this embodiment, the grinding load state is divided into a low load state. Suitable load conditions and high load conditions .

[0066] It should be noted that the historical grinding load status label is determined based on the feed adjustment records, continuous grinding operation status, and operation recovery results that have been formed during the actual production process of the target ball mill.

[0067] Specifically, for ball mills that can maintain continuous feeding, continuous grinding, and normal discharge, and do not require significant adjustments to increase or decrease material in multiple consecutive control cycles, the corresponding historical control cycles are determined as suitable load samples.

[0068] For historical operating phases in which the ball mill re-enters the appropriate load operating state after the ball mill is re-increased due to low material load in the mill during the production process, the historical control period before the implementation of the increased feeding control is defined as the low load sample.

[0069] For historical operating phases in which the ball mill re-enters the appropriate load operating state after the reduction or restriction of feed during production due to high material load inside the mill, the historical control period before the implementation of the reduction or restriction of feed control is defined as the high load sample.

[0070] Based on this, the historical grinding load status label is confirmed by the actual production control process of the target ball mill and the adjusted operating results, rather than being directly classified based on any single parameter among drive power, vibration response, or discharge mass flow rate.

[0071] In this embodiment, the load migration direction is used to indicate the direction of time change of the grinding load of the ball mill as the control cycle progresses, and does not indicate the direction of spatial movement of the material inside the ball mill.

[0072] In this embodiment, the load migration direction is divided into: the load increase direction. Stable load status and the direction of load reduction .

[0073] Meanwhile, to ensure that the migration direction label has a clear basis for judgment, the low load state, suitable load state, and high load state are assigned corresponding load levels in order of increasing grinding load: ; In the formula, This indicates the load level corresponding to a low load condition. This indicates the load level corresponding to a suitable load condition; This indicates the load level corresponding to a high load condition.

[0074] For two consecutive historical control cycles where grinding load status confirmation has been completed, the load migration direction label for the corresponding historical sample is determined based on the relationship between the load level corresponding to the later control cycle and the load level corresponding to the earlier control cycle: When the load level of the subsequent control cycle is higher than that of the previous control cycle, the load migration direction of the corresponding historical sample is marked as the load increase direction. ; When the load level corresponding to the next control cycle remains consistent with that of the previous control cycle, the corresponding historical sample is marked as a load steady state. ; When the load level of the subsequent control cycle is lower than that of the previous control cycle, the corresponding historical sample is marked as the direction of load reduction. .

[0075] Furthermore, in this embodiment, each historical grinding response feature set is... The grinding load state training sample set is composed of the grinding load state and its corresponding grinding load state label, and the grinding load state classification sub-model is established based on this.

[0076] Each historical grinding response feature set The load migration direction training sample set is composed of the load migration direction labels and their corresponding labels, and the load migration direction classification sub-model is established accordingly.

[0077] In this embodiment, each random forest classification sub-model includes multiple decision trees. During the model building process, random sampling with replacement is used to form training sub-samples corresponding to each decision tree from the corresponding training sample set. During the node partitioning process of each decision tree, some grinding response features are randomly selected to participate in the node partitioning. Each decision tree outputs the corresponding classification result, and the final classification result of the random forest classification sub-model is determined by majority voting.

[0078] Furthermore, under the continuous ball mill operation conditions described in this embodiment, the number of decision trees contained in each random forest classification sub-model can be 50 to 200, preferably 100.

[0079] It should be noted that when the number of decision trees is too small, the classification stability of the combination of different feed changes, drive power response, ball mill vibration response and discharge response may decrease; when the number of decision trees continues to increase, the improvement in the stability of the classification results tends to decrease, and at the same time, the amount of calculation when identifying the ball mill load state in each control cycle increases; therefore, this embodiment preferably uses 100 decision trees.

[0080] Furthermore, once the two random forest classification sub-models are established, they are saved as preset machine learning models for online state recognition during the actual continuous grinding process of the target ball mill, eliminating the need to retrain the models in each control cycle.

[0081] Specifically, for the current number Each control cycle will determine the current grinding response feature set. The data is input into the grinding load state classification sub-model; the grinding load state classification sub-model is represented as follows: The grinding load state corresponding to the current control cycle is: ; In the formula, This represents a sub-model for classifying grinding load conditions. Indicates the first The grinding response feature set corresponding to each control cycle; Indicates the first The grinding load state corresponding to each control cycle.

[0082] Furthermore, the grinding load state satisfies: ; The current grinding condition of the ball mill is determined to be one of the following: low load, suitable load, or high load.

[0083] It should be noted that the grinding load state is determined by a combination of changes in current feed, drive power, ball mill vibration response, discharge, and the response lag of corresponding operating parameters relative to changes in feed, rather than by any single operating parameter.

[0084] Furthermore, the grinding response feature set corresponding to the same current control cycle is... Input the load migration direction classification sub-model.

[0085] The load migration direction classification sub-model is represented as follows: The load migration direction corresponding to the current control cycle is: ; in, This represents a sub-model for classifying load migration direction. Indicates the first The load migration direction corresponding to each control cycle.

[0086] The load migration direction satisfies: ; These respectively indicate that the current grinding load is in one of the following directions: increasing load, stable load, or decreasing load.

[0087] It should be noted that if two control cycles have similar drive power, but their feed changes, ball mill vibration response, discharge response, and corresponding response lag relationships are different, then the load migration directions corresponding to the two control cycles may be different; therefore, this embodiment does not directly equate the drive power change direction with the grinding load migration direction.

[0088] Finally obtained ,in, This indicates the grinding load status corresponding to the current control cycle; This indicates the load migration direction corresponding to the current control cycle.

[0089] The grinding load state This is used to characterize the current ball milling process within a low-load, suitable-load, or high-load range; the load migration direction. It is used to characterize whether the current grinding load is changing in the direction of increasing, stabilizing, or decreasing.

[0090] It should be noted that... As input for the subsequent step S4 to determine the ball mill feed control type and feed control change amount, the ball mill feed control not only considers the current state of the grinding load, but also the direction of further changes in the current load.

[0091] Among them, the feed control type refers to the direction of change of the feed mass flow rate, which means adjusting the feed rate in the increasing direction (i.e., increasing feed), adjusting it in the decreasing direction (i.e., decreasing feed), or keeping it unchanged (i.e., maintaining feed). Specifically: Increasing the feed rate indicates that the target feed mass flow rate for the next control cycle is higher than the average feed mass flow rate for the current cycle. Reducing the feed rate indicates that the target feed mass flow rate for the next control cycle is lower than the average feed mass flow rate for the current cycle. "Maintain feed" means that the target feed mass flow rate for the next control cycle will remain at the current level or within the allowable small holding range.

[0092] S4. Determine the corresponding feed control parameters for the ball mill based on the grinding load state and load migration direction, and determine the expected grinding response for the next control cycle based on the feed control parameters. In continuous ball milling, feed control under the same grinding load is also affected by the direction of load migration. For example, if the load continues to increase under high load conditions, a larger feed reduction should be used, while if the load has already migrated in the direction of decreasing, a smaller feed reduction should be used to avoid over-adjustment. Therefore, in this embodiment, the feed control type and feed control variation are determined jointly based on the grinding load state and the direction of load migration.

[0093] Therefore, in this embodiment, the feeding control type and feeding control change amount corresponding to the current control cycle are jointly determined based on the current grinding load state of the ball mill and the corresponding load migration direction.

[0094] Specifically, let the first The feeding control type corresponding to each control cycle is: The feeding control types include increasing feeding, decreasing feeding, and maintaining feeding.

[0095] In one embodiment of the present invention, the corresponding feed control type and feed control amplitude level are determined based on the current grinding load state and the load migration direction. Specifically, when the ball mill is under low load and the load continues to decrease, the feed is increased and a larger feed control amplitude is used; when the mill is under low load and the load remains stable, the feed is increased and a conventional feed control amplitude is used; when the mill is under low load but the load has migrated in the direction of increase, the feed is increased and a smaller feed control amplitude is used.

[0096] When the ball mill is under suitable load and the load remains stable, maintain the current feed rate; when it is under suitable load and the load shifts towards increasing, slightly reduce the feed rate; when it is under suitable load and the load shifts towards decreasing, slightly increase the feed rate.

[0097] When the ball mill is under high load and the load continues to increase, reduce the feed rate and use a larger feed control amplitude; when it is under high load and the load remains stable, reduce the feed rate and use a normal feed control amplitude; when it is under high load but the load has shifted towards decreasing, reduce the feed rate and use a smaller feed control amplitude.

[0098] Therefore, the feed control not only reflects the current grinding load range of the ball mill, but also the subsequent trend of the grinding load.

[0099] Furthermore, let the first... The feed control variation determined for each control cycle based on the current grinding load state and load migration direction is: .

[0100] It should be noted that the change in feed control is... Different from the feed change in step S2 The two have different meanings; among them, the change in feed rate... Indicates the first The actual change in feed mass flow rate that has occurred in each control cycle relative to the previous control cycle; the feed control change is... Indicates the first The amount of change in feed control for the next control cycle is determined based on the current grinding state within each control cycle.

[0101] In this embodiment, the feed control change is determined based on the cycle average feed mass flow rate corresponding to the current control cycle and the corresponding feed control amplitude level: ; In the formula, Indicates the first The change in feed control for each control cycle is expressed in t / h. Indicates the first The average feed mass flow rate for each control cycle is expressed in t / h. This represents the feed control ratio coefficient determined based on the current grinding load state and load migration direction; it is a dimensionless parameter.

[0102] It should be noted that in continuous ore ball milling, continuous feeding by electronic belt scales, and ball mill response, there is a certain dynamic lag; the aforementioned feed control ratio coefficient It can be set according to different feed control amplitude levels. Among them, the smaller feed control amplitude corresponds to The value can be 0.01 to 0.02, with 0.015 being preferred; the corresponding value for conventional feed control is... A value of 0.02 to 0.04 is acceptable, with 0.03 being preferred; a larger feed control range corresponds to... The value can be 0.04 to 0.06, with 0.05 being preferred.

[0103] Specifically, when the feeding control type is held feeding, The corresponding change in feed control is zero.

[0104] Furthermore, if the feed control ratio is set too small, it may result in insufficient feed correction, causing the ball mill to remain in a state deviating from the appropriate grinding condition for a long time; if it is set too large, it may cause excessive adjustment of feed in a single operation due to the dynamic response lag in the ball milling process. Therefore, this embodiment adopts different feed control amplitudes according to the grinding load state and the load migration direction.

[0105] In this embodiment, the target feed mass flow rate corresponding to the next control cycle is determined based on the feed control type and the feed control change amount.

[0106] Specifically, when the feed control type is set to increase feed: ; When the feed control type is reduced feed: ; When the feed control type is hold feed: ; in, Indicates the first The target feed mass flow rate corresponding to each control cycle is expressed in t / h. Indicates the first The average feed mass flow rate for each control cycle is expressed in t / h. Indicates the first The change in feed control determined for each control cycle.

[0107] The target feed mass flow rate As the corresponding feed control parameter for the ball mill, it is used for feed control in the next control cycle.

[0108] Furthermore, the feed mass flow rate is calculated according to the aforementioned target. After the feeding control is executed, the ball mill's drive power, mechanical vibration, and discharge process will respond accordingly to changes in the grinding state inside the mill.

[0109] Since the ball mill itself has a dynamic process of feeding into the mill, grinding the material and discharging the material, this embodiment further determines the grinding response of the next control cycle that the current feeding control action usually corresponds to under the same or similar grinding conditions based on the historical control records formed during the previous normal continuous grinding process of the target ball mill, and uses it as a comparison benchmark for subsequent judgment on whether the ball mill response relationship has drifted.

[0110] Specifically, regarding the target ball mill's previously completed... Each historical control cycle can generate a corresponding historical control record: ; In the formula, Indicates the first Historical control records corresponding to each historical control cycle; Indicates the first The grinding load status corresponding to each historical control cycle; Indicates the first The load migration direction corresponding to each historical control cycle; Indicates the first The feeding control type corresponding to each historical control cycle; This indicates the first time after executing the feeding control. The change in drive power corresponding to each control cycle, in kW; This indicates the first time after executing the feeding control. The change in ball mill vibration response for each control cycle is expressed in m / s². This indicates the first time after executing the feeding control. The change in discharge mass flow rate corresponding to each control cycle is expressed in t / h.

[0111] Furthermore, regarding the current first For each control cycle, candidate historical control cycles with the same grinding load state, load migration direction, and feed control type as the current control cycle are first selected from the historical control records, i.e., those satisfying the following conditions: ; ; ; in, Indicates the cycle number of the current control cycle; This represents the candidate historical control cycle number selected from the historical control cycles preceding the current control cycle, and satisfies the following conditions: .

[0112] Furthermore, the historical reference control period is determined based on the difference between the feed control change corresponding to each candidate historical control period and the feed control change corresponding to the current control period.

[0113] Specifically, for the first The differences in feed control variation across candidate historical control cycles are as follows: ; in, Indicates the first The difference in feed control between each candidate historical control cycle and the current control cycle, in t / h; This indicates the change in feed control corresponding to the candidate historical control cycle; This indicates the change in feed control corresponding to the current control cycle.

[0114] It should be noted that the historical reference control cycle is selected from the candidate historical control cycles in order of increasing difference in the feed control variation.

[0115] In this embodiment, the number of historical reference control cycles can be 3 to 10, preferably 5. If the number of historical reference control cycles is too small, fluctuations in a single historical operation may have a significant impact on the expected grinding response; if the number is too large, it may introduce historical operating conditions that differ significantly from the current feed control action. Therefore, this embodiment preferably selects the 5 historical control cycles whose feed control changes are closest to the current control cycle.

[0116] When the number of candidate historical control periods is less than the preset number of historical reference control periods, all candidate historical control periods that meet the aforementioned screening conditions will be used as historical reference control periods.

[0117] Let the determined set of historical reference control periods be . The corresponding number of historical reference control periods is Based on the actual grinding response of the next control cycle after executing the corresponding feeding control in each historical reference control cycle, determine the expected grinding response corresponding to the current feeding control action.

[0118] Specifically, the expected change in drive power for the next control cycle is: ; The expected change in ball mill vibration response for the next control cycle is: ; The expected change in discharge mass flow rate for the next control cycle is: ; In the formula, This represents the set of historical reference control periods; This represents the number of historical reference control periods and is a dimensionless positive integer. Indicates the first The expected change in drive power for each control cycle, in kW; Indicates the first The expected change in ball mill vibration response for each control cycle is expressed in m / s². Indicates the first The expected change in discharge mass flow rate for each control cycle is expressed in t / h.

[0119] Therefore, the expected grinding response for the next control cycle is: ; in, This indicates the first step determined based on the current grinding load state, load migration direction, and current feed control action. The expected grinding response corresponding to each control cycle.

[0120] It should be noted that the feed control change is a ball mill feed control input, while the drive power change, ball mill vibration response change, and discharge mass flow rate change are ball mill operation responses formed after the feed control is executed. Therefore, in this embodiment, the feed control change is not used as a component parameter of the expected grinding response.

[0121] S5. Obtain the actual grinding response corresponding to the next control cycle, and determine the grinding response drift amount based on the deviation between the actual grinding response and the expected grinding response.

[0122] During long-term continuous grinding in a ball mill, the grinding media continuously participate in the impact and grinding of materials, and the liners bear the action of the grinding media and materials for an extended period. As the continuous operating time increases, the working states of the grinding media and liners may change, causing the interaction between the materials, grinding media, and the mill body to gradually change. Under these circumstances, even if the ball mill is under the same or similar grinding load and load migration direction, and performs the same or similar feeding control actions, the actual resulting drive power response, ball mill vibration response, and discharge response may gradually deviate from the historical response relationships corresponding to the previous stable continuous grinding stage.

[0123] Therefore, in this embodiment, entering the first... After the first control cycle, the data acquisition and cycle statistics are performed in the same way as in step S1 to obtain the first... The average drive power corresponding to each control cycle Periodic vibration response value and the average discharge mass flow rate per cycle .

[0124] Specifically, according to the method for determining the changes in operating parameters between adjacent control cycles in step S2, the first... The control cycle is relative to the first Actual drive power change over each control cycle , No. The control cycle is relative to the first The actual change in ball mill vibration response over one control cycle and the The control cycle is relative to the first Change in actual discharge mass flow rate during each control cycle Furthermore, the first The actual grinding response corresponding to each control cycle is: .

[0125] Specifically, according to step S4, the first... Expected grinding response for each control cycle The response deviations between the actual grinding response and the expected grinding response were determined separately: ; ; ; In the formula, Indicates the first The drive power response deviation corresponding to each control cycle is expressed in kW. Indicates the first The vibration response deviation of the ball mill corresponding to each control cycle is expressed in m / s². Indicates the first The material discharge response deviation corresponding to each control cycle is expressed in t / h.

[0126] It should be noted that the positive and negative relationships of each response deviation are used to indicate the direction of deviation of the corresponding actual response from the expected response; while the absolute value of each response deviation is used to indicate the degree of deviation of the corresponding actual response from the expected response.

[0127] Then it is further determined that: ; ; ; in, Indicates the first Absolute deviation of drive power response for each control cycle; Indicates the first Absolute deviation of ball mill vibration response for each control cycle; Indicates the first The absolute deviation of the material discharge response corresponding to each control cycle.

[0128] It should be noted that since the absolute deviation of the drive power response, the absolute deviation of the ball mill vibration response, and the absolute deviation of the discharge response have different physical dimensions, this embodiment determines the corresponding response absolute deviation normalization benchmark based on the historical control records formed during the stable continuous grinding stage of the target ball mill.

[0129] Select the historical normal control cycle in the stable continuous grinding stage from the historical control records of the target ball mill; the stable continuous grinding stage refers to the ball mill being in a state of continuous feeding, continuous grinding and continuous discharge, excluding historical control cycles corresponding to start-up, shutdown, no-load, maintenance, obvious material blockage and obvious measurement abnormalities.

[0130] For any historical normal control cycle to be evaluated, in the same way as determining the expected grinding response in step S4, based on the corresponding grinding load state, load migration direction, feed control type and feed control change, determine the corresponding historical reference control cycle from other historical control cycles, and further determine the historical expected grinding response corresponding to the historical normal control cycle to be evaluated.

[0131] When determining the expected historical grinding response corresponding to any historical normal control period to be evaluated, the historical normal control period to be evaluated itself is not used as the corresponding historical reference control period.

[0132] Based on the actual grinding response and the corresponding expected grinding response of each historical normal control cycle, the absolute deviation sequence of historical drive power response, the absolute deviation sequence of historical ball mill vibration response, and the absolute deviation sequence of historical discharge response are respectively formed.

[0133] After excluding data corresponding to startup, shutdown, no-load, maintenance, significant material blockage, and significant measurement anomalies, the maximum value in the corresponding historical response absolute deviation sequence is used as the upper limit of normal fluctuation of the corresponding response deviation, and the following are determined respectively: Drive power response absolute deviation normalized benchmark Normalized benchmark for absolute deviation of ball mill vibration response and the normalized benchmark for absolute deviation of material discharge response .

[0134] in, Indicates the absolute deviation used for drive power response The normalized benchmark for the absolute deviation of the drive power response after dimensionless processing is determined by the historical absolute deviation of the drive power response corresponding to the stable continuous grinding stage of the target ball mill, and the unit is kW.

[0135] Indicates the absolute deviation used for the vibration response of the ball mill. The normalized benchmark for the absolute deviation of the ball mill vibration response, which has been dimensionlessly processed, is determined by the historical absolute deviation of the ball mill vibration response, and its unit is m / s².

[0136] Indicates the absolute deviation used for discharging response. The normalized benchmark for the absolute deviation of the discharge response after dimensionless processing is determined by the historical absolute deviation of the discharge response, and the unit is t / h.

[0137] Among them, the normalized reference for absolute deviation of drive power response, the normalized reference for absolute deviation of ball mill vibration response, and the normalized reference for absolute deviation of discharge response are all parameters greater than zero; when the maximum normal value in the corresponding historical response absolute deviation sequence is zero, the preset minimum positive normalized reference is used instead to avoid invalid division in subsequent standardization calculations.

[0138] In one implementation, the maximum normal value in the corresponding historical response absolute deviation sequence is used as the corresponding normalization benchmark.

[0139] It should be noted that the normalized benchmark for absolute response deviation is used to measure the deviation between the actual response and the expected response, which is different from the response judgment scale used in step S2 to determine whether the operating parameters themselves form an effective response.

[0140] Determine the normalized benchmark for the absolute deviation of the drive power response. Normalized benchmark for absolute deviation of ball mill vibration response and the normalized benchmark for absolute deviation of material discharge response Then, the absolute deviations of each response corresponding to the current control cycle are dimensionless: ; ; ; in, , and They represent the first The standardized response deviation of the drive power corresponding to each control cycle is a dimensionless parameter. They represent the first The standardized response deviation of the ball mill vibration corresponding to each control cycle is a dimensionless parameter. Indicates the first The standardized response deviation of material discharge corresponding to each control cycle is a dimensionless parameter.

[0141] Each standardized response deviation is used to characterize the relative magnitude of the current corresponding absolute response deviation with respect to the scale of the absolute deviation of the same type of normal response during the historical stable continuous grinding stage of the target ball mill, thereby eliminating the dimensional differences between different response channels.

[0142] It should be noted that each standardized response deviation is only used to characterize the degree of deviation of the corresponding response channel; the positive or negative direction of each actual response relative to the expected response is still determined by the aforementioned driving power response deviation. Vibration response deviation of ball mill and material discharge response deviation reserve.

[0143] Furthermore, determine the first The grinding response drift corresponding to each control cycle: ; in, Indicates the first The grinding response drift corresponding to each control cycle is a dimensionless parameter.

[0144] The grinding response drift is used to comprehensively characterize the overall deviation of the actual drive power response, ball mill vibration response, and discharge response from the normal control conditions and corresponding grinding responses of the target ball mill during the historical stable continuous grinding stage, under the current grinding load state, load migration direction, and feed control action conditions.

[0145] S6. Based on the grinding response drift, the grinding load state and load migration direction determined by the preset machine learning model in the next control cycle are corrected and judged, and the corresponding grinding load state and load migration direction are determined based on the corrected judgment, thereby updating the corresponding feeding control parameters of the ball mill.

[0146] It should be noted that during the long-term continuous operation of the ball mill, the grinding response drift corresponding to a single control cycle may be affected by short-term fluctuations in the incoming material, local impacts, or measurement disturbances. Therefore, this embodiment does not directly perform state correction based on the grinding response drift of a single control cycle, but instead determines whether a persistent deviation in the response relationship has formed based on the grinding response drift of the most recent multiple consecutive control cycles.

[0147] Specifically, in this embodiment, entering the first... After one control cycle, the grinding response feature set corresponding to that control cycle is... Input the preset machine learning model to obtain the first... The initial grinding load state determined by the grinding load state classification sub-model for each control cycle and the The initial load migration direction determined by the load migration direction classification sub-model for each control cycle ; where superscript This indicates that the corresponding result has not yet been corrected for grinding response drift.

[0148] Furthermore, according to including the first The most recent control cycle, including the last one The degree of continuous grinding response drift is determined by the amount of grinding response drift corresponding to each continuous control cycle. ; in, Indicates as of the date The degree of continuous grinding response drift corresponding to each control cycle is a dimensionless parameter. Indicates the preset drift detection cycle number; This represents the period offset index, with a value of .

[0149] It should be noted that, considering the short-term operating disturbances and response duration during continuous ball milling, the preset drift judgment cycle number is determined. The number of control cycles can be 3 to 6, and in this embodiment, 4 control cycles are preferred; when When the size is too small, its ability to suppress single-cycle anomalous disturbances is limited; If the value is too large, it may increase the lag in judging changes in the continuous response relationship.

[0150] Furthermore, the degree of drift in the continuous grinding response is compared with a preset drift correction threshold. Comparison; where a preset drift correction threshold is used. It is a dimensionless parameter, determined based on the degree of drift in the historical continuous grinding response formed during the stable continuous grinding phase of the target ball mill, in order to distinguish between normal operation fluctuations and continuous response deviations that require correction.

[0151] It should be noted that, since the standardized response deviations in step S5 are all based on the normalized benchmark of the absolute deviation scale of the normal response corresponding to the historical stable continuous grinding stage of the target ball mill, the preset drift correction threshold is used in the continuous ore ball milling condition of this embodiment. The value range is set to 1.10 to 1.30. When the threshold is too small, normal incoming materials and operational fluctuations can easily trigger frequent corrections. When the threshold is too large, it may delay the identification of continuous changes in response relationship. Therefore, in this embodiment, 1.20 is preferred.

[0152] Specifically, this embodiment will The preset drift correction conditions have been determined.

[0153] When the preset drift correction condition is not met, the initial grinding load state is maintained. and initial load migration direction The current response deviation remains unchanged; when the preset drift correction condition is met, the current response deviation relationship is further determined.

[0154] Based on the drive power response deviation determined in step S5 Vibration response deviation of ball mill and material discharge response deviation Based on the corresponding normalized benchmark of absolute deviation of response, the following are determined respectively: ; ; ; In the formula, , as well as These represent the signed normalized response deviation of the drive power, the signed normalized response deviation of the ball mill vibration, and the signed normalized response deviation of the discharge, respectively, all of which are dimensionless parameters.

[0155] Furthermore, the first Current response deviation relationship for each control cycle ; The current response deviation relationship is used to jointly characterize the direction and relative degree of deviation of the actual drive power response, ball mill vibration response, and discharge response relative to the corresponding expected response. The sign of each signed normalized response deviation indicates the direction of deviation, and its absolute value indicates the relative magnitude of the deviation relative to the historical normal deviation scale.

[0156] It should be noted that the positive or negative deviation of a single response channel is not directly equivalent to an increase or decrease in the grinding load. In this embodiment, the current response deviation relationship composed of three response channels is used to match subsequent historical operating conditions in order to avoid directly inferring the grinding load status using a single operating parameter.

[0157] Furthermore, candidate historical control cycles with the same or similar feeding control conditions are selected from the historical control records of the target ball mill, and the historical response deviation relationship corresponding to each candidate historical control cycle is determined in the same way as the current control cycle: ; in, This indicates the sequence number of the historical control cycle preceding the current control cycle; Indicates the first The historical response deviation relationship corresponding to each candidate historical control cycle is used to jointly characterize the deviation direction and relative deviation degree of the three response channels of drive power, ball mill vibration and discharge relative to the corresponding historical expected grinding response in the historical control cycle; Indicates the first The driving power of each candidate historical control cycle has a signed standardized response deviation, which is a dimensionless parameter. Indicates the first The signed standardized response deviation of the ball mill vibration for each candidate historical control cycle is a dimensionless parameter. Indicates the first The signed standardized response deviation of the material discharge for each candidate historical control cycle is a dimensionless parameter.

[0158] Furthermore, based on the current response deviation relationship Deviation from historical response corresponding to each candidate historical control period The differences between them determine the corresponding response deviation relationship differences: ; in, Indicates the first The difference between the historical response deviation and the current response deviation corresponding to each candidate historical control cycle is a dimensionless parameter. The smaller the value, the closer the candidate historical control cycle is to the current control cycle in terms of the direction and relative degree of deviation in the three response channels of drive power, ball mill vibration, and discharge.

[0159] Furthermore, based on the differences in historical response deviations formed during the stable continuous grinding phase of the target ball mill, preset matching allowable conditions are set; wherein, the preset matching allowable conditions are represented by a preset response deviation matching threshold; specifically, when the difference in response deviations corresponding to a candidate historical control cycle is less than or equal to the preset response deviation matching threshold, the corresponding candidate historical control cycle is determined as a valid candidate historical operating condition; specifically, the preset response deviation matching threshold is a dimensionless parameter, determined based on the differences in historical response deviations under the same grinding load state, load migration direction, and feed control type during the stable continuous grinding phase of the target ball mill, and the maximum normal value among the historical response deviation differences obtained after excluding start-up, shutdown, no-load, maintenance, obvious material blockage, and obvious measurement anomalies is determined as the preset response deviation matching threshold.

[0160] Based on the order of the differences in response deviations corresponding to each valid candidate historical operating condition from smallest to largest, a preset number of historical control cycles are selected as historical reference operating conditions.

[0161] When there are no valid candidate historical working conditions that meet the preset matching allowable conditions, or when the number of valid candidate historical working conditions is insufficient to form a valid reference result, the grinding load state or load migration direction correction is not performed based on the historical matching result, and the initial grinding load state and initial load migration direction determined by the preset machine learning model are maintained.

[0162] In one implementation, the number of historical reference operating conditions can be 3 to 5, preferably 3. When the number of historical reference operating conditions is too small, occasional operational fluctuations in a single historical cycle may have a significant impact on the correction results. When the number is too large, historical operating conditions with significant differences from the current response deviation relationship may be introduced.

[0163] Based on the grinding load state and load migration direction corresponding to each historical reference working condition, the first... Reference grinding load state corresponding to each control cycle and reference load migration direction , where superscript This indicates that the corresponding parameters are reference results determined based on historical reference operating conditions.

[0164] In one implementation, the grinding load state and load migration direction corresponding to each historical reference working condition can be statistically analyzed, and the grinding load state that appears most frequently can be determined as the reference grinding load state. The load migration direction that occurs most frequently is determined as the reference load migration direction. .

[0165] Furthermore, when two or more grinding load conditions occur the same number of times, the response deviation relationship will differ. The grinding load state corresponding to the minimum historical reference working condition is determined as the reference grinding load state; when two or more load migration directions occur the same number of times, the difference in response deviation is considered. The load migration direction corresponding to the minimum historical reference operating condition is determined as the reference load migration direction.

[0166] Compare the reference grinding load state with the initial grinding load state. The reference load migration direction is compared with the initial load migration direction. The initial result is compared with the reference result; when the reference result is consistent with the initial result, the initial result is kept unchanged; when the reference result is inconsistent with the initial result, the initial result is corrected using the reference result; the initial grinding load state and the initial load migration direction are corrected and judged according to the reference grinding load state and the reference load migration direction.

[0167] Therefore, we obtain the first... The grinding load state determined after each control cycle correction and load migration direction .

[0168] According to the above and Following the method for determining the feed control parameters in step S4, the corresponding feed control type and feed control change amount are redefined, and the subsequent steps are updated accordingly. The target feed mass flow rate corresponding to each control cycle.

[0169] Under the current periodic index relationship, the first... The drift judgment and state correction results generated in the first control cycle are used to determine the first... The feeding control parameters corresponding to each control cycle are used to guide the grinding response drift correction results into the subsequent feeding control process.

[0170] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based intelligent control method for ball mill grinding processes, characterized in that, Includes the following steps: S1. Obtain grinding operation data of the ball mill in the current control cycle and historical control cycles, and construct ball mill operation sequence data; The grinding operation data includes at least the feed mass flow rate, drive power, ball mill vibration acceleration data, and discharge mass flow rate. S2. Determine the feed change, drive power change, ball mill vibration response change, and discharge change based on the ball mill running sequence data, and determine the response timing of drive power, ball mill vibration response, and discharge status relative to the feed change. Construct the grinding response feature set corresponding to the current control cycle based on each change and the corresponding response timing. S3. Input the grinding response feature set into a preset machine learning model to determine the current grinding load state of the ball mill and the corresponding load migration direction. S4. Determine the corresponding feed control parameters for the ball mill based on the grinding load state and load migration direction, and determine the expected grinding response for the next control cycle based on the feed control parameters. S5. Obtain the actual grinding response corresponding to the next control cycle, and determine the grinding response drift amount based on the deviation between the actual grinding response and the expected grinding response. S6. Based on the grinding response drift, the grinding load state and load migration direction determined by the preset machine learning model in the next control cycle are corrected and judged, and the corresponding grinding load state and load migration direction are determined based on the corrected judgment, thereby updating the corresponding feeding control parameters of the ball mill.

2. The intelligent control method for ball mill grinding process based on machine learning according to claim 1, characterized in that, In step S1, constructing the ball mill runtime sequence data specifically involves the following steps: The feed mass flow rate, drive power, ball mill vibration acceleration data, and discharge mass flow rate are periodically divided according to a preset control cycle. The corresponding cycle average feed mass flow rate, cycle average drive power, and cycle average discharge mass flow rate are determined based on the feed mass flow rate, drive power, and discharge mass flow rate in each control cycle, and the corresponding cycle vibration response value is determined based on the ball mill vibration acceleration data in each control cycle. The ball mill running sequence data are obtained by arranging the cycle average feed mass flow rate, cycle average drive power, cycle vibration response value, and cycle average discharge mass flow rate corresponding to the current control cycle and several consecutive historical control cycles before the current control cycle in the order of the control cycle.

3. The intelligent control method for ball mill grinding process based on machine learning according to claim 2, characterized in that, In step S2, constructing the grinding response feature set corresponding to the current control cycle involves the following steps: Based on the cycle average feed mass flow rate, cycle average drive power, cycle vibration response value, and cycle average discharge mass flow rate corresponding to the current control cycle and the previous control cycle, determine the feed change, drive power change, ball mill vibration response change, and discharge change corresponding to the current control cycle, respectively. The target feed change cycle that satisfies the preset feed change condition is determined from the ball mill operation sequence data. The target feed change cycle is used as the response reference cycle. In the subsequent continuous control cycle, the target response cycle that first satisfies the corresponding preset response condition for the drive power change, ball mill vibration response change and discharge change is determined respectively. Based on the period interval between each target response period and the response reference period, the number of drive power response lag periods, the number of ball mill vibration response lag periods, and the number of discharge response lag periods are determined respectively. Based on the changes in feed rate, drive power, ball mill vibration response, discharge rate, and the number of response lag cycles corresponding to the current control cycle, the grinding response feature set is constructed.

4. The intelligent control method for ball mill grinding process based on machine learning according to claim 3, characterized in that, In S3, the preset machine learning model includes a grinding load state classification sub-model and a load migration direction classification sub-model. Both the grinding load state classification sub-model and the load migration direction classification sub-model adopt a random forest classification model pre-established based on historical ball mill operation samples.

5. The intelligent control method for ball mill grinding process based on machine learning according to claim 4, characterized in that, Determining the current grinding load state of the ball mill and the corresponding load migration direction involves the following steps: The grinding response feature set corresponding to the current control cycle is input into the grinding load state classification sub-model and the load migration direction classification sub-model, respectively. The grinding load state of the ball mill is determined by the grinding load state classification sub-model, and the grinding load state includes at least low load state, suitable load state and high load state. The current load migration direction of the ball mill is determined by the load migration direction classification sub-model. The load migration direction includes at least the load increase direction, the load stability state, and the load decrease direction.

6. The intelligent control method for ball mill grinding process based on machine learning according to claim 5, characterized in that, In step S4, determining the feed control parameters corresponding to the ball mill involves the following steps: Based on the current grinding load status of the ball mill and the corresponding load migration direction, determine the feeding control type and feeding control change amount corresponding to the current control cycle; Based on the cycle average feed mass flow rate corresponding to the current control cycle, the feed control type, and the feed control change, the target feed mass flow rate corresponding to the next control cycle is determined, and the target feed mass flow rate is used as the feed control parameter corresponding to the ball mill. The feeding control types include increasing feeding, decreasing feeding, and maintaining feeding.

7. The intelligent control method for ball mill grinding process based on machine learning according to claim 6, characterized in that, In step S4, determining the expected grinding response for the next control cycle based on the feed control parameters includes the following steps: From the grinding operation data corresponding to the historical control cycle of the target ball mill, candidate historical control cycles with the same grinding load state, load migration direction, and feed control type as the current control cycle are determined. Based on the difference between the feed control change corresponding to each candidate historical control cycle and the feed control change corresponding to the current control cycle, the historical reference control cycle is determined. The changes in drive power, ball mill vibration response, and discharge mass flow rate are obtained for the next control cycle after the corresponding feed control is executed in the historical reference control cycle. Based on the changes in drive power, ball mill vibration response, and discharge mass flow rate corresponding to the historical reference control cycle, the expected changes in drive power, ball mill vibration response, and discharge mass flow rate for the next control cycle are determined, and the expected changes in drive power, ball mill vibration response, and discharge mass flow rate are used as the expected grinding response.

8. The intelligent control method for ball mill grinding process based on machine learning according to claim 7, characterized in that, In step S5, determining the grinding response drift based on the deviation between the actual grinding response and the expected grinding response involves the following steps: The cycle average drive power, cycle vibration response value and cycle average discharge mass flow rate corresponding to the next control cycle are obtained. Based on the difference between the parameters corresponding to the next control cycle and the current control cycle, the actual change in drive power, the actual change in ball mill vibration response and the actual change in discharge mass flow rate are determined respectively, and these are used as the actual grinding response. The drive power response deviation between the actual drive power change and the expected drive power change, the vibration response deviation between the actual ball mill vibration response change and the expected ball mill vibration response change, and the discharge response deviation between the actual discharge mass flow rate change and the expected discharge mass flow rate change are determined respectively. Based on the deviation between the historical actual grinding response and the historical expected grinding response corresponding to the stable continuous grinding stage of the target ball mill, the normalization benchmarks for the absolute deviation of the drive power response, the normalization benchmarks for the absolute deviation of the ball mill vibration response, and the normalization benchmarks for the absolute deviation of the discharge response are determined respectively. Based on the ratio of the absolute value of the drive power response deviation to the normalized reference of the absolute deviation of the drive power response, the ratio of the absolute value of the vibration response deviation to the normalized reference of the absolute deviation of the ball mill vibration response, and the ratio of the absolute value of the discharge response deviation to the normalized reference of the absolute deviation of the discharge response, the corresponding standardized response deviation is determined, and the grinding response drift is determined based on each of the standardized response deviations.

9. The intelligent control method for ball mill grinding process based on machine learning according to claim 8, characterized in that, In step S6, the feeding control parameters of the ball mill are updated according to the corrected grinding load state and load migration direction, involving the following steps: According to the most recent, including the next control cycle The grinding response drift amount corresponding to each continuous control cycle is used to determine the degree of continuous grinding response drift, and it is determined whether the degree of continuous grinding response drift meets the preset drift correction condition. The preset drift detection cycle number; When the preset drift correction condition is met, the current response deviation relationship is determined based on the drive power response deviation, ball mill vibration response deviation, and discharge response deviation corresponding to the next control cycle. From the historical control records of the target ball mill, determine the historical reference working condition that matches the current response deviation relationship, and based on the grinding load state and load migration direction corresponding to the historical reference working condition, correct and judge the grinding load state and load migration direction determined by the preset machine learning model in the next control cycle; Based on the correction judgment, the corresponding grinding load state and load migration direction are determined, and the feeding control parameters corresponding to the subsequent control cycle of the ball mill are updated.