A method and system for intelligent ball feeding dynamic control of ball mill based on real-time operating conditions

By aligning the real-time operating signals of the ball mill with a unified time window and evaluating data confidence, and combining feed disturbance compensation and hysteresis gating, a ball-addition request and instruction are generated, which solves the problem of mismatch in the amount of balls added during the replenishment of grinding media in the ball mill, and improves the stability of the equipment and the accuracy of the amount of balls added.

CN123102482APending Publication Date: 2026-11-17SHOUGANG LUANNAN MACHENG MINING CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611287625.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

During the replenishment of grinding media in a ball mill, due to real-time load changes and feed fluctuations, existing technologies struggle to achieve precise ball addition control, leading to a mismatch between the added ball quantity and demand. This can cause instantaneous changes in power, vibration, or acoustic signals, affecting the stable operation of the equipment.

Method used

By acquiring ball mill load-related signals and feed condition signals, unified time window alignment and data confidence evaluation are performed. Combined with feed disturbance compensation, ball load demand deviation is generated. Hysteresis gating and response coefficient adjustment are used to generate ball addition requests and instructions to ensure the accuracy and stability of ball addition.

Benefits of technology

It realizes intelligent ball feeding control of ball mill under real-time operating conditions, reduces mismatch caused by instantaneous fluctuations of a single signal, enhances adaptability to slow changes in operating conditions, and improves the stability of equipment operation and the accuracy of ball feeding.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123102482A_ABST
    Figure CN123102482A_ABST
Patent Text Reader

Abstract

This invention relates to the field of mineral processing technology, specifically to a method and system for intelligent ball mill ball loading dynamic control based on real-time operating conditions. This invention aligns the main motor load, cylinder vibration, or acoustic characterization with the feed rate, feed particle size index, and slurry concentration within a unified time window. It also integrates data confidence level and feed disturbance compensation into the calculation of ball load demand deviation. This allows the ball loading decision state to simultaneously characterize changes in mill load and external feed, maintaining a confirmed state even when signal alignment is insufficient or effective samples are lacking. This improves the distinction between continuous ball load demand deviation and short-term disturbances, thereby reducing mismatched ball loading caused by instantaneous fluctuations in a single signal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of mineral processing technology, specifically relating to a method and system for intelligent ball feeding dynamic control of ball mills based on real-time operating conditions. Background Technology

[0002] During the operation of a ball mill, the steel balls are gradually consumed due to wear, requiring the addition of balls to replenish the grinding media. A common method is to add balls according to a preset time and quantity, with the amount of balls added mainly based on a pre-set plan.

[0003] The real-time load of a ball mill is simultaneously affected by the feed rate, the state of the feed material, the state of the slurry, and the state of the grinding media. When these operating conditions change continuously, a deviation may occur between the fixed timing and quantity of ball addition and the current grinding media replenishment requirements. Short-term fluctuations in feed can also manifest in power, vibration, or acoustic signals in a manner similar to changes in ball load. If ball addition is triggered directly based on the instantaneous change of a single signal, it is easy to generate instructions that do not match actual needs.

[0004] To address the above problems, this invention proposes a method and system for intelligent ball feeding dynamic control of ball mills based on real-time operating conditions. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, the present invention aims to provide a method and system for intelligent ball mill ball feeding dynamic control based on real-time operating conditions.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for intelligent ball mill ball feeding dynamic control based on real-time operating conditions includes: The ball mill load-related signals and feed condition signals are acquired and aligned according to a unified time window to form a condition record; the load-related signals include at least one of the following: main motor active power, main motor current, effective value of cylinder vibration, and acoustic energy. Data confidence is evaluated based on operating condition records. The data confidence is determined by the proportion of valid sampling points, time alignment error, and outlier proportion. At the same time, the standardized deviation of the load-related signal is perturbed and compensated according to the changes in the feed condition signal to eliminate the influence of feed fluctuations on the load signal and obtain the ball load demand deviation, which characterizes the ball load replenishment demand inside the mill. When the data confidence level meets the threshold and the ball load demand deviation continues to reach the trigger threshold for at least N working condition windows, a ball addition request is generated, and the request state is maintained by hysteresis gating to avoid the deviation from repeatedly switching near the trigger boundary. In response to the ball loading request, the current ball load demand deviation is converted into ball loading mass according to the response coefficient, and the ball loading command is output after being rounded down by the smallest ball loading unit, and limited by the single upper limit and the rolling cumulative upper limit. The actual ball loading quality is obtained, and the response coefficient is updated online by recursively based on the change in ball load demand deviation according to the number of window delays before and after ball loading, so that the ball loading quality conversion for the next control cycle adopts the confirmed current mill response.

[0007] Preferably, the method further includes: projecting the original sampling records of each signal source onto a fixed-length operating window according to a unified clock; for signals with different sampling periods, aligning them by selecting the effective value closest to the target time according to their timestamp; when the time distance exceeds the set tolerance, marking the corresponding position as missing; replacing short-term missing signals with the middle value in this window; and retaining an unusable mark for signals with more than a set number of consecutive missing signals.

[0008] Preferably, the data confidence level is calculated as follows: the proportion of valid sampling points to the total number of sampling points within the operating window is calculated; the ratio of the maximum time alignment error to the alignment tolerance of each signal is calculated and limited to the range of zero to one; the proportion of outlier sampling points to the total number of valid sampling points is calculated; the above three items are assigned non-negative quality weights and then summed, and the result is limited to the range of zero to one. The larger the value, the higher the usability of the data in the current operating window; outlier sampling points are defined as points where the difference between the sampled value and the median value in the window exceeds a set multiple of the interquartile range.

[0009] Preferably, the ball load demand deviation is calculated as follows: the standardized deviation of each load-related signal relative to the reference operating range is calculated, and the results are weighted and summed according to a preset direction coefficient and weight; at the same time, the standardized change of each feed disturbance signal relative to the previous operating window is calculated, and the results are weighted and summed according to a preset disturbance compensation coefficient; the weighted sum of the load-related signals is subtracted from the weighted sum of the feed disturbance signals to obtain the ball load demand deviation after disturbance compensation.

[0010] Preferably, the hysteresis gating is configured with a trigger threshold and a release threshold, and the trigger threshold is greater than the release threshold. After the data validity gating passes, when the ball load demand deviation is less than or equal to the release threshold, the release judgment is executed first, the hysteresis state is set to request release and the instruction permission for this cycle is prohibited. When the ball load demand deviation is greater than or equal to the trigger threshold for N consecutive operating condition windows, and the set response observation interval has passed since the most recent request time, the hysteresis state is set to request validity and the instruction permission for this cycle is allowed. When the data confidence is lower than the confidence threshold or the feed disturbance state exceeds the allowable range, the trigger judgment and release judgment are not executed, the previous valid hysteresis state is maintained, and the instruction permission state for this cycle is set to prohibit.

[0011] Preferably, converting the current ball load demand deviation into ball loading mass specifically includes: when the ball loading request is valid, the current cycle instruction permission status is allowed, the ball loading execution unit is available, and the ball mill operation permission status is met, subtracting the release threshold from the current ball load demand deviation and dividing by the current response coefficient to obtain the unrestricted ball loading mass; then comparing this value with the smaller value between the single ball loading limit and the remaining allowable mass within the rolling time period, and taking the smaller one; then rounding down by the smallest ball loading unit to obtain the instruction mass for this operation; when the instruction mass for this operation exceeds the allowable mass of a single ball loading pulse, it is split into multiple pulses, with a set response observation interval left between adjacent pulses.

[0012] A ball mill intelligent ball feeding dynamic control system based on real-time operating conditions includes: The operating condition acquisition unit is connected to the operating condition detection end of the ball mill and is used to receive load-related signals and ore feeding operating condition signals, and form operating condition records by aligning them according to a unified time window; The state estimation unit, connected to the operating condition acquisition unit, is used to normalize the operating condition records, evaluate data confidence, and compensate for feed disturbances, generating ball load demand deviation, data confidence, and feed disturbance status; the data confidence is jointly determined by the proportion of effective sampling points, time alignment error, and outlier proportion; The ball-adding decision unit, connected to the state estimation unit, is used to generate a ball-adding request when the data confidence meets the threshold and the ball load demand deviation continues to reach the trigger threshold for at least N working condition windows, and the request state is maintained by hysteresis gating. The instruction generation unit, connected to the ball-adding decision unit, is used to respond to the ball-adding request. It converts the current ball load demand deviation into ball-adding mass according to the response coefficient, and outputs the ball-adding instruction after rounding down by the smallest ball-adding unit, limiting the single upper limit and the rolling cumulative upper limit. The ball-adding execution unit, connected to the instruction generation unit, is used to execute ball-adding according to the ball-adding instruction, and obtain the actual ball-adding mass and execution status by the weight difference before and after weighing; The feedback update unit is connected to the state estimation unit and the ball loading execution unit respectively. It is used to recursively update the response coefficient online and output it to the instruction generation unit based on the actual ball loading quality, execution status and the change in ball load demand deviation of the set window number before and after ball loading.

[0013] Preferably, the operating condition acquisition unit projects the original sampling records of each signal source onto a fixed-length operating condition window according to a unified clock. For signals with different sampling periods, the unit selects the effective value closest to the target time according to its timestamp for alignment. When the time distance exceeds the set tolerance, it is marked as missing. For short-term missing data, the middle value in this window is used as the replacement.

[0014] Preferably, the state estimation unit assigns non-negative quality weights to the proportion of effective sampling points, the proportion of time alignment error, and the proportion of outliers, and then performs a weighted summation, limiting the result to the range of zero to one to obtain the data confidence level; the state estimation unit calculates the standardized deviation of each load-related signal and performs a weighted summation according to a preset direction coefficient and weight, calculates the standardized change of each feed disturbance signal relative to the previous operating window and performs a weighted summation according to a preset disturbance compensation coefficient, and subtracts the weighted sum of the feed disturbance signals from the weighted sum of the load-related signals to obtain the ball load demand deviation.

[0015] Preferably, the feedback update unit forms a valid response sample when the execution status is normal, the actual ball loading mass is greater than zero, the data confidence level after delay reaches a threshold, and the ore feeding disturbance between the two working condition windows before and after ball loading is lower than a set threshold; the ball load demand deviation before ball loading is subtracted from the ball load demand deviation after ball loading, and then divided by the actual ball loading mass to obtain the current response observation value; the response coefficient is updated using a first-order recursive method, that is, the new response coefficient is equal to the weighted average of the current response coefficient and the current response observation value, and the result is limited to the preset upper and lower limits; the current response coefficient is maintained when the execution is abnormal or the response sample conditions are not met; The ball feeding execution unit includes a ball storage hopper, a weighing unit that supports the ball storage hopper and continuously outputs the weight value, and a ball feeding gate connected between the ball outlet of the ball storage hopper and the ball guiding channel. The outlet of the ball guiding channel leads to the ball feeding inlet of the ball mill. The weighing unit records the stable weight before and after the ball feeding gate operation. The actual ball feeding mass is obtained by subtracting the weight after the operation from the weight before the operation.

[0016] Beneficial effects 1. This invention aligns the main motor load, cylinder vibration or acoustic characterization with the feed rate, mill particle size index and slurry concentration within a unified time window, and combines data confidence and feed disturbance compensation into the calculation of ball load demand deviation. This allows the state of entering the ball loading decision to simultaneously characterize mill load changes and external feed changes, and maintains the confirmed state when signal alignment or effective samples are insufficient. This improves the distinction between continuous ball load demand deviation and short-term disturbances, thereby reducing mismatched ball loading caused by instantaneous fluctuations of a single signal.

[0017] 2. This invention employs a hysteresis gating system consisting of deviation persistence determination, trigger and release thresholds, and coordinates response coefficient conversion, minimum ball-adding unit quantization, single-time upper limit and rolling cumulative upper limit on the ball-adding command. Under boundary conditions, it maintains the request state and converts larger commands into ball-adding pulses with response intervals. At the same time, it ensures that each pulse falls within the measurable range of the ball-adding execution unit and maintains the observable interval between adjacent actions, stabilizing the actual amount added each time. This alleviates the problems of deviation between the fixed quantity and the current ball load requirement and frequent command switching.

[0018] 3. This invention uses the actual ball loading mass obtained from the weighing difference, the execution status, and the ball load demand deviation after the set delay to collaboratively form a response sample. When the data confidence level meets the standard and the feed disturbance is within the set range, the response coefficient is recursively updated, so that the ball loading mass conversion of the next control cycle adopts the confirmed current mill response, and samples with execution abnormalities or feed step changes are excluded from parameter updates, thereby enhancing the adaptability of control parameters to slow operating condition changes and reducing the ball loading deviation caused by long-term use of fixed conversion relationships. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.

[0021] Example 1 Please see Figure 1 As shown in the figure, this embodiment provides a method for intelligent ball mill ball feeding dynamic control based on real-time operating conditions. The method includes the following steps: S1. The working condition acquisition unit receives load-related signals and ore feeding working condition signals, aligns them according to a unified time window, and forms a working condition record. The system receives load-related signals and feed condition signals from the ball mill's operating condition detection terminal, and combines the timestamp of this sampling, signal value, signal quality status, and operating permission status into an original sampling record. The load-related signal includes at least one of the following: main motor active power, main motor current, effective value of cylinder vibration, and acoustic energy; the feed condition signal includes at least the feed rate, and may include one of the following: grinding particle size index and slurry concentration.

[0022] The operating condition acquisition unit projects the original sampling records onto a scale of [length missing] according to a uniform clock. Within the operating condition window, the preferred window length is 60 seconds, with a sliding interval of 10 seconds. For signals with different sampling periods, the most recent valid value is selected based on its timestamp. When the time distance exceeds the alignment tolerance, the corresponding position is marked as missing. Short-term missing signals are replaced with the median value within the window, while signals with more than a set number of consecutive missing signals are marked as unusable. This step outputs operating condition records organized according to a unified time reference for normalization and quality evaluation in step S2. Time alignment establishes a correspondence between the load signal and the feed signal within the same operating condition window, maintaining a consistent time reference for disturbance compensation in step S2.

[0023] S2. The state estimation unit normalizes, evaluates the quality of, and compensates for feed disturbances in the working condition records, generating ball load demand deviation, data confidence, and feed disturbance state. Read the operating condition record output in step S1, and calculate the median, interquartile range, rate of change, and effective sample ratio for each available signal; When the difference between a sampled value and the median exceeds a set multiple of the interquartile range, the sampled value is replaced by the median and included in the outlier count. The ratio of the number of valid sampled points of all required signals within the current window to the total number of required sampled points is used as the outlier count. The maximum time alignment error of each signal is divided by the alignment tolerance and then limited to 0 to 1 as the standard. The proportion of outlier sampling points to valid sampling points is used as... The data confidence level is calculated using the following formula:

[0024] In the formula, , and The quality weights are non-negative and the sum of the three is 1; when the number of valid sampling points is 0, the weights will be... Set it to 1. This yields... The value ranges from 0 to 1, with a larger value indicating a higher degree of data availability for the current operating condition window. The quality weight and alignment tolerance are fixed during the debugging phase to ensure that records under the same operating condition can reproduce the same level of data confidence.

[0025] Let the first The first working condition window The window statistics of each signal are Its center value and scale within the reference operating range are respectively and The standardization deviation is:

[0026] In the formula, This is the dimensionless standardized deviation. For signal index; Index for operating condition window; and A physical unit with a corresponding signal; If positive, the value represents the standard deviation or robust scalar value of the corresponding signal within the reference operating range. If the value is less than the set lower limit, use that lower limit instead to maintain the stability of the normalization calculation.

[0027] During commissioning, directional coefficients are set for the load-related signals respectively. and weight ,make A positive change uniformly indicates an increase in the demand for ball charge replenishment. The ball charge demand deviation is:

[0028] In the formula, This refers to the dimensionless ball load demand deviation; A set of load-related signal indices; This is a set of indices for ore feed disturbance signals; The weights are non-negative and the sum of all weights is 1; Take 1 or -1; For indexing ore disturbance signals; The non-negative disturbance compensation coefficient; For the first The standardized change of the feed disturbance signal relative to the previous operating window; This indicates the summation of all items within the corresponding set. If the feed disturbance signal is missing, the feed disturbance status of the corresponding window is set to unconfirmed, and subsequent steps maintain the historical hysteresis state and prohibit the formation of new instructions for this cycle.

[0029] This step outputs , The feed disturbance status is used to establish continuous conditions for step S3. Aligning the operating condition records with disturbance compensation within this step allows step S3 to determine the ball-addition demand based on load and feed changes within the same time window.

[0030] S3. The ball-adding decision unit generates a ball-adding request and the permission status of the instruction for this cycle based on data confidence, deviation persistence and hysteresis gating. The ball-adding decision unit reads the output of step S2. , The system checks the feed disturbance status and reads the deviation duration count, hysteresis status, latch-up request, and the most recent request time saved from the previous control cycle. Each decision is executed in the order of data validity gating, release decision, and trigger decision. When... When the level is below the confidence threshold or the ore feeding disturbance exceeds the allowable range, the release and trigger judgments are not executed. The previous valid hysteresis state and request latch are maintained, the permission state of the instruction in this cycle is set to prohibited, and this window is marked as a delayed judgment. The confidence threshold is preferably 0.85.

[0031] After the data validation gate passes, when When the threshold is less than or equal to the release threshold, the release judgment takes priority, setting the hysteresis state to request release, resetting the request latch, and setting the instruction permission state for this cycle to prohibit; otherwise, when continuous When a working condition window is greater than or equal to the trigger threshold, and the request latch is not set or the response observation interval has elapsed since the most recent request time, the hysteresis state is set to request valid, the request latch is set, and a ball-adding request is generated. At the same time, the instruction permission state for this cycle is set to allowed. The trigger threshold and release threshold are preferably 0.2 and 0.08, respectively.

[0032] After the generated ball-adding request is read in step S4, the current request event is consumed; the instruction permission status remains prohibited until the end of the response observation interval. The trigger threshold is greater than the release threshold; when the deviation is between the two thresholds or the consecutive window number has not been met, the historical hysteresis status is maintained but no new instruction permission is generated.

[0033] This step outputs the ball-adding request and the current status. The hysteresis state and the current cycle instruction permission state are used for step S4 to calculate the ball-adding quality. The confidence level of step S2, together with the persistence, hysteresis, and latch request of this step, reduces repeated triggering near the boundary and prevents new ball-adding instructions from being generated before the low-confidence window or before the response has been observed to be completed.

[0034] S4. When the ball-adding request is valid and the instruction is permitted, the instruction generation unit converts the demand deviation into the ball-adding quality, and outputs the ball-adding instruction after quantization, amplitude limiting and pulse splitting. The instruction generation unit reads the ball-adding request output in step S3. The system reads the hysteresis state and the current cycle instruction permission state, and simultaneously reads the response coefficient. The rolling time period includes the ball mass added, the availability status of the ball adding execution unit, and the operational permission status of the ball mill. Response coefficient. It represents the observable reduction in the deviation of the ball load requirement per unit of actual ball mass, and is expressed as the reciprocal of the mass unit and is greater than zero.

[0035] When the ball-adding request is valid, the current cycle instruction permission status is enabled, the ball-adding execution unit is available, and the ball mill's operating permission status is met, the current instruction is obtained using the following formula:

[0036] In the formula, To ensure the quality of this ball-adding instruction; This is a quantization operation that rounds down to the smallest ball-adding unit; To limit the value between a given lower and upper limit; To remove the threshold; This represents the current response coefficient; This is the maximum number of balls that can be added in a single instance. The remaining allowable quality within the rolling time period; This indicates taking the smaller of the two numbers; mass is expressed in kilograms or a unit of mass consistent with the weighing unit. When the calculated value is less than the smallest ball-adding unit, Output 0; when When the mass exceeds the allowable mass of a single ball-adding pulse, the instruction generation unit splits it into several pulses and leaves a set response observation interval between adjacent pulses.

[0037] When the instruction permission status for the current cycle is prohibited, the ball-adding execution unit is unavailable, or the ball mill's operating permission status is not met, the instruction generation unit does not call the aforementioned quality conversion formula, sets the current cycle instruction quality to 0, and does not generate new ball-adding pulses. This step outputs a ball-adding instruction containing instruction quality, pulse sequence, calculation time, and associated operating condition window, for execution in step S5. The request status in step S3 and the instruction permission status for the current cycle, in conjunction with the quantization limit in this step, convert the confirmed and executable continuous demand into a discrete control quantity that meets the capabilities of the ball-adding execution unit.

[0038] S5. The ball-adding execution unit executes the ball-adding operation according to the instruction, and obtains the actual ball-adding mass and execution status by the weight difference before and after weighing. The ball-adding execution unit reads the ball-adding command output in step S4. The weighing unit first records the weight of the ball storage hopper before the action, then opens the goal according to the command of the current pulse, and the steel ball enters the ball mill through the ball guide channel; after the goal is closed and the weighing value stabilizes, the weighing unit records the weight after the action. The actual ball-adding mass is obtained by subtracting the weight after the action from the weight before the action. .

[0039] The ball-adding execution unit will The instruction quality is compared with the current pulse. When the difference is within the set tolerance, a normal execution state is established; when the difference exceeds the set tolerance, an abnormal execution state is established, and subsequent pulses wait for the device state to recover. This step outputs... The execution status and instruction association identifiers are used in step S6 to match the operating condition windows before and after ball addition. The instruction quality given in step S4 is cross-checked with the actual quality obtained in this step, so that step S6 can establish a response sample based on the actual quality entering the ball mill. The preferred tolerance is 5 kg.

[0040] S6. The feedback update unit updates the response coefficient for use in the next cycle based on the actual ball-adding quality, execution status, and the deviation in delay requirements before and after ball-adding. The feedback update unit reads the output of step S5. The execution status and instruction association identifier are used to read the demand deviation before adding the ball. Let the sliding interval of the operating condition window be... Response delay window number If the integer is positive, then the response delay time elapsed since the end of execution is . The feedback update unit is in the first... The demand deviation is read from the new output of the status estimation unit in the individual operating condition window. When the execution status is normal, When the data confidence level is greater than 0, reaches the threshold after a delay, and the feed disturbance between the two operating windows is lower than the set threshold, then... This serves as a sample response.

[0041] Initial response coefficients During the commissioning phase, this was achieved through a controlled ball loading: when the feed disturbance was within acceptable limits and the operating data was valid, a known mass was added. ,according to calculate When controlled ball addition is not performed, the equipment's historical calibration value or current value can be used. and The preset values ​​between them are used as And update it according to the formula after obtaining the first valid response sample.

[0042] The response coefficients are updated using the following formula:

[0043] In the formula, The response coefficient to be used in the next control cycle; The update coefficient has a value range greater than 0 and less than or equal to 1, preferably 0.2; and These are the lower and upper limits of the allowable range for the response coefficient, respectively, both of which are greater than 0 and Less than ; This represents the number of operating condition windows spanned from the end of execution to the reading of the demand deviation after ball addition; other symbols have the same meaning as defined above. If an execution exception occurs or any response sample condition is not met, the current condition is maintained. And mark this record as not to participate in the update.

[0044] This step outputs and response sample status, The data is read from step S4 of the next control cycle. The actual ball-adding mass in step S5 is combined with the delayed operating condition response of this step, so that the next calculation in step S4 is based not only on the deviation, but also on the current response that has been confirmed by the completed execution.

[0045] Example 2 Please see Figure 2As shown, this embodiment provides an intelligent ball mill ball-adding dynamic control system based on real-time operating conditions, including an operating condition acquisition unit 110, a state estimation unit 120, a ball-adding decision unit 130, an instruction generation unit 140, a ball-adding execution unit 150, and a feedback update unit 160. This system 100 is connected to the operating condition detection end and ball-adding inlet of the ball mill 200.

[0046] The operating condition acquisition unit 110 is connected to the operating condition detection terminal of the ball mill 200, reads load-related signals and feed operating condition signals, aligns them with a unified clock according to the timestamp, forms an operating condition record, and outputs it to the state estimation unit 120. The state estimation unit 120 performs window statistics, outlier processing, reference interval normalization, and feed disturbance compensation on the operating condition record, forming ball load demand deviation, data confidence level, and feed disturbance status, which are then read by the ball addition decision unit 130.

[0047] The ball-loading decision unit 130 stores the continuous deviation count, hysteresis state, request latch, and the time of the most recent request. It only outputs a ball-loading request, current demand deviation, and current cycle instruction permission status to the instruction generation unit 140 when the data confidence level reaches a set threshold, the ball load demand deviation continuously meets the triggering conditions, and the feed disturbance state is within the allowable range. Low-confidence or disallowable feed disturbances only maintain a historical hysteresis state and do not generate new instruction permissions. The instruction generation unit 140 reads the response coefficients maintained by the feedback update unit 160 and the availability status of the ball-loading execution unit 150 and the operating permission status of the ball mill 200. Only when the current cycle instruction permission status, execution availability status, and operating permission status are all satisfied does it convert the demand deviation into unconstrained ball-loading mass, and then generate a ball-loading instruction based on the minimum ball-loading unit, single-time upper limit, and rolling cumulative upper limit.

[0048] The ball-feeding execution unit 150 includes a ball storage hopper, a weighing unit, a ball-feeding gate, and a ball-guiding channel. The weighing unit supports the ball storage hopper and continuously outputs the weight value. The ball-feeding gate is connected between the ball outlet of the ball storage hopper and the ball-guiding channel. The outlet of the ball-guiding channel leads to the ball-feeding inlet of the ball mill 200. The ball-feeding execution unit 150 controls the ball-feeding gate's action according to the ball-feeding command, uses the weight difference before and after the action to form the actual ball-feeding mass and execution status, and outputs both to the feedback update unit 160.

[0049] The feedback update unit 160 reads the actual ball loading quality and execution status output by the ball loading execution unit 150, and reads the newly formed ball load demand deviation from the status estimation unit 120 after the set response delay window number is reached. When execution is normal, data confidence reaches the threshold, and the ore feeding disturbance state is within the allowable range, the feedback update unit 160 generates a response sample, updates the response coefficient, and outputs it to the instruction generation unit 140; when execution is abnormal or the response sample conditions are not met, the current response coefficient is maintained. The initial response coefficient, obtained through debugging and calibration and limited to the allowable range, is used during the first run. It also saves the sample states that were not updated for subsequent operational status verification. In this way, the data flow, ball loading instruction flow, and actual ball loading feedback flow of the system 100 form a continuous loop.

[0050] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for intelligent ball mill ball feeding dynamic control based on real-time operating conditions, characterized in that, include: Acquire ball mill load-related signals and ore feed condition signals, align them according to a unified time window, and form a condition record. Load-related signals include at least one of the following: main motor active power, main motor current, effective value of cylinder vibration, and acoustic energy. Data confidence is evaluated based on operating condition records. The data confidence is determined by the proportion of valid sampling points, time alignment error, and outlier proportion. At the same time, the standardized deviation of the load-related signal is perturbed and compensated according to the changes in the feed condition signal to eliminate the influence of feed fluctuations on the load signal and obtain the ball load demand deviation, which characterizes the ball load replenishment demand inside the mill. When the data confidence level meets the threshold and the ball load demand deviation continues to reach the trigger threshold for at least N working condition windows, a ball addition request is generated, and the request state is maintained by hysteresis gating to avoid the deviation from repeatedly switching near the trigger boundary. In response to the ball loading request, the current ball load demand deviation is converted into ball loading mass according to the response coefficient, and the ball loading command is output after being rounded down by the smallest ball loading unit, and limited by the single upper limit and the rolling cumulative upper limit. The actual ball loading quality is obtained, and the response coefficient is updated online by recursively based on the change in ball load demand deviation according to the number of window delays before and after ball loading, so that the ball loading quality conversion for the next control cycle adopts the confirmed current mill response.

2. The intelligent ball-feeding dynamic control method for ball mills based on real-time operating conditions according to claim 1, characterized in that, Also includes: The original sampling records of each signal source are projected onto a fixed-length operating window according to a unified clock. For signals with different sampling periods, the effective value closest to the target time is selected for alignment according to its timestamp. When the time distance exceeds the set tolerance, the corresponding position is marked as missing. For short-term missing signals, the number of bits in this window is used for replacement. For signals with continuous missing signals exceeding the set number, an unusable mark is retained.

3. The intelligent ball-feeding dynamic control method for ball mills based on real-time operating conditions according to claim 1, characterized in that, The data confidence level is calculated as follows: the proportion of valid sampling points to the total number of sampling points that should be present in the operating window is calculated; the ratio of the maximum time alignment error to the alignment tolerance of each signal is calculated and limited to the range of zero to one; the proportion of outlier sampling points to the total number of valid sampling points is calculated; after assigning non-negative quality weights to the above three items, they are weighted and summed, and the result is limited to the range of zero to one. The larger the value, the higher the availability of data in the current operating window. Outlier sampling points are defined as points where the difference between the sampled value and the middle value in the window exceeds a set multiple of the interquartile range.

4. The intelligent ball-feeding dynamic control method for a ball mill based on real-time operating conditions according to claim 1, characterized in that, The ball load demand deviation is calculated as follows: the standardized deviation of each load-related signal relative to the reference operating range is calculated, and the results are weighted and summed according to preset direction coefficients and weights; at the same time, the standardized change of each feed disturbance signal relative to the previous operating window is calculated, and the results are weighted and summed according to preset disturbance compensation coefficients; the weighted sum of the load-related signals is subtracted from the weighted sum of the feed disturbance signals to obtain the ball load demand deviation after disturbance compensation.

5. The intelligent ball-feeding dynamic control method for a ball mill based on real-time operating conditions according to claim 1, characterized in that, Hysteresis gating is configured with a trigger threshold and a release threshold, and the trigger threshold is greater than the release threshold. After the data validity gating passes, when the ball load demand deviation is less than or equal to the release threshold, the release judgment is executed first, the hysteresis state is set to request release and the instruction permission for this cycle is prohibited. When the ball load demand deviation is greater than or equal to the trigger threshold for N consecutive operating conditions, and the set response observation interval has elapsed since the most recent request time, the hysteresis state is set to request valid and the instruction permission for this cycle is allowed; when the data confidence is lower than the confidence threshold or the ore feeding disturbance state exceeds the allowable range, the trigger judgment and release judgment are not executed, the previous valid hysteresis state is maintained, and the instruction permission state for this cycle is set to prohibited.

6. The intelligent ball-feeding dynamic control method for a ball mill based on real-time operating conditions according to claim 1, characterized in that, Converting the current ball load demand deviation into ball loading mass specifically involves: when the ball loading request is valid, the current cycle instruction permission status is allowed, the ball loading execution unit is available, and the ball mill operation permission status is met, subtracting the release threshold from the current ball load demand deviation and dividing by the current response coefficient yields the unrestricted ball loading mass; then comparing this value with the smaller of the single ball loading limit and the remaining allowable mass within the rolling time period, and taking the smaller one; subsequently rounding down by the smallest ball loading unit to obtain the instruction mass for this cycle; when the instruction mass for this cycle exceeds the allowable mass of a single ball loading pulse, it is split into multiple pulses, with a set response observation interval left between adjacent pulses.

7. A ball mill intelligent ball feeding dynamic control system based on real-time operating conditions, characterized in that, include: The operating condition acquisition unit is connected to the operating condition detection end of the ball mill and is used to receive load-related signals and ore feeding operating condition signals, and form operating condition records by aligning them according to a unified time window; The state estimation unit, connected to the operating condition acquisition unit, is used to normalize the operating condition records, evaluate data confidence, and compensate for feed disturbances, generating ball load demand deviation, data confidence, and feed disturbance status; the data confidence is jointly determined by the proportion of effective sampling points, time alignment error, and outlier proportion; The ball-adding decision unit, connected to the state estimation unit, is used to generate a ball-adding request when the data confidence meets the threshold and the ball load demand deviation continues to reach the trigger threshold for at least N working condition windows, and the request state is maintained by hysteresis gating. The instruction generation unit, connected to the ball-adding decision unit, is used to respond to the ball-adding request. It converts the current ball load demand deviation into ball-adding mass according to the response coefficient, and outputs the ball-adding instruction after rounding down by the smallest ball-adding unit, limiting the single upper limit and the rolling cumulative upper limit. The ball-adding execution unit, connected to the instruction generation unit, is used to execute ball-adding according to the ball-adding instruction, and obtain the actual ball-adding mass and execution status by the weight difference before and after weighing; The feedback update unit is connected to the state estimation unit and the ball loading execution unit respectively. It is used to recursively update the response coefficient online and output it to the instruction generation unit based on the actual ball loading quality, execution status and the change in ball load demand deviation of the set window number before and after ball loading.

8. The intelligent ball-feeding dynamic control system for a ball mill based on real-time operating conditions according to claim 7, characterized in that, The operating condition acquisition unit projects the original sampling records of each signal source onto a fixed-length operating condition window according to a unified clock. For signals with different sampling periods, it selects the effective value closest to the target time according to its timestamp for alignment. When the time distance exceeds the set tolerance, it is marked as missing. For short-term missing data, the middle value in this window is used as the replacement.

9. The intelligent ball-feeding dynamic control system for a ball mill based on real-time operating conditions according to claim 7, characterized in that, The state estimation unit assigns non-negative quality weights to the proportion of effective sampling points, the proportion of time alignment error, and the proportion of outliers, and then sums them up by weighting. The result is then limited to the range of zero to one to obtain the data confidence level. The state estimation unit calculates the standardized deviation of each load-related signal and sums it up by weighting it according to the preset direction coefficient and weight. It also calculates the standardized change of each feed disturbance signal relative to the previous operating window and sums it up by weighting it according to the preset disturbance compensation coefficient. The weighted sum of the load-related signals is subtracted from the weighted sum of the feed disturbance signals to obtain the ball load demand deviation.

10. The intelligent ball-feeding dynamic control system for a ball mill based on real-time operating conditions according to claim 7, characterized in that, The feedback update unit generates a valid response sample when the execution status is normal, the actual ball loading mass is greater than zero, the data confidence level after delay reaches the threshold, and the ore feeding disturbance between the two working condition windows before and after ball loading is lower than the set threshold. The ball loading demand deviation before ball loading is subtracted from the ball loading demand deviation after ball loading, and then divided by the actual ball loading mass to obtain the current response observation value. The response coefficient is updated using a first-order recursive method, that is, the new response coefficient is equal to the weighted average of the current response coefficient and the current response observation value, and the result is limited to the preset upper and lower limits. The current response coefficient is maintained when the execution is abnormal or the response sample conditions are not met. The ball feeding execution unit includes a ball storage hopper, a weighing unit that supports the ball storage hopper and continuously outputs the weight value, and a ball feeding gate connected between the ball outlet of the ball storage hopper and the ball guiding channel. The outlet of the ball guiding channel leads to the ball feeding inlet of the ball mill. The weighing unit records the stable weight before and after the ball feeding gate operation. The actual ball feeding mass is obtained by subtracting the weight after the operation from the weight before the operation.