Material particle size monitoring method and system for basic cobalt carbonate production

Abstract

The invention discloses a material particle size monitoring method and system for basic cobalt carbonate production, and relates to the technical field of material particle size monitoring. Through a multi-mode sensing data fusion and intelligent decomposition algorithm, the particle size monitoring precision under the working conditions of high solid content and strong turbulence is remarkably improved; according to the method, physical constraint and a dynamic splitting mechanism are combined, seed crystal particles and abnormal secondary nucleation particles are effectively distinguished, and the problem that double peaks are misjudged as broad peaks in a traditional method is avoided; meanwhile, on the basis of a current reference threshold strategy of sliding window statistics, different batches and working condition changes are adaptively matched, the manual intervention requirement is reduced, stirring parameters are adjusted and quantified into a function of a dispersion uniformity index, the limitation of empirical control is broken through, frequency domain compensation and adaptive filtering design are additionally introduced, and the dispersion uniformity is improved. And the influence of complex interference sources such as bubbles and mechanical vibration is obviously inhibited.

Description

Technical Field

[0001] The present invention relates to the technical field of material particle size monitoring, and particularly to a method and system for monitoring the particle size of materials used in the production of cobalt basic carbonate. Background Art

[0002] In the continuous production process of cobalt basic carbonate, the uniform dispersion of seeds is the core link for controlling the unimodal distribution of product particle size; currently, mainstream production lines generally adopt online laser particle size analyzers or ultrasonic attenuation spectroscopy analysis technologies to dynamically adjust the stirring intensity and the seed replenishment speed by monitoring the change of particle size in the reaction kettle in real time; however, under the conditions of high solid content > 25% and strong turbulent stirring speed > 1000 r / min, seed particles are prone to form a competitive growth mechanism with newly formed particles due to uneven local concentration gradient and fluid shear force distribution, ultimately resulting in a significant bimodal characteristic in the particle size distribution curve; this phenomenon is particularly prominent when preparing lithium cobaltate precursor with high tap density, directly affecting the compaction performance and cycle life of the battery cathode material; Although in recent years, light scattering signal analysis algorithms and multi-sensor fusion schemes based on machine learning have emerged, they still have essential defects in the bimodal distribution scenario. Traditional feature extraction methods are difficult to distinguish the seed-induced peak from the abnormal secondary nucleation peak, and mismerge the bimodal peaks into a broad peak for processing; in addition, the particle size optical response model relying on offline calibration cannot adapt to the real-time changing flow field environment, resulting in insufficient decoupling accuracy of the mixed distribution; for example, in the patent disclosed in Publication No. CN104418388B, although the eddy current enhanced dispersion technology adopted improves the initial distribution uniformity of seeds, during the continuous reaction process, the periodic appearance of stirring dead zones will still cause the local seed concentration to fluctuate beyond the critical value, and the existing monitoring system lacks effective sensing ability for such microscale heterogeneities.

[0003] In view of the above problems, some solutions alleviate the bimodal phenomenon by combining a pulsed seed injection strategy with an adaptive optical compensation algorithm, including: using high-frequency, small-dose pneumatic injection to replace batch feeding to reduce the risk of local supersaturation; introducing a particle size distribution prediction model based on transfer learning to infer the seed dispersion state through historical data; however, such methods are limited by the hardware response speed and the model generalization ability, and still exhibit control lag or even mis-triggering when dealing with sudden flow field disturbances, such as stirring shaft vibration offset and feed pipe blockage, and it is difficult to fundamentally suppress the formation of bimodal distribution. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a method and system for monitoring the particle size of materials used in the production of cobalt basic carbonate to solve the problem that the existing monitoring methods are difficult to distinguish the mixed distribution of seeds and newly formed particles under high solid content, and the frequent occurrence of particle size bimodal phenomenon caused by control lag restricts the product consistency.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for monitoring the particle size of materials used in the production of cobalt basic carbonate, which includes: Step S1, collecting multi-source sensing data in the reaction kettle in real time, where the multi-source sensing data includes laser scattering intensity angular distribution signals, ultrasonic attenuation spectrum signals, and stirring shaft rotation speed signals; Step S2, generating a first particle size distribution curve based on the laser scattering intensity angular distribution signal, and simultaneously generating a second particle size distribution curve according to the ultrasonic attenuation spectrum signal; Step S3, inputting the first particle size distribution curve and the second particle size distribution curve into a multi-modal distribution decomposition model, and outputting the decomposed seed particle distribution sub-curve and the new particle distribution sub-curve; Step S4, calculating the peak proportion of the seed particle distribution sub-curve, and when the peak proportion is lower than a preset threshold, generating a stirring parameter adjustment instruction; Step S5, sending the stirring parameter adjustment instruction to the stirring drive device to dynamically adjust the inclination angle and rotation speed gradient of the stirring paddle.

[0007] As a preferred scheme of the method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to the present invention, wherein: the construction of the multi-modal distribution decomposition model includes: Performing Gaussian mixture modeling on the particle size distribution curves in the historical production data, and constraining the particle size distribution of the seed particles to conform to the preset nominal distribution characteristics; Using KL divergence to dynamically evaluate the deviation degree between the real-time distribution curve and the historical distribution curve, and when the deviation degree exceeds the set critical value, triggering the distribution curve splitting operation; Outputting the decomposition result including the seed particle distribution sub-curve and the new particle distribution sub-curve; The multi-modal distribution decomposition model includes a feature alignment layer, a cross-attention layer, and a physical constraint layer, wherein: The feature alignment layer performs timestamp synchronization and spatial registration on the laser scattering intensity angular distribution signal and the ultrasonic attenuation spectrum signal; The cross-attention layer calculates the correlation weight matrix of the two modal signals in the frequency domain-spatial domain; The physical constraint layer enforces the satisfaction of the particle volume conservation equation.

[0008] As a preferred scheme of the method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to the present invention, wherein: the distribution curve splitting operation includes the following steps: a) Detecting the peak-valley ratio of the real-time distribution curve, and determining it as a potential double peak when the valley value is higher than 20% of the peak heights on both sides; b) Perform second derivative zero-crossing detection on the suspected bimodal region to locate the center position of the sub-peak; c) Dynamically increase the number of components of the Gaussian mixture model according to the sub-peak spacing.

[0009] As a preferred solution of the method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to the present invention, wherein: in step S3, the Gaussian mixture modeling of the historical particle size distribution and the way of constraining the seed nominal distribution are as follows: For the centered particle size variable Use Gaussian components to construct a mixture density function: , wherein, represents the weight of the th Gaussian component, represents the total number of components of the Gaussian mixture model, represents the variance of the th component, represents the mean of the th component, represents the normalized and centered particle size variable, represents the probability density of the mixture model at the particle size ; Introduce the KL divergence constraint on the component numbered and the seed nominal distribution to make this component close to the preset characteristics. Let the overall optimization objective be the sum of the log-likelihood and the constrained divergence: , wherein, represents the total number of historical samples, represents the particle size value of the th sample, represents the constraint strength coefficient, represents the Kullback-Leibler divergence between two normal distributions, represents the seed nominal mean, represents the seed nominal variance, represents the component index assigned to the seed characteristics, and respectively represent the mean and variance of the th component; , wherein, in the denominator, and correspond to the variances of the compared distributions before and after, represents the square of the mean deviation between the two distributions, and the constant 1 is used to standardize the divergence baseline; Adopt improved EM iterative minimization , including: E-step: Calculate the posterior probability: , M-step: , wherein, represents the posterior probability that the th sample belongs to the th component, represents the density value of the th Gaussian component at , represents the particle size measurement value of the th historical sample after normalization and centering processing, represents the weight of the th component after updating in this round of M-step, represents the updated mean value, represents the updated variance; For the th component, add gradient correction after M-step: , wherein, represents the learning rate for regular gradient correction, and repeat the iteration until , represents the mean value of the th component belonging to the seed classification index after updating in this round of M-step and gradient correction, represents the variance of the th component belonging to the seed classification index after updating in this round of M-step and gradient correction.

[0010] As a preferred scheme of the method for monitoring the particle size of materials for the production of cobalt basic carbonate according to the present invention, wherein: the step of generating the stirring parameter adjustment instruction includes: Calculate the dispersion uniformity index according to the peak width of the seed particle distribution sub-curve; Establish the mapping relationship between the dispersion uniformity index and the inclination angle of the stirring paddle to generate the inclination compensation amount; Based on the current stirring speed and the material viscosity, calculate the step size of the rotational speed gradient adjustment.

[0011] As a preferred scheme of the method for monitoring the particle size of materials for the production of cobalt basic carbonate according to the present invention, wherein: in step S4, when calculating the peak ratio and judging whether to trigger stirring adjustment according to the current reference threshold, take the mixing weight of the seed component in the decomposition model as the peak ratio , for working condition fluctuations, a current reference threshold based on sliding window statistics is designed , the process includes: The mixing weight represents the proportion of seed particles in the overall distribution, and is used to measure the seed abundance. The current reference threshold is adaptively adjusted through historical data , Among them, represents the mixing weight of the seed component, that is, the weight of the component in the Gaussian mixture model; The window statistical radius is set to sub-sampling times, and calculate the mean and standard deviation of the historical peak proportion: , Among them, represents the historical sampling window length, represents the time of the peak proportion, and are the mean and standard deviation within this window respectively; Based on the statistical results, after setting the safety factor , the current reference threshold is: , Among them, represents the safety factor, which is used to adjust the triggering sensitivity, represents the current reference threshold; When the real-time peak proportion is lower than the current reference threshold , an instruction to adjust the stirring parameters is immediately generated to restore the seed abundance.

[0012] In the second aspect, the present invention provides a material particle size monitoring system for the production of cobalt basic carbonate, including a multi-source sensing module, which includes a laser scattering probe array, an ultrasonic transceiver and a stirring shaft encoder installed on the side wall of the reaction kettle; a signal processing module, connected to the multi-source sensing module, configured to perform noise reduction processing on the original sensing data and generate a first particle size distribution curve and a second particle size distribution curve; a distribution decomposition module, integrating the multi-modal distribution decomposition model, receiving the first particle size distribution curve and the second particle size distribution curve and outputting the decomposition result; a control decision-making module, connected to the distribution decomposition module, configured to generate an instruction to adjust the stirring parameters according to the peak proportion; an actuator interface module, which converts the stirring parameter adjustment instruction into a drive signal and sends it to the stirring drive device.

[0013] As a preferred solution of the material particle size monitoring system for cobalt basic carbonate production according to the present invention, wherein: the arrangement mode of the laser scattering probe array is as follows: Three detection planes are arranged in the axial height direction of the reaction kettle, and three probes are annularly distributed at 120° in each plane; The included angle between the incident optical path of each probe and the central line of the stirring shaft is 15°-30°, and the adjacent layer probes are arranged with a 40° circumferential dislocation.

[0014] As a preferred solution of the material particle size monitoring system for cobalt basic carbonate production according to the present invention, wherein: the signal processing module includes: An adaptive filtering unit for eliminating the periodic mechanical vibration noise caused by the rotation of the stirring blades; A frequency domain compensation unit for dynamically adjusting the frequency band weight coefficient of the ultrasonic attenuation spectrum according to the real-time solid content data; A data synchronization unit for data synchronization.

[0015] As a preferred solution of the material particle size monitoring system for cobalt basic carbonate production according to the present invention, wherein: the control decision module further includes: An emergency intervention unit, which automatically triggers the start of the seed addition device when it is detected that the peak particle size of the new particle distribution sub-curve exceeds 1.5 times the peak particle size of the seed particles; The discharge port of the seed addition device is provided with an ultrasonic atomizing head, and the atomization particle size is controlled within the range of 50-100μm.

[0016] The beneficial effects of the present invention are as follows: through the multi-modal sensing data fusion and intelligent decomposition algorithm, the present invention significantly improves the particle size monitoring accuracy under the conditions of high solid content and strong turbulence; by combining physical constraints and dynamic splitting mechanisms, the present invention effectively distinguishes seed particles and abnormal secondary nucleation particles, avoiding the problem of misjudging double peaks as wide peaks in traditional methods; at the same time, based on the dynamic threshold strategy of sliding window statistics, it adaptively matches different batches and working condition changes, reduces the need for manual intervention, quantifies the adjustment of stirring parameters as a function of the dispersion uniformity index, breaks through the limitations of empirical control, and additionally introduces frequency domain compensation and adaptive filtering designs to significantly suppress the influence of complex interference sources such as bubbles and mechanical vibrations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0018] Figure 1Schematic flow diagram of the method for monitoring the particle size of materials used in the production of cobalt basic carbonate in Example 1.

[0019] Figure 2 Schematic framework diagram of the system for monitoring the particle size of materials used in the production of cobalt basic carbonate in Example 1. Detailed implementation manners

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings of the specification.

[0021] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or selectively exclusive embodiment with other embodiments.

[0023] Example 1, referring to Figure 1 and Figure 2 , this example provides a method for monitoring the particle size of materials used in the production of cobalt basic carbonate, including the following steps: Step S1, collect multi-source sensing data in the reaction kettle in real time. The multi-source sensing data includes laser scattering intensity angular distribution signals, ultrasonic attenuation spectrum signals, and stirring shaft rotation speed signals; Step S2, generate a first particle size distribution curve based on the laser scattering intensity angular distribution signal, and at the same time generate a second particle size distribution curve according to the ultrasonic attenuation spectrum signal; Step S3, input the first particle size distribution curve and the second particle size distribution curve into a multi-modal distribution decomposition model, and output the decomposed seed particle distribution sub-curve and the new particle distribution sub-curve; The construction of the multi-modal distribution decomposition model includes: Perform Gaussian mixture modeling on the particle size distribution curves in the historical production data, and constrain the particle size distribution of the seed particles to conform to the preset nominal distribution characteristics; Dynamically evaluate the deviation degree between the real-time distribution curve and the historical distribution curve using the KL divergence. When the deviation degree exceeds the set critical value, trigger the distribution curve splitting operation; Output the decomposition result including the seed particle distribution sub-curve and the new particle distribution sub-curve; The multi-modal distribution decomposition model includes a feature alignment layer, a cross-attention layer, and a physical constraint layer, where: The feature alignment layer performs timestamp synchronization and spatial registration on the laser scattering intensity angular distribution signal and the ultrasonic attenuation spectrum signal; The cross-attention layer calculates the correlation weight matrix of the two modal signals in the frequency-spatial domain; The physical constraint layer enforces the satisfaction of the particle volume conservation equation, making the total volume of the decomposed sub-curves consistent with the original distribution; The distribution curve splitting operation includes the following steps: a) Detect the peak-valley ratio of the real-time distribution curve. When the valley value is higher than 20% of the peak heights on both sides, it is determined as a potential double peak; b) Perform a second derivative zero-crossing detection on the suspected double-peak region to locate the center position of the sub-peak; c) Dynamically increase the number of components of the Gaussian mixture model according to the sub-peak spacing; In step S3, the method of Gaussian mixture modeling of the historical particle size distribution and the constraint of the seed nominal distribution is as follows: For the centered particle size variable Adopt Gaussian components to construct the mixture density function: , where, Represents the weight of the th Gaussian component, Represents the total number of components of the Gaussian mixture model, Represents the variance of the th component, Represents the mean of the th component, Represents the normalized and centered particle size variable, Represents the probability density of the mixture model at the particle size ; Introduce the KL divergence constraint on the component numbered and the seed nominal distribution to make this component close to the preset features. Let the overall optimization objective be the sum of the log-likelihood and the constraint divergence: , where, Represents the total number of historical samples, Represents the particle size value of the th sample, Represents the constraint intensity coefficient, Represents the Kullback-Leibler divergence between two normal distributions, Represents the seed nominal mean, Represents the seed nominal variance, Indicates the component index assigned to the seed feature, and respectively represent the mean and variance of the th component; , where the and in the denominator are the variances of the compared distributions before and after, represents the squared deviation of the means of the two distributions, and the constant 1 is used to standardize the divergence baseline; Adopt improved EM iteration to minimize , including: E-step: Calculate the posterior probability: , M-step: , where, represents the posterior probability that the th sample belongs to the th component, represents the density value of the th Gaussian component at , represents the particle size measurement value of the th historical sample after normalization and centering, represents the weight of the th component after update in this round of M-step, represents the updated mean, represents the updated variance; For the components, additional gradient correction is applied after the M-step: , where, represents the learning rate for regular gradient correction, and repeat the iteration until , represents the mean of the components belonging to the seed classification index after update and gradient correction in this round of M-step, represents the variance of the components belonging to the seed classification index after update and gradient correction in this round of M-step; Specifically, here the Gaussian mixture model is tightly coupled with the nominal distribution of the target component through KL divergence, so that the selected components always expand around the preset peak position of the seed without deviation. In the EM iteration, additional gradient correction is applied after the M-step, so that the nominal distribution constraint is naturally incorporated into the parameter update in the form of an analytical gradient, without numerical approximation, improving the optimization efficiency. At the same time, it can be adjusted by Dependence degree of the control model on the prior; Step S4: Calculate the peak proportion of the seed particle distribution sub-curve. When the peak proportion is lower than the preset threshold, generate a stirring parameter adjustment instruction; The generation steps of the stirring parameter adjustment instruction include: Calculate the dispersion uniformity index according to the peak width of the seed particle distribution sub-curve; Establish the mapping relationship between the dispersion uniformity index and the inclination angle of the stirring blade, and generate the inclination angle compensation amount; Based on the current stirring speed and the material viscosity, calculate the adjustment step size of the speed gradient; In step S4, when calculating the process of triggering the stirring adjustment according to the peak proportion and judging with the current reference threshold, take the mixing weight of the seed component in the decomposition model as the peak proportion , and design the current reference threshold based on the sliding window statistics for the working condition fluctuation , and the process includes: The mixing weight represents the proportion of the seed particles in the overall distribution, and is used to measure the seed abundance. The current reference threshold is adaptively adjusted through historical data, , wherein, represents the mixing weight of the seed component, that is, the weight of the component in the Gaussian mixture model; The window statistical radius is set to sub-samplings, and calculate the mean and standard deviation of the historical peak proportion: , wherein, represents the length of the historical sampling window, represents the time of the peak proportion, and are the mean and standard deviation within this window respectively; Based on the statistical results, set the safety factor and then, the current reference threshold is: , wherein, represents the safety factor, which is used to adjust the triggering sensitivity, represents the current reference threshold; When the real-time peak proportion is lower than the current reference threshold , immediately generate a stirring parameter adjustment instruction to restore the seed abundance; Specifically, here, the mixing weight of the seed component directly output by the Gaussian mixture model is used as the seed peak proportion to avoid the computational overhead caused by the secondary numerical integration. The current reference threshold Then, historical trends and fluctuation information are introduced through sliding window statistics, and the safety factor can be flexibly set, enabling the judgment rule to not only quickly respond to insufficient seeds but also resist short-term abnormal interference. Compared with a fixed threshold, the dynamic threshold is more adaptable, suitable for different batches and working conditions, and can significantly reduce false triggers and missed triggers; Step S5: Send the stirring parameter adjustment instruction to the stirring drive device to dynamically adjust the inclination angle and rotational speed gradient of the stirring blades.

[0024] This embodiment also provides a monitoring system for the particle size of materials used in the production of cobalt basic carbonate, including: A multi-source sensing module, including a laser scattering probe array, an ultrasonic transceiver, and a stirring shaft encoder installed on the side wall of the reaction kettle; The arrangement method of the laser scattering probe array is as follows: Three detection planes are set in the axial height direction of the reaction kettle, and three probes are distributed in a 120° circular pattern in each plane; The included angle between the incident light path of each probe and the center line of the stirring shaft is 15° - 30°, and the adjacent layer probes are arranged with a 40° circumferential offset; A signal processing module, connected to the multi-source sensing module, configured to perform noise reduction processing on the original sensing data and generate a first particle size distribution curve and a second particle size distribution curve; The signal processing module includes: An adaptive filtering unit for eliminating periodic mechanical vibration noise caused by the rotation of the stirring blades; A frequency domain compensation unit for dynamically adjusting the frequency band weight coefficient of the ultrasonic attenuation spectrum according to real-time solid content data; A data synchronization unit for data synchronization, keeping the acquisition time stamp deviation between the laser scattering signal and the ultrasonic signal within 10 ms; In the frequency domain compensation unit, a dynamic adjustment mechanism for the frequency band weight based on real-time solid content is designed: The frequency spectrum is divided into frequency bands, and for the th frequency band, the solid content sensitivity coefficient is obtained, and the unnormalized weight is generated through exponential mapping: , where represents the solid content sensitivity coefficient corresponding to the th frequency band, represents the currently measured real-time solid content of the material, represents the reference solid content level, represents the frequency band 's unnormalized weight, , represents the total number of frequency bands; Normalize all frequency bands to obtain dynamic weights: , where, represents the normalized weight of the th frequency band, represents the sum of all unnormalized weights; Apply this weight to the original ultrasonic attenuation spectrum coefficient to obtain the compensated frequency band attenuation value: , where, represents the original ultrasonic attenuation spectrum coefficient of the th frequency band, represents the frequency band attenuation value for particle size inversion after compensation; Regarding the sensitivity of the dynamic weight to the change in solid content, it can be derived: , where, represents the change rate of the weight of the th frequency band with respect to the solid content, is the sum of the weighted sensitivities of all frequency bands; Specifically, the mechanism introduced here uses logarithmic exponential mapping to embed the information of the deviation of the real-time solid content from the reference value into the frequency band weights, ensuring that as the solid content increases or decreases, each frequency band automatically obtains gain or attenuation compensation. The exponential mapping and normalization steps ensure that the weights are always positive and sum to 1, thus not changing the total spectral energy. The solid content sensitivity coefficient can be calibrated based on experiments. The response differences of different frequency bands to the solid content can be flexibly adjusted through this parameter. The derivative formula of the quadratic expansion provides the gradient direction for online adaptive learning, which can be further used for subsequent fine-tuning of the sensitivity coefficient or closed-loop optimization. The overall scheme is simple in calculation, strong in real-time performance, does not require frequent retraining, and only needs to read the current solid content to complete the compensation, providing a more stable ultrasonic signal input for particle size distribution inversion; The distribution decomposition module integrates a multi-modal distribution decomposition model, receives the first particle size distribution curve and the second particle size distribution curve, and outputs the decomposition result; The control decision module is connected to the distribution decomposition module and is configured to generate a stirring parameter adjustment instruction according to the peak ratio; The control decision module further includes: The emergency intervention unit automatically triggers the start of the seed addition device when it detects that the peak particle size of the newly generated particle distribution sub-curve exceeds 1.5 times the peak particle size of the seed particles; The discharge port of the seed addition device is provided with an ultrasonic atomizing head, and the atomization particle size is controlled within the range of 50-100 μm; The actuator interface module converts the stirring parameter adjustment instruction into a drive signal and sends it to the stirring drive device.

[0025] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for monitoring the particle size of materials used in the production of cobalt basic carbonate, characterized in that, including Step S1: Collect multi-source sensing data in the reactor in real time. The multi-source sensing data includes laser scattering intensity angular distribution signals, ultrasonic attenuation spectrum signals, and agitator shaft rotation speed signals; Step S2: Generate a first particle size distribution curve based on the laser scattering intensity angular distribution signals, and simultaneously generate a second particle size distribution curve according to the ultrasonic attenuation spectrum signals; Step S3: Input the first particle size distribution curve and the second particle size distribution curve into a multi-modal distribution decomposition model, and output the decomposed seed particle distribution sub-curve and the new particle distribution sub-curve; Step S4: Calculate the peak proportion of the seed particle distribution sub-curve. When the peak proportion is lower than a preset threshold, generate a stirring parameter adjustment instruction; Step S5: Send the stirring parameter adjustment instruction to the stirring drive device to dynamically adjust the inclination angle and rotation speed gradient of the stirring blades.

2. The method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to claim 1, characterized in that, The construction of the multi-modal distribution decomposition model includes: Perform Gaussian mixture modeling on the particle size distribution curves in the historical production data, and constrain the seed particle size distribution to conform to a preset nominal distribution characteristic; Dynamically evaluate the deviation degree between the real-time distribution curve and the historical distribution curve using KL divergence. When the deviation degree exceeds a set critical value, trigger the distribution curve splitting operation; Output the decomposition result including the seed particle distribution sub-curve and the new particle distribution sub-curve; The multi-modal distribution decomposition model includes a feature alignment layer, a cross-attention layer, and a physical constraint layer, where: The feature alignment layer performs timestamp synchronization and spatial registration on the laser scattering intensity angular distribution signals and the ultrasonic attenuation spectrum signals; The cross-attention layer calculates the correlation weight matrix of the two modal signals in the frequency-domain - spatial domain; The physical constraint layer enforces the satisfaction of the particle volume conservation equation.

3. The method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to claim 2, characterized in that, The distribution curve splitting operation includes the following steps: a) Detect the peak-valley ratio of the real-time distribution curve. When the valley value is higher than 20% of the peak heights on both sides, it is determined as a potential double peak; b) Perform second derivative zero-crossing detection on the suspected double peak region to locate the center position of the sub-peak; c) Dynamically increase the number of components of the Gaussian mixture model according to the sub-peak spacing.

4. The method for monitoring the particle size of materials used in the production of cobalt basic carbonate according to claim 3, characterized in that, In Step S3, the method of Gaussian mixture modeling of the historical particle size distribution and the seed nominal distribution constraint is: For the centered particle size variable Adopt Gaussian components to construct a mixture density function: ， Among them, represents the weight of the th Gaussian component, represents the total number of components of the Gaussian mixture model, represents the th component's variance, represents the th component's mean, represents the particle size variable after normalization and centering, represents the probability density of the mixture model at the particle size ; Introduce the KL divergence constraint on the component numbered and the nominal distribution of seeds so that this component is close to the preset features. Let the overall optimization objective be the sum of the log-likelihood and the constrained divergence: ， Among them, represents the total number of historical samples, represents the particle size value of the th sample, represents the constraint strength coefficient, represents the Kullback-Leibler divergence between two normal distributions, represents the seed nominal mean, represents the seed nominal variance, represents the component index assigned to the seed characteristics, and respectively represent the mean and variance of the th component; , where the and correspond to the variances of the compared distributions before and after, represents the squared deviation of the means of the two distributions, and the constant 1 is used to standardize the divergence baseline; Adopt improved EM iterative minimization , including: E-step: Calculate the posterior probability: ， M-step: ， Among them, represents the posterior probability that the -th sample belongs to the -th component, represents the density value of the -th Gaussian component at , represents the particle size measurement value of the -th historical sample after normalization and centering, represents the weight of the -th component after the M-step update in this round, represents the updated mean, represents the updated variance; For components, append gradient correction after M steps: ， Among them, represents the learning rate for regular gradient correction, and the iteration is repeated until , represents the mean value of the components belonging to the seed classification index after being updated in the current M-step and corrected by the gradient, represents the variance of the components belonging to the seed classification index after being updated in the current M-step and corrected by the gradient.

5. A method for monitoring the particle size of materials used in the production of cobalt basic carbonate as claimed in claim 1, characterized in that, The generation steps of the stirring parameter adjustment instruction include: Calculate the dispersion uniformity index according to the peak width of the seed particle distribution sub-curve; Establish the mapping relationship between the dispersion uniformity index and the inclination angle of the stirring blades to generate an inclination compensation amount; Calculate the adjustment step size of the rotation speed gradient based on the current stirring speed and the material viscosity.

6. The particle size monitoring method for the production of cobalt basic carbonate according to claim 5, characterized in that, In step S4, when calculating the peak ratio to determine the triggering of stirring adjustment according to the current reference threshold, the mixing weight of the seed component in the decomposition model is taken as the peak ratio , for the working condition fluctuations, a current reference threshold based on sliding window statistics is designed , and the process includes: The mixing weight represents the proportion of seed particles in the overall distribution, and is used to measure the seed abundance. The current reference threshold is adaptively adjusted through historical data, ， Among them, represents the mixing weight of the seed component, that is, the weight of the component in the Gaussian mixture model; Set the window statistical radius to Subsample and calculate the mean and standard deviation of the historical peak ratio: ， Among them, represents the length of the historical sampling window, represents the moment of the peak occupancy ratio, and are the mean and standard deviation within this window respectively; Set a safety factor based on the statistical results After that, the current reference threshold is: ， Among them, represents a safety factor used to adjust the triggering sensitivity, represents the current reference threshold; When the real-time peak occupancy ratio is lower than the current reference threshold immediately generate a stirring parameter adjustment instruction to restore the seed abundance.

7. A material particle size monitoring system for the production of cobalt basic carbonate, based on the method for monitoring the particle size of the material for the production of cobalt basic carbonate according to any one of claims 1 to 6, characterized in that, including: A multi-source sensing module, including a laser scattering probe array, an ultrasonic transceiver, and an agitator shaft encoder installed on the side wall of the reactor; A signal processing module, connected to the multi-source sensing module, configured to perform noise reduction processing on the original sensing data and generate a first particle size distribution curve and a second particle size distribution curve; A distribution decomposition module, integrating the multi-modal distribution decomposition model, receiving the first particle size distribution curve and the second particle size distribution curve and outputting the decomposition result; A control decision-making module, connected to the distribution decomposition module, is configured to generate a stirring parameter adjustment instruction according to the peak ratio; An actuator interface module converts the stirring parameter adjustment instruction into a driving signal and sends it to the stirring driving device.

8. A particle size monitoring system for the production of cobalt basic carbonate according to claim 7, characterized in that, The arrangement mode of the laser scattering probe array is as follows: Three detection planes are arranged in the axial height direction of the reaction kettle, and three probes are annularly distributed at 120° in each plane; The included angle between the incident optical path of each probe and the center line of the stirring shaft is 15° - 30°, and the probes of adjacent layers are arranged with a 40° circumferential dislocation.

9. A particle size monitoring system for the production of cobalt basic carbonate according to claim 7, characterized in that, The signal processing module includes: An adaptive filtering unit for eliminating the periodic mechanical vibration noise caused by the rotation of the stirring blades; A frequency domain compensation unit dynamically adjusts the frequency band weight coefficient of the ultrasonic attenuation spectrum according to the real-time solid content data; A data synchronization unit for data synchronization.

10. A particle size monitoring system for the production of cobalt basic carbonate as described in claim 7, characterized in that, The control decision-making module further includes: An emergency intervention unit automatically triggers the start of the seed addition device when it is detected that the peak particle size of the new particle distribution sub-curve exceeds 1.5 times the peak particle size of the seed particles; The discharge port of the seed addition device is provided with an ultrasonic atomizing head, and the atomization particle size is controlled in the range of 50 - 100 μm.

Images