Carbon nanomaterial slurry dispersion quality on-line monitoring and feedback device and method

Abstract

The present application belongs to the technical field of carbon nanomaterial preparation, and particularly relates to a kind of carbon nanomaterial slurry dispersion quality on-line monitoring and feedback device and method.The device obtains the absorbance, rheology and potential signal of slurry by multi-sensing cooperation, and generates dispersion degree cooperative index by coupling, thereby constructing real-time dispersion phase space to map dynamic dispersion state as transient phase point and process trajectory.Based on the preloaded golden stable basin model, the state of the state deviation is solved to the trend invasion vector, and then the process tuning instruction driving the state to converge to the optimal area is synthesized in reverse through the process rheology inversion kernel, and is compiled as an industrial control signal to realize closed-loop control.The present application solves the defects of existing technology such as monitoring lag, single dimension, inability to dynamically analyze and intelligent closed-loop control, and realizes real-time, accurate and adaptive optimization of slurry dispersion quality.

Description

Technical Field

[0001] This invention relates to the field of carbon nanomaterial preparation technology, and in particular to an online monitoring and feedback device and method for the dispersion quality of carbon nanomaterial slurries. Background Technology

[0002] In high-tech industries such as new energy, advanced composite materials, and electronic packaging, carbon nanomaterials (such as carbon nanotubes and graphene) have become key fundamental materials due to their excellent mechanical, electrical, and thermal properties. The full realization of these properties highly depends on the formation of a highly uniform and stable dispersion slurry in a solvent or resin matrix. Therefore, controlling the dispersion quality is a core process step in the preparation and application of related materials. In current industrial practice, the assessment of dispersion quality mainly relies on periodic offline sampling, which is then sent to a laboratory for testing using equipment such as laser particle size analyzers, scanning electron microscopes, or ultraviolet spectrophotometers. This approach has a significant time blind spot; from sampling and testing to obtaining analytical results, it often takes several hours or even longer, failing to capture instantaneous state changes and signs of instability during dynamic dispersion processes such as high-speed stirring and ultrasonic treatment. When the test results show that the quality is unqualified, a large amount of slurry has already been scrapped, resulting in serious raw material waste and production interruption.

[0003] To address the lag issues of offline detection, the industry has gradually introduced some online monitoring technologies, such as installing online viscometers or near-infrared probes. However, existing online solutions suffer from the following systemic drawbacks: First, they are limited to a single monitoring dimension, typically focusing only on individual parameters such as viscosity or concentration. They fail to simultaneously acquire multi-dimensional key information reflecting dispersion uniformity (e.g., absorbance), rheological properties (e.g., viscosity), and colloidal stability (e.g., zeta potential), leading to a one-sided assessment of complex dispersion states. More critically, existing technologies only monitor and alarm at the threshold level for isolated parameters, lacking the ability to deeply couple multi-source heterogeneous data and construct a comprehensive state model that can reflect the dynamic evolution of the dispersion process in real time. Because they cannot form a quantitative characterization and trend calculation of the process dynamics, existing systems cannot automatically derive precise control strategies from real-time state models, nor can they translate these strategies into closed-loop control commands that can drive production line actuators. This results in the production process still heavily relying on manual experience for delayed and crude intervention, failing to achieve real-time, adaptive optimization control of slurry dispersion quality.

[0004] Therefore, this invention proposes an online monitoring and feedback device and method for the dispersion quality of carbon nanomaterial slurry. Summary of the Invention

[0005] This invention provides an online monitoring and feedback device and method for the dispersion quality of carbon nanomaterial slurries. By integrating multiple sensors, a real-time dispersion phase space and process trajectory are constructed. Based on the model, the situational intrusion vector is calculated, and control commands are synthesized in reverse to achieve closed-loop control. This solves the core defects of existing technologies, such as one-sided monitoring, lack of dynamic models and intelligent closed-loop control, and realizes online and adaptive optimization control of slurry dispersion quality.

[0006] This invention provides an online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurries, comprising: A multi-parameter flow field probe array is used to simultaneously acquire the absorbance fingerprint, rheological response spectrum and interface potential signal of the slurry; The dispersed phase space evolution modeling module is used to couple absorbance fingerprints with rheological response spectra to generate a dispersion co-index, and construct a real-time dispersed phase space with the dispersion co-index as the horizontal axis and the interface potential signal as the vertical axis, mapping synchronous data into transient phase points in the real-time dispersed phase space; The process trajectory entropy state perception module is used to connect the transient phase points of the time series to form the process evolution trajectory, and calculate the current entropy state of the process evolution trajectory based on the preloaded golden stable basin model, and solve the situation invasion vector of the invasion into the golden stable basin. The inverse mapping synthesis engine is controlled to receive the situational intrusion vector and synthesizes the process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin through the embedded process rheological inversion kernel. The industrial protocol injection and entropy lake module is used to compile the process rheological tuning wave into a control pulse sequence that can be recognized by the corresponding industrial equipment and inject it into the production line. At the same time, it deposits the real-time dispersed phase space, process evolution trajectory and process rheological tuning wave data of the whole cycle into the process entropy lake.

[0007] Preferably, the multi-parameter flow field probe array includes: Photon acquisition probes are used to acquire absorbance fingerprints that characterize the spatial distribution of nanoparticle clusters online. A shear stress acquisition probe is used to acquire rheological response spectra characterizing the internal structural strength of slurries online. The double-layer acquisition probe is used to acquire interfacial potential signals that characterize the interaction potential between particles online.

[0008] Preferably, the dispersed-state phase space evolution modeling module includes: The synergy index synthesis submodule is used to divide the real-time acquired absorbance fingerprint by the characteristic viscosity value in the rheological response spectrum to calculate the synergy index of dispersion, which characterizes the synergistic relationship between dispersion uniformity and flow resistance. The phase space mapping submodule is used to construct a two-dimensional real-time dispersed phase space with the dispersion co-exponent as the horizontal axis and the interface potential signal as the vertical axis, and to map the dispersion co-exponent and interface potential signal at each moment into a transient phase point in the real-time dispersed phase space.

[0009] Preferably, the process trajectory entropy state sensing module includes: The trajectory chain generation submodule is used to connect transient phase points that appear in the real-time dispersed phase space in a time sequence to generate process evolution trajectories that characterize the dynamic evolution of the dispersion process. The entropy state solution submodule contains a gold stable basin model; the gold stable basin model defines a closed, optimal gold stable basin region in the real-time dispersed phase space. The entropy state solution submodule is used to calculate the vector from the latest transient phase point to the target point in the core region of the golden stable basin, which is used as the situational intrusion vector; where the direction of the situational intrusion vector is defined as the intrusion direction and the magnitude is defined as the intrusion intensity. Simultaneously, the recent rate of curvature change of the evolution trajectory is calculated as the trajectory entropy value; the trajectory entropy value is used to characterize the trend of disorder in the process.

[0010] Preferably, the stable gold basin model is a dynamically deformable basin generated by combining a Gaussian mixture model and a convex hull algorithm based on multiple sets of historical best production batch data clusters. The boundary morphology, area, and core target location of the dynamically deformable basin can adaptively deform based on the input carbon nanomaterial crystal structure code and slurry formulation hash value.

[0011] Preferably, the controlled inverse mapping synthesis engine includes: The inversion solution submodule contains a process rheology inversion kernel. The process rheology inversion kernel takes the situational invasion vector and trajectory entropy as inputs. By solving an optimization problem constrained by the slurry rheological constitutive equation and DLVO theory, it inversely maps out the precise adjustment amounts of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field. The tuned wave compilation submodule is used to compile the precise adjustments of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field into a set of process rheological tuned waves with specific amplitude, frequency, and phase relationships in the time domain, according to the equipment response characteristics and process safety boundaries.

[0012] Preferably, the process rheological inversion kernel is a hybrid architecture model. The core is a structure-rheological relationship predictor built by a graph neural network, and the periphery is coupled with an online optimizer based on a model predictive control framework to solve for the optimal adjustment amount under multiple constraints.

[0013] Preferably, the industrial protocol injection and entropy lake module includes: The protocol injection gateway submodule supports the decomposition and adaptation of process rheological tuning waves into OPCUA information models, EtherCAT drive messages, and time-sensitive network scheduling commands, enabling lossless injection with heterogeneous industrial control systems. The process entropy lake submodule is used to archive and store each frame of real-time dispersed phase space snapshots, process evolution trajectory fragments, situational intrusion vectors, and process rheological tuning waves in an immutable chain structure with nanosecond-level timestamps, forming a process entropy lake with complete causal relationships.

[0014] Preferred options also include: The panoramic perception and early warning cockpit includes a phase space holographic projection unit and an entropy change early warning unit, wherein: Phase space holographic projection unit is used to present a three-dimensional dynamic hologram of real-time dispersed phase space, golden stable basin and process evolution trajectory in three dimensions using augmented reality technology; The entropy change precursor warning unit is used to trigger an entropy increase precursor warning when the trajectory entropy value monotonically increases beyond a set threshold within a continuous period, or when the intrusion direction of the situation intrusion vector continuously moves away from the target point in the centrifugal domain.

[0015] This invention provides a method for online monitoring and feedback of the dispersion quality of carbon nanomaterial slurries, including: Simultaneously acquire the absorbance fingerprint, rheological response spectrum, and interfacial potential signal of the slurry; The absorbance fingerprint is coupled with the rheological response spectrum to generate the dispersion co-index. The dispersion co-index is used as the horizontal axis and the interface potential signal is used as the vertical axis to construct a real-time dispersed phase space. The synchronous data is mapped to the transient phase points in the real-time dispersed phase space. The transient phase points of the time series are connected to form the process evolution trajectory. Based on the pre-loaded gold stable basin model, the current entropy state of the process evolution trajectory is calculated, and the situation invasion vector of the invasion into the gold stable basin is solved. Receive the situational intrusion vector and, through the embedded process rheological inversion kernel, inversely synthesize a process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin. The process rheometry tuned wave is compiled into a control pulse sequence that can be recognized by the corresponding industrial equipment and injected into the production line. At the same time, the real-time dispersed phase space, process evolution trajectory and process rheometry tuned wave data of the whole cycle are deposited into the process entropy lake.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: By integrating a multi-parameter flow field probe array, it overcomes the shortcomings of existing online monitoring technologies, such as their single dimension and inability to comprehensively and synchronously perceive the dispersion uniformity, rheological properties, and colloidal stability of slurries; by using a dispersed phase space evolution modeling module and a process trajectory entropy state perception module, it solves the problems of existing technologies that can only perform isolated parameter threshold alarms and lack deep fusion of multi-source heterogeneous data to construct dynamic process evolution models; finally, by leveraging a regulatory inverse mapping synthesis engine and an industrial protocol injection and entropy lake module, it achieves a seamless closed loop from real-time state perception and intelligent trend analysis to precise regulation command generation and execution, completely changing the traditional production mode of only monitoring without control or relying on delayed manual intervention, and achieving real-time, adaptive, model-driven closed-loop optimization control of carbon nanomaterial slurry dispersion quality.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a diagram showing the overall architecture and data flow of the online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry in this embodiment of the invention. Figure 2 This is a diagram illustrating the core calculation and control path of the online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry in this embodiment of the invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 As shown, this invention provides an embodiment of an online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurries, comprising: A multi-parameter flow field probe array is used to simultaneously acquire the absorbance fingerprint, rheological response spectrum and interface potential signal of the slurry; The dispersed phase space evolution modeling module is used to couple absorbance fingerprints with rheological response spectra to generate a dispersion co-index, and construct a real-time dispersed phase space with the dispersion co-index as the horizontal axis and the interface potential signal as the vertical axis, mapping synchronous data into transient phase points in the real-time dispersed phase space; The process trajectory entropy state perception module is used to connect the transient phase points of the time series to form the process evolution trajectory, and calculate the current entropy state of the process evolution trajectory based on the preloaded golden stable basin model, and solve the situation invasion vector of the invasion into the golden stable basin. The inverse mapping synthesis engine is controlled to receive the situational intrusion vector and synthesizes the process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin through the embedded process rheological inversion kernel. The industrial protocol injection and entropy lake module is used to compile the process rheological tuning wave into a control pulse sequence that can be recognized by the corresponding industrial equipment and inject it into the production line. At the same time, it deposits the real-time dispersed phase space, process evolution trajectory and process rheological tuning wave data of the whole cycle into the process entropy lake.

[0022] In this embodiment, the absorbance fingerprint of the slurry is the absorbance time sequence and peak distribution at characteristic wavelengths within the wavelength range of 300-800 nm, and the rheological response spectrum is 1-100 s. -1 The viscosity change curve under shear rate, along with the storage modulus and loss modulus data, and the interface potential signal, which is the Zeta potential detection value of the nanoparticles (accuracy ±1mV), are all captured synchronously through a multi-parameter flow field probe array to comprehensively reflect the dispersion uniformity, internal structural strength, and interparticle interaction potential.

[0023] In this embodiment, the absorbance fingerprint is coupled with the rheological response spectrum to generate a dispersion synergy index. Specifically, the ratio of the characteristic parameters of the absorbance fingerprint to the steady-state viscosity value corresponding to the shear rate of 50 s⁻¹ in the rheological response spectrum is taken, with a value ranging from 0.8 to 1.2, simultaneously characterizing the fit between dispersion uniformity and flow resistance. The characteristic parameters of the absorbance fingerprint specifically refer to the absorbance peak value corresponding to the characteristic wavelength within the 300-800 nm wavelength range that most sensitively reflects the spatial distribution of the nanoparticle group. This characteristic wavelength is determined by analyzing historical best production batch data clusters, and is typically the wavelength corresponding to the characteristic absorption peak of carbon nanomaterials (such as the 550-650 nm range for carbon nanotubes and the 400-500 nm range for carbon black). The magnitude and stability of its absorbance peak value are directly related to the particle dispersion uniformity and are the core basis for calculating the dispersion synergy index.

[0024] In this embodiment, synchronous data refers to the combined data of absorbance fingerprint, rheological response spectrum and interface potential signal collected at the same time, which is the basis for constructing real-time dispersed phase space and generating transient phase points.

[0025] In this embodiment, the pre-loaded gold stable basin model is a dynamic region generated by Gaussian mixture model and convex hull algorithm based on historical best production batch data. The boundary has a dispersion synergy index of 0.9-1.1 and an absolute value of interface potential of 30-50mV, and can adaptively deform according to material type and formulation.

[0026] In this embodiment, the current entropy state of the process evolution trajectory is a value calculated based on the recent rate of change of curvature of the trajectory, which is used to characterize the disorder trend of the dispersion process. The threshold is 0.5. If the value exceeds the threshold and increases for three consecutive cycles, it is judged as abnormal.

[0027] In this embodiment, the situational invasion vector of the invading gold stable basin is the vector from the latest transient phase point to the target point in the core region. The direction represents the deviation direction, the magnitude represents the deviation intensity, and the quantification of the difference between the dispersed state and the optimal interval.

[0028] In this embodiment, the embedded process rheological inversion kernel is a hybrid architecture that integrates graph neural networks and model predictive control. Taking the situational intrusion vector and entropy value as input, it derives the optimal adjustment amount under the constraints of the rheological constitutive equation and DLVO theory. Its core design and execution logic are clear and reproducible. The specific implementation process and key details are as follows: From an architectural perspective, this hybrid architecture consists of two layers: a graph neural network structure-rheological relationship prediction layer and a model prediction control optimization layer. These two layers have a clear data interaction logic: the output of the graph neural network structure-rheological relationship prediction layer directly serves as the input constraint for the model prediction control optimization layer, ensuring that the two layers collaboratively derive the optimal adjustment amount. Specifically, the graph neural network structure-rheological relationship prediction layer employs a graph convolutional network architecture, abstracting the dispersion system of carbon nanomaterial slurry into graph structure data—using nanoparticles (carbon nanotubes / carbon black), dispersant molecules, NMP solvent micro-elements as nodes of the graph, and particle-dispersant adsorption, interparticle repulsion / attraction, and solvent-particle wetting as edges. Simultaneously, nodes are assigned characteristic attributes (such as the number of active sites on the particle surface, dispersant molecule chain length, and solvent viscosity), and edges are assigned weights (such as adsorption strength and force magnitude). This layer learns dispersion through a message passing mechanism of neighborhood feature aggregation and node state updates. System characteristics: First, the mean aggregation function is used to collect feature information in the neighborhood of each node (e.g., the chain length features of the dispersant nodes around the particle node and the viscosity features of the solvent node). Then, through linear transformation and activation function, the neighborhood features after aggregation are fused with the node's own features to update the node's hidden state. Finally, the dispersion state-rheological parameter correlation model is output. Based on the input situational intrusion vector (reflecting the degree of deviation of the dispersion state) and trajectory entropy value (reflecting the disorder of the dispersion process), the rheological parameter change trend of the slurry under different process adjustments (e.g., the change curve of the characteristic viscosity of the slurry after adjusting the speed of the disperser) can be predicted.

[0029] The model predictive control optimization layer is based on the dispersed state-rheological parameter correlation model output by the graph neural network layer. It combines the preset slurry rheological constitutive equation and DLVO theory as constraints, and solves for the optimal adjustment amount through a rolling logic of prediction-optimization-feedback. The rheological constitutive equation describes the correlation between slurry shear stress, shear rate, and viscosity. Its key parameters are determined through offline experiments: three sets of shear stress data at different shear rates are used, and fitting methods are employed to obtain specific values ​​for yield stress, consistency coefficient, and flow behavior index (e.g., the yield stress of a carbon nanotube-carbon black mixed slurry (mass ratio 1:1) is 5 Pa, the consistency coefficient is 20 Pa·sⁿ, and the flow behavior index is 0.8). The DLVO theory describes the interparticle interaction potential energy. Based on the interparticle distance, the sum of van der Waals attraction potential energy and double-layer repulsion potential energy is calculated to ensure that the interparticle repulsion potential energy reaches the set standard after adjustment, avoiding agglomeration. Specifically, this layer first predicts the slurry rheological parameters (characteristic viscosity, shear stress) and dispersion state (interfacial potential, dispersant addition index) corresponding to different process adjustment schemes (such as disperser speed ±10%, dispersant addition ±0.2%) within the next 5 sampling periods (10 seconds each). Then, it defines an optimization objective function: with the dispersant addition index returning to 1.0 ± 0.05 and the absolute value of the interfacial potential maintained at ≥30mV, and the rheological parameter fluctuation range ≤10%, a weighted combination function is constructed (the weight allocation is: dispersion state index accounts for 0.6, and rheological parameter index accounts for 0.4). Finally, within the process safety boundary (disperser speed not exceeding 5000 rpm, dispersant addition ±0.2%), the layer further optimizes the slurry rheological parameters (characteristic viscosity, shear stress) and dispersion state (interfacial potential, dispersant addition ±0.2%). Under the constraints of adding no more than 3% of the total solid content and reactor temperature fluctuation of ±2℃, a quadratic programming algorithm is used to solve the optimization problem within the finite time domain (5 sampling periods). The output includes three types of optimal adjustment values: energy input spectrum adjustment of the disperser (e.g., speed increase of 15%), chemical potential input gradient adjustment of the dispersant (e.g., dispersant addition rate increase of 0.02 mol / (L・min)), and thermodynamic field adjustment of the reactor (e.g., temperature increase of 1℃). The calculation results of the adjustment values ​​can be verified by historical best production batch data (taking 5 historical best batches, inputting the same situational invasion vector and entropy value, the average time for the adjustment value output by this architecture to make the dispersion state return to the stable range is 40 seconds, which is 60% more efficient than traditional manual adjustment).

[0030] Furthermore, the key parameters of this hybrid architecture (such as the initial values ​​of the weight matrix of the graph neural network, the sampling period of the model predictive control, and the prediction time domain) were all determined through offline training and validation: the weight matrix of the graph neural network was initialized normally, and the bias term was initialized to 0. The optimizer was trained on a dispersed dataset containing 200 different material formulations (carbon nanotube / carbon black mass ratio of 1:9 to 9:1, dispersant type BYK-163 / Solsperse24000) to ensure that the prediction error of the dispersed state-rheological parameters was ≤5%; the sampling period of the model predictive control was set to 10 seconds (to match the response speed of the slurry rheological parameters), and the prediction time domain was set to 5 periods (to balance prediction accuracy and computational complexity) to ensure that the architecture can stably derive the optimal adjustment amount under different dispersed scenarios.

[0031] In this embodiment, a process rheological tuning wave that can drive the transient phase point from the current entropy state to the target point in the core region of the gold-stabilized basin is synthesized in reverse through an embedded process rheological inversion kernel. That is, the adjustment parameters of the disperser energy, dispersant addition amount, and reactor temperature are calculated in reverse according to the degree of deviation and compiled into a control signal with specific amplitude, frequency and phase. Details are as follows: Step 1: Quantitative classification of deviation degree (for parameter adjustment calibration): First, by using the magnitude (intrusion intensity) of the intrusion vector and the trajectory entropy value, the degree of deviation from the dispersed state is divided into three levels, and the adjustment intensity benchmarks corresponding to different levels are clarified: Slight deviation: Intrusion intensity ≤ 0.2 and trajectory entropy value ≤ 0.3, adjustment intensity 5%-10% of the base value; Moderate deviation: Intrusion intensity 0.2-0.5 or trajectory entropy value 0.3-0.5, adjust the intensity by 10%-20% of the base value; Severe deviation: Invasion intensity > 0.5 or trajectory entropy value > 0.5, adjustment intensity is 20%-30% of the base value; the base value is determined by the historical best batch data (e.g., the basic speed of the disperser is 3000 rpm, the basic amount of dispersant added is 1% of the total solid content, and the basic temperature of the reactor is 25℃).

[0032] Step 2: Reverse mapping adjustment parameters (corresponding to deviation and process parameters): Based on the decentralized state-process parameter correlation model of the process rheology inversion kernel, the specific adjustment values ​​of the three types of core parameters are derived in reverse according to the deviation level and deviation direction: Dispersant energy adjustment parameters: The greater the invasion intensity, the higher the shear energy needs to be increased. The adjustment parameter is the speed change value (e.g., for slight deviation, increase the speed by 300 rpm; for moderate deviation, increase the speed by 600 rpm). If the invasion direction indicates insufficient dispersion uniformity, the high-speed shearing time should be extended simultaneously (e.g., for moderate deviation, extend by 2 minutes). Dispersant addition adjustment parameters: When the interfacial potential signal is low (particles are prone to agglomeration), increase the amount of dispersant added according to the degree of deviation (e.g., slight deviation +0.1% total solids content, severe deviation +0.3%). The adjustment parameters are addition rate + total amount added (e.g., addition rate 0.02mol / (L・min), total amount 0.2% total solids content). Reactor temperature adjustment parameters: When the slurry viscosity is too high (high flow resistance), moderately increase the temperature to reduce the viscosity (e.g., increase the temperature by 1℃ for every 0.1 coherent index unit deviation). The adjustment parameter is the temperature change value (range ±2℃) to avoid sudden temperature changes that could damage the dispersion state.

[0033] Step 3: Compile into a control signal with specific amplitude, frequency, and phase: The above adjustment parameters are converted into time-domain control signals according to the mapping rule of parameter type → signal characteristics to ensure that the equipment can respond accurately: Amplitude: The magnitude of the corresponding adjustment parameter (e.g., for speed adjustment +600 rpm, the amplitude is 600; for temperature adjustment +1℃, the amplitude is 1). The larger the amplitude, the stronger the adjustment. Frequency: Corresponds to the execution rhythm of the adjustment actions (the dispersant speed adjustment frequency is 10Hz to ensure a smooth speed increase; the dispersant addition frequency is 5Hz to match the metering pump response speed; the temperature adjustment frequency is 1Hz to avoid drastic fluctuations). Phase: The timing relationship of multi-parameter coordinated adjustment (e.g., start the dispersant speed adjustment first (phase 0°), start the dispersant addition 0.5 seconds later (phase 90°), start the temperature adjustment 1 second later (phase 180°)) to avoid adjustment conflict.

[0034] Step 4: Signal Timing Coordination and Format Adaptation The control signals corresponding to the three types of parameters are integrated according to their phase relationship, and then adapted to the communication protocol of industrial equipment (such as OPCUA, EtherCAT) to generate a control pulse sequence: Time-sensitive coordination: Through time-sensitive network scheduling commands, ensure that speed increase, dispersant addition, and temperature adjustment are executed synchronously according to preset phases (e.g., dispersant is added only after the speed reaches the target value). Format adaptation: The amplitude, frequency, and phase information are encapsulated into a signal format that the device can recognize (such as the data stream of EtherCAT drive messages or the standardized data object of OPCUA). After being injected into the production line equipment, the equipment performs adjustment actions according to the signal characteristics, and finally pulls the dispersed state back to the golden stable basin.

[0035] The entire process requires no human intervention. All mapping rules and signal parameters are determined through training with historical data. Those skilled in the art can reproduce the complete calculation and compilation process based on the above logic to ensure that the control signal accurately matches the deviation state and achieves rapid regression of the dispersion quality.

[0036] In this embodiment, the target point in the core region of the gold stable basin is the optimal dispersion state point in the model, corresponding to a dispersion synergy index of 1.0 and an absolute value of interface potential of 40mV, which is the target benchmark for regulation.

[0037] In this embodiment, the process rheological tuning wave is a time-domain signal that carries adjustment parameters. It integrates the control logic of the disperser, dispersant, and reactor, and adapts to the equipment response characteristics and process safety boundaries.

[0038] In this embodiment, the process rheometry tuning wave is compiled into a control pulse sequence that can be recognized by the corresponding industrial equipment and injected into the production line. That is, it is adapted to industrial protocols such as OPCUA and EtherCAT, and converted into 10-100Hz control commands to drive the production line equipment to make precise adjustments.

[0039] In this embodiment, the real-time dispersed phase space throughout the entire cycle is a two-dimensional dynamic space that runs through the entire process of slurry dispersion. It records the transient phase points and process evolution trajectory at each moment, and fully presents the dynamic changes of the dispersion state.

[0040] To simultaneously and accurately acquire key original signals characterizing the dispersion state of slurry from three fundamental levels—optical, rheological, and electrochemical—a multi-parameter flow field probe array is proposed, including: Photon acquisition probes are used to acquire absorbance fingerprints that characterize the spatial distribution of nanoparticle clusters online. A shear stress acquisition probe is used to acquire rheological response spectra characterizing the internal structural strength of slurries online. The double-layer acquisition probe is used to acquire interfacial potential signals that characterize the interaction potential between particles online.

[0041] In this embodiment, the spatial distribution state of the nanoparticle group refers to the degree of uniform dispersion of carbon nanoparticles in the slurry, including whether the particles agglomerate, the size of the agglomerates and the uniformity of the spatial distribution, which is the core indicator for evaluating the dispersion quality.

[0042] In this embodiment, the absorbance fingerprint characterizing the spatial distribution of the nanoparticle group is obtained by continuously collecting absorbance data in the wavelength range of 300-800nm ​​using a photon trapping probe, forming a signal set that includes the temporal variation pattern and the distribution of characteristic wavelength peaks. The more uniform the particle distribution, the more stable the absorbance signal.

[0043] In this embodiment, the photon trapping probe is an online detection element that integrates ultraviolet-visible spectrophotometry detection. It can be directly inserted into the slurry flow field and capture absorbance signals in real time without sampling, ensuring that the detection data is consistent with the actual dispersion state.

[0044] In this embodiment, the shear stress acquisition probe is an online sensor with rheological detection capabilities, capable of applying stress for 1-100 seconds. -1 The shear rate is measured, and the viscosity, storage modulus and loss modulus data of the slurry are collected simultaneously to accurately capture the internal structural response of the slurry.

[0045] In this embodiment, the rheological response spectrum characterizing the internal structural strength of the slurry is obtained online by continuously applying gradient shear force to the slurry during the dispersion process using a shear stress acquisition probe, recording the changes in the rheological parameters of the slurry at different shear rates, and forming a continuous spectrum reflecting the internal structural strength. The higher the structural strength, the more stable the rheological parameters.

[0046] In this embodiment, the double-layer acquisition probe is an online detection element based on the principle of electrophoresis, which can capture the potential changes of the double layer of nanoparticles in the slurry in real time and directly output the Zeta potential signal with a detection accuracy of ±1mV.

[0047] In this embodiment, the interface potential signal characterizing the interaction potential between particles is obtained online by detecting the potential difference of the double layer on the surface of nanoparticles in real time through a double layer acquisition probe. The larger the absolute value of this signal, the stronger the repulsive force between particles and the better the dispersion stability; conversely, the smaller the signal, the easier it is to agglomerate.

[0048] like Figure 2 As shown, to fuse heterogeneous physical signals and transform them into a quantitative index that comprehensively reflects the synergistic relationship between dispersion uniformity and flow resistance, and to construct an intuitive and information-rich state representation space based on this index, a dispersed state phase space evolution modeling module is proposed, including: The synergy index synthesis submodule is used to divide the real-time acquired absorbance fingerprint by the characteristic viscosity value in the rheological response spectrum to calculate the synergy index of dispersion, which characterizes the synergistic relationship between dispersion uniformity and flow resistance. The phase space mapping submodule is used to construct a two-dimensional real-time dispersed phase space with the dispersion co-exponent as the horizontal axis and the interface potential signal as the vertical axis, and to map the dispersion co-exponent and interface potential signal at each moment into a transient phase point in the real-time dispersed phase space.

[0049] In this embodiment, the characteristic viscosity value in the rheological response spectrum refers to the shear rate of 50 s in the rheological response spectrum. -1 The steady-state viscosity of the slurry is such that the shear rate is close to the normal operating conditions of industrial dispersion production, and can objectively reflect the actual flow resistance characteristics of the slurry.

[0050] In this embodiment, the absorbance fingerprint acquired in real time is divided by the characteristic viscosity value in the rheological response spectrum to calculate the dispersion synergy index, which characterizes the synergistic relationship between dispersion uniformity and flow resistance. Specifically, the absorbance peak of the characteristic wavelength in the absorbance fingerprint is taken as the numerator and the characteristic viscosity value is taken as the denominator. When the ratio result is in the range of 0.8-1.2, it indicates that the dispersion uniformity and flow resistance are well matched. The characteristic wavelength is the specific wavelength in the 300-800nm ​​ultraviolet-visible detection band of carbon nanomaterial slurry that most sensitively reflects the spatial dispersion uniformity of nanoparticle groups. This wavelength is determined based on the inherent optical absorption characteristics of the corresponding carbon nanomaterial. The characteristic wavelength of carbon nanotube slurry is selected in the range of 550-650nm, the characteristic wavelength of carbon black slurry is selected in the range of 400-500nm, and the wavelength corresponding to the peak value with the strongest absorbance response in the mixed system is selected for carbon nanotube and carbon black mixed slurry. The characteristic wavelength can be determined by offline calibration. The optimal stable dispersion slurry with the same formula is scanned in the whole band. The wavelength with the largest absorbance peak value and the best correlation with the particle aggregation state is determined as the characteristic wavelength used to calculate the dispersion synergy index of this type of slurry. The corresponding absorbance peak value can directly characterize the dispersion uniformity of nanoparticles in the slurry.

[0051] In this embodiment, dispersion uniformity refers to whether carbon nanoparticles are evenly distributed in the slurry without obvious agglomerates. The stability of absorbance fingerprint is its core characterization criterion, and the smaller the absorbance fluctuation, the better the dispersion uniformity.

[0052] In this embodiment, flow resistance refers to the obstruction encountered by the slurry during pipeline transportation, stirring and dispersion, etc., and is mainly determined by the viscosity of the slurry. The larger the characteristic viscosity value, the greater the flow resistance.

[0053] In this embodiment, a two-dimensional real-time dispersed phase space is constructed with the dispersion synergy index as the horizontal axis and the interface potential signal as the vertical axis. The dispersion synergy index and the interface potential signal at each moment are mapped to a transient phase point in the real-time dispersed phase space. The horizontal axis reflects the synergy level of dispersion uniformity and flow resistance, and the vertical axis reflects the dispersion stability between particles. The combination of the two can comprehensively locate the current dispersion state.

[0054] like Figure 2 As shown, to not only observe the instantaneous state but also track the dynamic evolution trend of the process and quantify the deviation and disorder trend of the current state from the optimal stable region, a process trajectory entropy state sensing module is proposed, including: The trajectory chain generation submodule is used to connect transient phase points that appear in the real-time dispersed phase space in a time sequence to generate process evolution trajectories that characterize the dynamic evolution of the dispersion process. The entropy state solution submodule contains a gold stable basin model; the gold stable basin model defines a closed, optimal gold stable basin region in the real-time dispersed phase space. The entropy state solution submodule is used to calculate the vector from the latest transient phase point to the target point in the core region of the golden stable basin, which is used as the situational intrusion vector; where the direction of the situational intrusion vector is defined as the intrusion direction and the magnitude is defined as the intrusion intensity. Simultaneously, the recent rate of curvature change of the evolution trajectory is calculated as the trajectory entropy value; the trajectory entropy value is used to characterize the trend of disorder in the process.

[0055] In this embodiment, the gold stable basin model defines a closed and optimal gold stable basin region in the real-time dispersed phase space. The boundary of this region is determined by training with historical best production batch data, corresponding to a dispersion coordination index of 0.9-1.1 and an absolute value of interface potential signal of 30-50mV. The dispersion state within this region can ensure uniform particle dispersion and meet the requirements of slurry flow characteristics, making it an ideal dispersion range for industrial production.

[0056] In this embodiment, the vector from the latest transient phase point to the target point in the core region of the golden stable basin is calculated as the situational intrusion vector. The direction of the situational intrusion vector is defined as the intrusion direction, that is, the orientation of the latest transient phase point from the target point in the core region, and the magnitude is defined as the intrusion intensity, that is, the straight-line distance between the two in the phase space. The direction and magnitude of the vector together quantify the specific situation of the current dispersion state deviating from the optimal range, providing a precise basis for subsequent regulation.

[0057] In this embodiment, the recent rate of curvature change of the process evolution trajectory is used as the trajectory entropy value. The trajectory entropy value is used to characterize the disorder trend of the process. The larger the rate of curvature change, the more violent the fluctuation of the dispersion state. The higher the trajectory entropy value, the more obvious the disorder trend. When the entropy value exceeds 0.5 and increases monotonically for three consecutive monitoring cycles, it indicates that the dispersion process may deviate from the stable track and timely regulation needs to be initiated.

[0058] To enable the criteria for determining the optimal stable region to adapt to different materials and formulations and reflect the dynamic and personalized nature of the process window, a golden stable basin model is proposed. This model is a dynamic deformation basin generated by combining a Gaussian mixture model and a convex hull algorithm based on multiple sets of historical optimal production batch data clusters. The boundary morphology, area, and core target location of the dynamically deformable basin can adaptively deform based on the input carbon nanomaterial crystal structure code and slurry formulation hash value.

[0059] In this embodiment, the historical best production batch data cluster refers to a set of past production batch data that meets the dispersion quality requirements (such as dispersion uniformity, good stability, and high production efficiency). It includes the absorbance fingerprint, rheological response spectrum, interface potential signal and corresponding production process parameters of each batch, and is the basic data source for constructing the gold stable basin model.

[0060] In this embodiment, the gold stable basin model is a dynamically deformable basin generated based on multiple sets of historical best production batch data clusters through a combination of Gaussian mixture model and convex hull algorithm. The Gaussian mixture model is used to mine the distribution patterns of optimal dispersion states in the data clusters, and the convex hull algorithm is used to define the boundaries of the optimal distribution area. The combination of the two forms a closed basin that covers all optimal dispersion states. The core logic is as follows: Gaussian mixture model for discovering optimal distribution patterns: First, the historical best production batch data cluster (including the dispersion coordination index and interface potential signal of each batch) is input into the Gaussian mixture model. The model divides the data cluster into multiple Gaussian distribution sub-clusters through iterative calculation (using the EM algorithm). Each sub-cluster corresponds to a probability distribution of the optimal dispersion state. The model outputs the center coordinates (core value of the optimal state), variance (state fluctuation range), and weight (sub-cluster proportion) of each sub-cluster. This accurately describes the distribution characteristics of the optimal dispersion state in phase space and filters out all valid data points that meet the dispersion quality standards.

[0061] The convex hull algorithm defines the closed boundary: Based on all the effective data points selected by the Gaussian mixture model, the convex hull algorithm (such as Graham's scan method) is used to fit the contours of these points: After sorting the data points by their polar angles, the outermost effective data points are connected in sequence to form a convex polygon. This convex polygon is the initial closed boundary of the golden stable basin, ensuring that the boundary can completely enclose all the optimal dispersed state data points without omission.

[0062] Model co-processing optimizes basin boundaries: The sub-cluster variance output by the Gaussian mixture model is used as weights to fine-tune the initial boundary generated by the convex hull algorithm. For sub-cluster regions with high probability density, the boundary is appropriately shrunk to improve basin accuracy. For edge regions with low probability density but still optimal, they are retained within the basin. This ultimately forms a closed, stable golden basin that covers all optimal dispersion states and eliminates invalid and redundant regions. Furthermore, the basin can dynamically deform by simultaneously adjusting the sub-cluster partitioning rules of the Gaussian mixture model and the fitting weights of the convex hull algorithm based on the carbon nanomaterial crystal structure encoding and slurry formulation hash value.

[0063] In this embodiment, the boundary morphology, area, and core target location of the dynamically deformable basin can be adaptively deformed based on the input carbon nanomaterial crystal structure code and slurry formulation hash value, ensuring that the model can accurately match the corresponding optimal dispersion interval under different materials and formulations, thereby improving the adaptability of the device.

[0064] In this embodiment, the carbon nanomaterial crystal structure encoding is a digital identifier that characterizes the core properties of carbon nanomaterials. It includes key parameters such as the aspect ratio of carbon nanotubes, the number of graphene layers, the specific surface area of ​​carbon black, and the degree of crystal defects, and is used to distinguish the dispersion characteristics of different types of carbon nanomaterials.

[0065] In this embodiment, the slurry formulation hash value is a unique identifier generated based on the core formulation parameters of the slurry, covering information such as the concentration of carbon nanomaterials, the type and proportion of dispersant, and the type and ratio of solvent, which is used to accurately match the optimal dispersion state boundary of the corresponding formulation.

[0066] like Figure 2 As shown, to inversely and accurately map the quantized signal representing state deviation into executable, multi-parameter coordinated process adjustment quantities, a controllable inverse mapping synthesis engine is proposed, including: The inversion solution submodule contains a process rheology inversion kernel. The process rheology inversion kernel takes the situational invasion vector and trajectory entropy as inputs. By solving an optimization problem constrained by the slurry rheological constitutive equation and DLVO theory, it inversely maps out the precise adjustment amounts of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field. The tuned wave compilation submodule is used to compile the precise adjustments of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field into a set of process rheological tuned waves with specific amplitude, frequency, and phase relationships in the time domain, according to the equipment response characteristics and process safety boundaries.

[0067] In this embodiment, a process rheology inversion kernel is embedded. This kernel takes the state intrusion vector and trajectory entropy as inputs, and solves an optimization problem constrained by the slurry rheological constitutive equation and DLVO theory. It then inversely maps the precise adjustments to the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field. The core principle is to combine the degree of deviation and disorder trend of the dispersion state, under the constraints of the slurry flow characteristics and the principle of interparticle interaction, to derive the key process parameter adjustment schemes that allow the dispersion state to return to the optimal range. The entire process uses the state intrusion vector and trajectory entropy as core inputs, and the slurry rheological constitutive equation and DLVO theory as hard constraints, completing the precise derivation in four steps: The first step is to clarify the specific degree and direction of deviation of the dispersion state through the situational intrusion vector. The larger the magnitude of the intrusion vector, the more serious the deviation of the current dispersion state from the target point. The direction of the intrusion vector directly points to the specific problem type, such as the imbalance of dispersion uniformity, abnormal flow resistance, or insufficient particle stability. At the same time, the trajectory entropy value is used to judge the chaotic trend of the dispersion process. The higher the trajectory entropy value, the more violent the fluctuation of the dispersion state and the greater the risk of subsequent deterioration. The combination of the two can quantify the intensity level and core adjustment direction that need to be adjusted. The second step is to use the slurry rheological constitutive equation as the constraint basis for the slurry flow characteristics. This equation clarifies the inherent correlation between slurry viscosity, shear stress, and flow resistance. When deriving and adjusting the parameters, it is necessary to ensure that the adjusted disperser energy input and reactor temperature parameters do not exceed the slurry flow characteristics boundary defined by this law, so as to avoid the problem of sudden increase in slurry flow resistance and fluidity failure. The third step is to use the DLVO theory as a constraint on the interaction between particles. This theory clarifies the balance between attraction and repulsion between nanoparticles in the slurry. When deriving and adjusting the parameters, it is necessary to ensure that the chemical potential input gradient of the dispersant after adjustment can keep the repulsion between particles in a stable range, meet the core requirement of preventing particle agglomeration, and ensure dispersion stability. The fourth step involves determining the adjustment intensity and direction based on the aforementioned deviation and disorder trend. Within the dual constraints of the slurry flow characteristics and the principle of inter-particle interaction, the specific adjustment amount of the corresponding process parameters is matched in reverse. For cases of insufficient dispersion uniformity and obvious disorder trend, the increase range of the disperser energy input spectrum is derived. For the risk of particle agglomeration, the adjustment rate of the dispersant chemical potential input gradient is derived. For abnormal slurry flow resistance, the temperature control range of the reactor thermodynamic field is derived. Finally, the specific adjustment parameters of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field are accurately obtained. All adjustment parameters meet the process safety boundary requirements and can drive the dispersion state to quickly return to the optimal range of the golden stable basin.

[0068] In this embodiment, the optimization problem constituting the slurry rheological constitutive equation and the DLVO theory means that when seeking process adjustment parameters, it is necessary to follow the flow characteristics of the slurry under different shear conditions (rheological constitutive equation) and the principle of attraction-repulsion between particles (DLVO theory) to ensure that the adjusted process not only conforms to the physicochemical properties of the slurry, but also effectively controls the particle dispersion state, and avoids dispersion failure or slurry property damage due to parameter abrupt changes. Following the flow characteristics of slurry under different shear conditions (rheological constitutive equation) means that when deriving process adjustment parameters, it is crucial to strictly adhere to the inherent laws governing the change of viscosity and shear stress of the slurry with shear rate. This ensures that the adjusted energy input range of the disperser will not cause abrupt changes in shear stress or viscosity exceeding the suitable range, keeping the slurry's flow resistance within an acceptable range for industrial production. This aligns with the slurry's inherent physical flow characteristics and avoids problems such as poor slurry flowability and internal structural damage caused by excessively high or low shear energy. Following the principle of interparticle attraction-repulsion (DLVO theory) means that when deriving the dispersant chemical potential input gradient adjustment parameters, it is essential to strictly adhere to the balance between attraction and repulsion between nanoparticles within the slurry. This ensures that the adjusted dispersant addition... The dosage and addition rate can maintain the repulsive force between particles at a stable level that can suppress agglomeration, matching the interfacial chemical characteristics of the particles themselves. This avoids problems such as insufficient repulsive force leading to agglomeration due to improper dispersant control, or excessive repulsive force causing system instability. At the same time, under dual constraints, the temperature adjustment range of the reactor thermodynamic field is determined to ensure that temperature changes do not simultaneously disrupt the flow characteristics of the slurry and the balance of interaction between particles. Ultimately, the values ​​and ranges of all process adjustment parameters are adapted to the physicochemical properties of the slurry. This not only effectively corrects deviations in the dispersion state through parameter adjustment and drives it to the optimal range, but also avoids adverse consequences such as dispersion failure and slurry property damage caused by various parameter mutations, ensuring that the process control process is safe and effective.

[0069] In this embodiment, the energy input spectrum of the disperser refers to the time-domain distribution characteristics of the energy output by the disperser during operation, including the variation law of stirring speed and shear power, which directly affects the agglomeration and dispersion effect of particles in the slurry.

[0070] In this embodiment, the chemical potential input gradient of the dispersant refers to the rate of concentration change and spatial distribution gradient of the dispersant in the slurry, which determines the coating efficiency and uniformity of the dispersant on the particle surface, and thus affects the particle dispersion stability.

[0071] In this embodiment, the thermodynamic field of the reactor refers to the temperature distribution state inside the reactor that affects the properties of the slurry. Temperature changes affect the viscosity of the slurry, the activity of the dispersant, and the strength of the interaction between particles, and are an important environmental parameter in the dispersion process.

[0072] In this embodiment, the precise adjustment amounts for the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field refer to the specific adjustment values ​​or ranges of change calculated for the above three key process dimensions. The magnitude and direction of the adjustment amount are determined by the degree of deviation of the situational intrusion vector and the chaotic trend of the trajectory entropy value. The degree of deviation of the situational intrusion vector is directly reflected by its magnitude. The larger the magnitude, the more serious the deviation of the current dispersion state from the target point, and the larger the value of the required process adjustment. The smaller the magnitude, the smaller the adjustment value. The direction of the intrusion vector clearly points to the specific dimension of the deviation from the dispersion state. If the vector direction points to insufficient dispersion uniformity, the adjustment amount of the energy input spectrum of the disperser should be increased accordingly. If the vector direction points to the risk of particle agglomeration and low interface potential, the adjustment amount of the chemical potential input gradient of the dispersant should be increased accordingly. If the vector direction points to abnormal flow resistance and viscosity imbalance, the temperature parameters of the thermodynamic field of the reactor should be adjusted accordingly. In this way, the control direction of the three types of process adjustment amounts is determined.

[0073] The trajectory entropy value reflects the trend of disorder in the dispersion process. The higher the entropy value, the more violent the fluctuation of the dispersion state and the greater the risk of subsequent deterioration. It will amplify the adjustment range proportionally on the basis of the adjustment amount determined by the situational invasion vector. When the entropy value is in the normal range, the basic adjustment amount will remain unchanged. At the same time, the trend of trajectory entropy value will further calibrate the adjustment direction. If the entropy value shows a continuous upward trend, the proportion of continuous control parameters will be increased in the predetermined adjustment direction to ensure that the adjustment action can not only match the deviation of the current dispersion state, but also specifically suppress the trend of disorder and deterioration in the dispersion process. Finally, the specific magnitude and control direction of the adjustment amount of the energy input spectrum of the disperser, the chemical potential input gradient of the dispersant, and the thermodynamic field of the reactor are accurately determined, so as to achieve the accurate return of the dispersion state to the target point of the golden stable basin.

[0074] In this embodiment, the precise adjustments to the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field are compiled into a set of process rheological tuning waves with specific amplitude, frequency, and phase relationships in the time domain, based on the equipment response characteristics and process safety boundaries. Essentially, this transforms abstract parameter adjustment requirements into time-domain control signals that the equipment can directly respond to, ensuring precise and stable execution of the adjustment actions. Details are as follows: The first step is to establish a precise mapping relationship between adjustment quantities and core signal characteristics. Three types of adjustment quantities—the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field—are respectively mapped to the amplitude characteristics of the process rheological tuning wave. The amplitude magnitude is directly equivalent to the specific value of the adjustment quantity, and the adjustment direction determines the positive or negative attribute of the amplitude (e.g., an increase of 500 rpm in the disperser speed corresponds to an amplitude of +500; a decrease of 1°C in the reactor temperature corresponds to an amplitude of -1). Through this mapping rule, abstract process adjustment values ​​are first transformed into quantifiable signal amplitude parameters, achieving a preliminary visualization of the adjustment requirements.

[0075] The second step involves defining the signal frequency and phase based on the equipment's response characteristics. For different process equipment with varying response speeds, corresponding signal frequencies are matched: dispersers, being high-speed devices, have their adjustment signals set to 10-20Hz to ensure smooth, shock-free speed adjustments; dispersant metering pumps, with moderate response speeds, have signals set to 5-10Hz to avoid sudden changes in the addition rate; and reactor temperature control systems, with slower responses, have signals set to 1-5Hz to prevent drastic temperature fluctuations. Simultaneously, signal phases are defined based on the timing logic of process control: with disperser energy adjustment as the baseline phase (0°), dispersant addition adjustment lags by 0.5 seconds (corresponding to phase 90°), and reactor temperature adjustment lags by 1 second (corresponding to phase 180°). This phase difference enables coordinated adjustments across multiple devices, avoiding control conflicts.

[0076] The third step is to integrate multi-dimensional signals to form a time-domain tuned wave. The amplitude-frequency-phase parameters corresponding to the three types of adjustment quantities mentioned above are integrated into a single process rheological tuned wave according to the time-domain synchronization rules: On the time-domain axis, the signal value at each moment is composed of the sub-signals corresponding to the three types of adjustment quantities. The superposition logic follows the principle of phase priority and amplitude superposition (that is, first determine the occurrence sequence of each sub-signal according to the phase difference, and then superimpose the amplitude during the overlapping period), and finally form a continuous time-domain signal curve that fully carries all process adjustment requirements.

[0077] The fourth step is to optimize the signal by verifying its response characteristics and process safety boundaries. The integrated time-domain signal is compared and verified with the response thresholds of the disperser, metering pump, and temperature control device (e.g., the maximum response frequency of the disperser is 20Hz, and the upper limit of the amplitude corresponding to the maximum addition rate of the metering pump) and process safety boundaries (e.g., the energy adjustment amplitude of the disperser does not exceed ±30% of the rated value, and the temperature adjustment amplitude does not exceed ±2℃). If the signal amplitude exceeds the safety boundary, the amplitude is corrected according to the upper limit of the boundary. If the signal frequency exceeds the equipment response range, the frequency is lowered to the equipment adaptation range. If there is a conflict in the phase timing, the phase difference is finely adjusted (adjustment step size is 0.1 seconds) until the signal is fully adapted to the equipment characteristics and process safety requirements.

[0078] After the above four-step transformation, the abstract parameter adjustment requirements have been converted into time-domain control signals (process rheological tuning waves) that the equipment can directly recognize. Dispersers, metering pumps, and other equipment can accurately execute corresponding adjustment actions by analyzing the amplitude, frequency, and phase information of the signal, while ensuring a smooth and safe adjustment process without dispersion failure or equipment damage caused by parameter abrupt changes. Those skilled in the art can fully reproduce the transformation process based on the above mapping rules, frequency / phase setting standards, and integrated verification procedures.

[0079] In this embodiment, the equipment response characteristics refer to the response speed, execution accuracy, and tolerance range of industrial equipment such as dispersers, metering pumps, and temperature control devices to control signals. When compiling process rheometry tuning waveforms, it is necessary to adapt to the equipment characteristics to avoid signal mismatch leading to adjustment failure or equipment damage.

[0080] In this embodiment, the process safety boundary refers to the limit range of process parameters that ensure production safety and product quality, including the disperser energy input not exceeding ±30% of the rated power, the dispersant addition rate not exceeding 0.05 mol / (L・min), and the reaction vessel temperature fluctuation not exceeding ±2℃, to prevent safety hazards or deterioration of slurry performance caused by parameters exceeding the limits.

[0081] In this embodiment, the process rheological tuning wave with a specific amplitude, frequency and phase relationship in the time domain is the core control signal that carries the process adjustment requirements. The amplitude corresponds to the magnitude of the adjustment, the frequency corresponds to the execution rhythm of the adjustment action, and the phase corresponds to the timing relationship of the coordinated adjustment of multiple devices. The three work together to achieve precise and coordinated control of the distributed process.

[0082] To effectively handle the complex nonlinear relationships of slurry systems and solve for the globally optimal adjustment strategy under multiple physical and safety constraints, a hybrid architecture model is proposed for the process rheological inversion kernel. The core is a structure-rheological relationship predictor constructed by a graph neural network, coupled with an online optimizer based on a model predictive control framework to solve for the optimal adjustment amount under multiple constraints.

[0083] In this embodiment, the process rheological inversion kernel is a hybrid architecture model. The core is a structure-rheological relationship predictor built from a graph neural network. This predictor learns the correlation between the slurry dispersion structure and rheological properties in historical data, enabling it to accurately predict the changing trends of slurry rheological parameters under different process adjustments. The externally coupled online optimizer, based on a model predictive control framework, uses the situational intrusion vector and trajectory entropy as target inputs, and employs multiple constraints including the slurry rheological constitutive equation, DLVO theory, and process safety boundaries. Through rolling optimization calculations, it selects the optimal process adjustment amount that allows the dispersion state to quickly return to the golden stable basin, ensuring that the control actions are precise and meet actual production needs. Specific details are as follows: The hybrid architecture consists of two layers, which are arranged in a progressive relationship of prediction, constraint, and optimization, with each layer having clearly defined functions and implementation details: 1. Core Layer: Structure-Rheological Relationship Predictor Built with Graph Neural Network: This predictor is the core of the kernel's perception and prediction capabilities. Its core function is to learn and establish a precise correlation between three factors: slurry dispersion structure (e.g., particle dispersion uniformity, agglomeration state), process adjustment parameters (dispersant speed, dispersant dosage, reactor temperature), and rheological property parameters (characteristic viscosity, shear stress). This allows for accurate prediction of the changing trends of slurry rheological parameters under different process adjustments. Its specific implementation details are as follows: Data input: The training data is based on the historical best production batch data cluster. The data consists of two parts: input data is process adjustment parameters and dispersion state parameters (process adjustment parameters: disperser speed range 1000-5000 rpm, dispersant addition range 0.5%-3% total solids content, reactor temperature range 20-30℃; dispersion state parameters: absorbance fingerprint characteristic peak value, interface potential signal). The output data is the corresponding slurry rheological property parameters (steady-state viscosity at a shear rate of 50 s⁻¹, shear stress at different shear rates).

[0084] Network Structure Design: A two-layer Graph Convolutional Network (GCN) architecture is adopted to abstract the slurry dispersion system into a graph structure for learning: ① Node Definition: Carbon nanoparticle dispersant molecules and solvent micro-elements are set as the core nodes of the graph, and each node is assigned characteristic attributes (e.g., particle nodes: particle size, number of surface active sites; dispersant nodes: molecular chain length, adsorption activity; solvent nodes: viscosity, dielectric constant); ② Edge Definition: Particle-dispersant adsorption, interparticle repulsion / attraction, and solvent-particle wetting are set as the edges of the graph, and edge weights are assigned according to the intensity of the interaction (e.g., strong adsorption is weighted at 0.8-1.0, and weak adsorption is weighted at 0.2-0.4).

[0085] Training and prediction logic: Training is completed through a process of neighborhood feature aggregation, node state update, and outputting the correlation model. First, the mean aggregation function is used to collect feature information within the neighborhood of each node (such as the chain length features of the dispersant nodes surrounding the particle node and the viscosity features of the solvent node). Then, the neighborhood features and the node's own features are fused through linear transformation and ReLU activation function to update the node state, and finally, the structure-rheological relationship model is trained. When using it, the current situation intrusion vector (reflecting the degree of deviation of the dispersion state) + trajectory entropy value (reflecting the disorder of the dispersion process) + initial process parameters are input, and the model can output the trend curves of the changes in slurry rheological parameters (characteristic viscosity, shear stress) under different process adjustment schemes (such as predicting the decreasing trend of slurry characteristic viscosity in the next 10 monitoring periods after the disperser speed is increased by 20%).

[0086] 2. Outer Layer: Online Optimizer of the Model Predictive Control Framework This optimizer is the core of the kernel's decision-making and control. Its core function is to select the optimal process adjustment amount that allows the dispersed state to quickly return to the golden stable basin, based on the prediction results provided by the structure-rheology predictor and combined with multiple constraints. Its specific implementation details are as follows: Target input: Directly receive two core signals—the situational intrusion vector (including deviation direction: such as insufficient dispersion uniformity or excessive flow resistance; including deviation degree: quantized by the modulus) and the trajectory entropy value (including the chaotic trend: such as a continuous rise indicates increased chaos).

[0087] Specific applications of multiple constraints: ① Slurry rheological constitutive equation constraint: The Herschel-Bulkley type rheological constitutive equation is adopted (parameters such as yield stress and consistency coefficient are determined by offline experimental fitting). The shear stress, shear rate, and viscosity of the slurry after process adjustment must conform to the law of the equation to avoid the problem of slurry flow failure caused by sudden shear stress. ② DLVO theory constraint: Based on the balance relationship between van der Waals attraction potential energy and double electric layer repulsion potential energy between particles in DLVO theory, the absolute value of the repulsion potential energy between particles after adjustment is constrained to ≥30kT (k is Boltzmann constant, T is absolute temperature) to ensure that particles do not agglomerate. ③ Process safety boundary constraint: It is specified that the disperser speed should not exceed 5000 rpm, the amount of dispersant added should not exceed 3% of the total solid content, and the temperature fluctuation of the reactor should not exceed ±2℃ to avoid safety hazards caused by parameter exceeding the limit.

[0088] Rolling optimization calculation logic: A rolling optimization strategy combining prediction time domain and control time domain is adopted: ① Prediction stage: Based on the output of the structure-rheological relationship predictor, predict the dispersion state change trajectory (dispersion synergy index, interface potential signal change curve) corresponding to 3-5 candidate process adjustment schemes (e.g., Scheme 1: rotation speed +200 rpm, dispersant addition +0.2%; Scheme 2: rotation speed +300 rpm, temperature +1℃) within the next 5 monitoring cycles (10 seconds per cycle); ② Optimization stage: Define the optimization objective function (with the objective of maintaining the absolute value of the interface potential at ≥30mV and the rheological parameter fluctuation range ≤10% by rapidly regressing the dispersion synergy index to 1.0±0.05, with a weight allocation of 0.6 for the dispersion state index and 0.4 for the rheological parameter index), and calculate the objective function value for each candidate scheme; ③ Screening phase: Select the scheme with the optimal objective function value (i.e., closest to the ideal dispersion state) and satisfy all constraints, and extract the dispersion machine energy input spectrum adjustment, dispersant chemical potential input gradient adjustment, and reactor thermodynamic field adjustment from the scheme as the optimal process adjustment for the final output.

[0089] II. Core data interaction logic of the two-tier architecture: The two-layer architecture achieves collaborative work through unidirectional input and bidirectional verification, ensuring unambiguous data flow: ① Unidirectional input: The output of the structure-rheology predictor (the trend of rheological parameter changes under different process adjustments) directly serves as the basis for prediction by the online optimizer of model predictive control, allowing the optimizer to perform multi-scheme prediction; ② Bidirectional verification: When the optimizer is screening candidate schemes, if it finds that the rheological parameter change trend of a certain scheme exceeds the constraint range (such as violating the rheological constitutive equation), it will feed the scheme back to the predictor. The predictor will re-output the accurate rheological parameters under the scheme, and the optimizer will verify again to ensure that the final selected optimal adjustment amount conforms to the prediction law and satisfies the constraint conditions.

[0090] III. Complete Kernel Workflow (Reproducible Steps): Data acquisition: Real-time acquisition of the current slurry's intrusion vector (magnitude, direction) and trajectory entropy value (value, trend of change); Trend prediction: Input the above data into the structure-rheology predictor and output the rheological parameter change trends corresponding to 3-5 sets of candidate process adjustment schemes; Constraint optimization: The online optimizer aims to quickly regress the optimal value from the dispersed state. It combines the rheological constitutive equation, DLVO theory and process safety boundary to perform rolling optimization calculations on candidate solutions and select the solution with the optimal objective function value. Output results: Output the energy input spectrum adjustment of the disperser (e.g., rotation speed +250 rpm), the chemical potential input gradient adjustment of the dispersant (e.g., addition rate +0.02 mol / (L·min)), and the thermodynamic field adjustment of the reactor (e.g., temperature +1℃) corresponding to the optimal scheme.

[0091] To ensure that intelligent control commands can drive actual industrial equipment without damage and reliably, and to fully record the production process for traceability, analysis, and optimization, an industrial protocol injection and entropy lake module is proposed, including: The protocol injection gateway submodule supports the decomposition and adaptation of process rheological tuning waves into OPCUA information models, EtherCAT drive messages, and time-sensitive network scheduling commands, enabling lossless injection with heterogeneous industrial control systems. The process entropy lake submodule is used to archive and store each frame of real-time dispersed phase space snapshots, process evolution trajectory fragments, situational intrusion vectors, and process rheological tuning waves in an immutable chain structure with nanosecond-level timestamps, forming a process entropy lake with complete causal relationships.

[0092] In this embodiment, the process rheometry tuning wave is decomposed and adapted into an OPCUA information model, EtherCAT drive messages, and time-sensitive network scheduling instructions to achieve lossless injection with heterogeneous industrial control systems. The core is to split and convert the unified process control signal according to the format requirements of different industrial protocols to ensure that key control information is not lost during transmission and execution, while adapting to the communication rules of different types of control systems.

[0093] In this embodiment, the OPCUA information model is a general industrial communication information specification. Through standardized data structures and interfaces, it encapsulates the control parameters (such as amplitude and frequency) in process rheometry tuning waves into data objects that can be recognized across platforms, thereby achieving interoperability between devices from different manufacturers.

[0094] In this embodiment, the EtherCAT drive message is a communication data frame based on the EtherCAT real-time industrial Ethernet protocol. It encapsulates the control command of the process rheometry tuning wave according to high real-time requirements, and can quickly transmit it to the execution equipment such as dispersers and metering pumps to ensure the immediate response of the adjustment action.

[0095] In this embodiment, the time-sensitive network scheduling command is a scheduling control signal based on the time-sensitive network (TSN) protocol, used to coordinate the control timing of multiple devices, ensuring that multiple adjustment actions corresponding to the process rheological tuning wave are executed synchronously according to a preset phase relationship, and avoiding timing disorder from affecting the control effect.

[0096] In this embodiment, the process rheological tuning wave is decomposed and adapted into an OPCUA information model, an EtherCAT driver message, and a time-sensitive network scheduling command. This involves first splitting the core parameters of the tuning wave, such as amplitude, frequency, and phase, and then converting and encapsulating the data according to the format specifications of the three protocols to form control signals adapted to different communication scenarios.

[0097] In this embodiment, a heterogeneous industrial control system refers to an industrial control system composed of different manufacturers, different communication protocols, and different functional modules, including different types of control devices such as distributed control systems (DCS) and programmable logic controllers (PLCs), whose communication interfaces and data formats differ.

[0098] To present complex internal state models and early warning information to operators in an intuitive and immersive way, and to achieve predictive early warnings earlier than traditional alarms, the following additional measures are proposed: The panoramic perception and early warning cockpit includes a phase space holographic projection unit and an entropy change early warning unit, wherein: Phase space holographic projection unit is used to present a three-dimensional dynamic hologram of real-time dispersed phase space, golden stable basin and process evolution trajectory in three dimensions using augmented reality technology; The entropy change precursor warning unit is used to trigger an entropy increase precursor warning when the trajectory entropy value monotonically increases beyond a set threshold within a continuous period, or when the intrusion direction of the situation intrusion vector continuously moves away from the target point in the centrifugal domain.

[0099] In this embodiment, augmented reality technology is used to present a three-dimensional dynamic hologram of the real-time dispersed phase space, the golden stable basin, and the process evolution trajectory. This is achieved by using augmented reality equipment to transform the two-dimensional real-time dispersed phase space, the closed golden stable basin region, and the dynamic process evolution trajectory into a three-dimensional image that can be observed intuitively. Operators can view the spatial distribution and changing trends of the dispersed state through an interactive interface, thereby improving the intuitiveness of monitoring.

[0100] In this embodiment, the continuous period refers to the continuous monitoring period divided according to a preset time interval. The duration of each period is matched with the response speed of the slurry dispersion process. The default setting is 10 seconds / period to ensure that the continuous change trend of the dispersion state can be captured in a timely manner.

[0101] In this embodiment, when the trajectory entropy value monotonically increases beyond a set threshold within a continuous period, or when the intrusion direction of the situational intrusion vector continuously deviates from the target point in the centrifugal domain, an entropy increase precursor warning is triggered. The core is to identify the trend of deterioration of the dispersed state in advance through two different dimensions of anomaly judgment logic, so as to avoid waiting until the parameters exceed the standard before alarming, and to reserve sufficient time for regulation.

[0102] In this embodiment, the threshold value refers to a pre-set critical value of trajectory entropy, which is used to determine whether the disorder of the dispersion process is abnormal. The default value is 0.5, which can be flexibly adjusted according to the characteristics of different carbon nanomaterials and slurry formulations.

[0103] In this embodiment, the invasion direction of the situational intrusion vector refers to the deviation of the latest transient phase point from the target point in the core region of the golden stable basin. It is determined by the coordinate difference in phase space and intuitively reflects the specific direction of the dispersion state deviating from the optimal interval.

[0104] In this embodiment, the intrusion direction of the situational intrusion vector continuously moving away from the target point in the centrifugal domain means that within three or more consecutive monitoring cycles, the deviation direction of the transient phase point is always far away from the target point in the centrifugal domain, with no trend of return, indicating that the dispersion state is continuously deteriorating and emergency control needs to be initiated.

[0105] In this embodiment, the entropy increase precursor warning is an early warning mechanism that informs operators of the risk of deterioration in the dispersion status through sound and light signals and interface prompts. Unlike traditional alarms that are triggered after parameters exceed the standard, it can achieve early detection and early control, reducing the probability of unqualified dispersion quality.

[0106] like Figure 1 As shown, this invention provides an embodiment of an online monitoring and feedback method for the dispersion quality of carbon nanomaterial slurries, comprising: Simultaneously acquire the absorbance fingerprint, rheological response spectrum, and interfacial potential signal of the slurry; The absorbance fingerprint is coupled with the rheological response spectrum to generate the dispersion co-index. The dispersion co-index is used as the horizontal axis and the interface potential signal is used as the vertical axis to construct a real-time dispersed phase space. The synchronous data is mapped to the transient phase points in the real-time dispersed phase space. The transient phase points of the time series are connected to form the process evolution trajectory. Based on the pre-loaded gold stable basin model, the current entropy state of the process evolution trajectory is calculated, and the situation invasion vector of the invasion into the gold stable basin is solved. Receive the situational intrusion vector and, through the embedded process rheological inversion kernel, inversely synthesize a process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin. The process rheometry tuned wave is compiled into a control pulse sequence that can be recognized by the corresponding industrial equipment and injected into the production line. At the same time, the real-time dispersed phase space, process evolution trajectory and process rheometry tuned wave data of the whole cycle are deposited into the process entropy lake.

[0107] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. An online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry, characterized in that, include: A multi-parameter flow field probe array is used to simultaneously acquire the absorbance fingerprint, rheological response spectrum and interface potential signal of the slurry; The dispersed phase space evolution modeling module is used to couple absorbance fingerprints with rheological response spectra to generate a dispersion co-index, and construct a real-time dispersed phase space with the dispersion co-index as the horizontal axis and the interface potential signal as the vertical axis, mapping synchronous data into transient phase points in the real-time dispersed phase space; The process trajectory entropy state perception module is used to connect the transient phase points of the time series to form the process evolution trajectory, and calculate the current entropy state of the process evolution trajectory based on the preloaded golden stable basin model, and solve the situation invasion vector of the invasion into the golden stable basin. The inverse mapping synthesis engine is controlled to receive the situational intrusion vector and synthesizes the process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin through the embedded process rheological inversion kernel. The industrial protocol injection and entropy lake module is used to compile the process rheological tuning wave into a control pulse sequence that can be recognized by the corresponding industrial equipment and inject it into the production line. At the same time, it precipitates the real-time dispersed phase space, process evolution trajectory and process rheological tuning wave data of the whole cycle into the process entropy lake. The process trajectory entropy state sensing module includes: The trajectory chain generation submodule is used to connect transient phase points that appear in the real-time dispersed phase space in a time sequence to generate process evolution trajectories that characterize the dynamic evolution of the dispersion process. The entropy state solution submodule contains a gold stable basin model; the gold stable basin model defines a closed, optimal gold stable basin region in the real-time dispersed phase space. The entropy state solution submodule is used to calculate the vector from the latest transient phase point to the target point in the core region of the golden stable basin, which is used as the situational intrusion vector; where the direction of the situational intrusion vector is defined as the intrusion direction and the magnitude is defined as the intrusion intensity. Simultaneously, the recent rate of curvature change of the process evolution trajectory is calculated as the trajectory entropy value; the trajectory entropy value is used to characterize the disorder trend of the process. Among them, the Gold Stable Basin Model is a dynamic deformable basin generated by combining Gaussian mixture model and convex hull algorithm based on multiple sets of historical best production batch data clusters. The boundary morphology, area, and target location of the dynamically deformable basin can adaptively deform based on the input carbon nanomaterial crystal structure code and slurry formulation hash value. Among them, regulating the inverse mapping synthesis engine includes: The inversion solution submodule contains a process rheology inversion kernel. The process rheology inversion kernel takes the situational invasion vector and trajectory entropy as inputs. By solving an optimization problem constrained by the slurry rheological constitutive equation and DLVO theory, it inversely maps out the precise adjustment amounts of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field. The tuned wave compilation submodule is used to compile the precise adjustments of the disperser energy input spectrum, the dispersant chemical potential input gradient, and the reactor thermodynamic field into a set of process rheological tuned waves with specific amplitude, frequency, and phase relationships in the time domain, according to the equipment response characteristics and process safety boundaries. The process rheological inversion kernel is a hybrid architecture model. At its core is a structure-rheological relationship predictor built by a graph neural network, coupled with an online optimizer based on a model predictive control framework to solve for the optimal adjustment amount under multiple constraints.

2. The online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry according to claim 1, characterized in that, Multi-parameter flow field probe array, including: Photon acquisition probes are used to acquire absorbance fingerprints that characterize the spatial distribution of nanoparticle clusters online. A shear stress acquisition probe is used to acquire rheological response spectra characterizing the internal structural strength of slurries online. The double-layer acquisition probe is used to acquire interfacial potential signals that characterize the interaction potential between particles online.

3. The online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry according to claim 1, characterized in that, The dispersed-state phase space evolution modeling module includes: The synergy index synthesis submodule is used to divide the real-time acquired absorbance fingerprint by the characteristic viscosity value in the rheological response spectrum to calculate the synergy index of dispersion, which characterizes the synergistic relationship between dispersion uniformity and flow resistance. The phase space mapping submodule is used to construct a two-dimensional real-time dispersed phase space with the dispersion co-exponent as the horizontal axis and the interface potential signal as the vertical axis, and to map the dispersion co-exponent and interface potential signal at each moment into a transient phase point in the real-time dispersed phase space.

4. The online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry according to claim 1, characterized in that, The industrial protocol injection and entropy lake module includes: The protocol injection gateway submodule supports the decomposition and adaptation of process rheological tuning waves into OPCUA information models, EtherCAT drive messages, and time-sensitive network scheduling commands, enabling lossless injection with heterogeneous industrial control systems. The process entropy lake submodule is used to archive and store each frame of real-time dispersed phase space snapshots, process evolution trajectory fragments, situational intrusion vectors, and process rheological tuning waves in an immutable chain structure with nanosecond-level timestamps, forming a process entropy lake with complete causal relationships.

5. The online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry according to claim 1, characterized in that, Also includes: The panoramic perception and early warning cockpit includes a phase space holographic projection unit and an entropy change early warning unit, wherein: Phase space holographic projection unit is used to present a three-dimensional dynamic hologram of real-time dispersed phase space, golden stable basin and process evolution trajectory in three dimensions using augmented reality technology; The entropy change precursor warning unit is used to trigger an entropy increase precursor warning when the trajectory entropy value monotonically increases beyond a set threshold within a continuous period, or when the intrusion direction of the situation intrusion vector continuously moves away from the target point in the centrifugal domain.

6. A method for online monitoring and feedback of the dispersion quality of carbon nanomaterial slurry, characterized in that, The online monitoring and feedback device for the dispersion quality of carbon nanomaterial slurry according to any one of claims 1 to 5 includes: Simultaneously acquire the absorbance fingerprint, rheological response spectrum, and interfacial potential signal of the slurry; The absorbance fingerprint is coupled with the rheological response spectrum to generate the dispersion co-index. The dispersion co-index is used as the horizontal axis and the interface potential signal is used as the vertical axis to construct a real-time dispersed phase space. The synchronous data is mapped to the transient phase points in the real-time dispersed phase space. The transient phase points of the time series are connected to form the process evolution trajectory. Based on the pre-loaded gold stable basin model, the current entropy state of the process evolution trajectory is calculated, and the situation invasion vector of the invasion into the gold stable basin is solved. Receive the situational intrusion vector and, through the embedded process rheological inversion kernel, inversely synthesize a process rheological tuned wave that can drive the transient phase point from the current entropy state to the target point in the core domain of the gold stable basin. The process rheometry tuned wave is compiled into a control pulse sequence that can be recognized by the corresponding industrial equipment and injected into the production line. At the same time, the real-time dispersed phase space, process evolution trajectory and process rheometry tuned wave data of the whole cycle are deposited into the process entropy lake.

Images