Preparation and control method of high-dispersibility ultra-fine silver powder for photovoltaic silver paste

Abstract

The present application relates to the technical field of metal powder manufacturing, and particularly relates to a preparation and regulation method of high-dispersibility ultra-fine silver powder for photovoltaic silver paste. Firstly, the present application carries out deep decoupling analysis on the motor torque signal, obtains response lag time and torque roughness index, realizes accurate quantification of different micro characteristics of soft clusters and hard clusters, and can eliminate the masking of macroscopic viscosity to micro characteristics in high solid content system; secondly, different dynamic thresholds are set for soft clusters and hard clusters respectively, which can reduce the bottom noise fluctuation caused by different batch equipment wear and viscosity change in the reaction stage, improve the accuracy of real-time distinguishing of soft clusters and hard clusters, and is beneficial to long-term stable operation in industrial field; finally, different regulation strategies are formulated for soft clusters and hard clusters by distinguishing different states of the silver paste, which can avoid the traditional one-size-fits-all regulation, realizes the treatment according to the disease for different states, and significantly improves the stability and efficiency of the preparation of high-dispersibility silver powder.

Description

Technical Field

[0001] This invention relates to the field of metal powder manufacturing technology, specifically to a method for preparing and controlling highly dispersible ultrafine silver powder for photovoltaic silver paste. Background Technology

[0002] Ultrafine silver powder for photovoltaic silver paste is usually prepared by liquid-phase chemical reduction. In order to reduce production costs, industrial production tends to use reaction systems with high solid content and large volume. In the later stages of the reaction in high solid content systems, as a large number of solid particles are generated, the rheological properties of the paste will change significantly, causing the shear energy transmitted by the stirring blades to dissipate at the microscale. The micro-mixing efficiency near the feeding nozzle cannot match the chemical reduction rate. Under this condition, if the newly added reducing agent cannot be quickly dispersed by the surrounding dense medium, its local excess will lead to two micro-failure modes: First, the polymeric dispersant (such as PVP) will undergo molecular chain entanglement due to excessively high local concentration or charge shielding, losing its steric hindrance and causing soft agglomeration; Second, excessively high local supersaturation will directly induce the explosive growth and adhesion of silver nuclei, forming irreversible micron-sized hard agglomerates.

[0003] Existing technologies typically control the reaction process by monitoring macroscopic viscosity or stirring torque. However, in large-volume, high-viscosity systems, the macroscopic torque signal mainly reflects the overall rheological resistance of the slurry, while microscopic molecular chain entanglement and localized hard agglomeration both manifest as resistance fluctuations in the macroscopic signal. Existing monitoring methods struggle to distinguish between these two fundamentally different failure modes. This lack of differentiation leads to a simplistic control strategy: in industrial settings, the only option is often to blindly increase the stirring speed. However, for soft agglomerates, simple mechanical shearing often fails to untangle the molecular chains and may even exacerbate the shear degradation of the polymeric dispersant; for hard agglomerates, conventional stirring shear forces are insufficient to break them up. Therefore, without introducing expensive and easily clogged online particle size analyzers, existing equipment cannot detect and differentiate between dispersant failure (soft cluster formation) and hard agglomeration formation in high-concentration systems in real time, making it difficult to achieve stable and efficient preparation of highly dispersible silver powder. Summary of the Invention

[0004] To address the technical problem of existing technologies being unable to detect and distinguish between dispersant failure and hard agglomeration formation in high-concentration systems in real time without introducing expensive and easily clogged online particle size analyzers, thus hindering the stable and efficient preparation of highly dispersible ultrafine silver powder, the present invention aims to provide a method for controlling the preparation of highly dispersible ultrafine silver powder for photovoltaic silver paste. The specific technical solution adopted is as follows: This invention proposes a method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste, the method comprising: During the silver paste stirring process, all real-time torque initial sequences of the motor stirring are acquired; all real-time torque initial sequences are fused and preprocessed to determine the real-time torque fused sequence; all reference torque initial sequences and reference torque fused sequences are acquired. The response lag time is determined based on the time distribution benchmark deviation of the real-time torque fusion sequence; in the frequency domain, the corresponding torque roughness index is determined based on the time-series fluctuation benchmark deviation of the torque value in the real-time torque fusion sequence. Based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in response lag time, a dynamic first threshold is determined; in the frequency domain, based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in torque roughness index, a dynamic second threshold is determined. The numerical relationships between the dynamic first threshold and the response lag time, and between the dynamic second threshold and the torque roughness index are analyzed to determine the silver paste stirring state. Based on the silver paste stirring state, silver powder preparation control is implemented.

[0005] Furthermore, the calculation method for the real-time torque fusion sequence includes: In all real-time torque initial sequences, calculate the average value of the torque values ​​at the same index time, determine the initial fusion value at each index time, arrange all the initial fusion values, and determine the reference initial fusion sequence; The reference initial fusion sequence is standardized to determine the standard initial fusion value at each index time. The difference between the standard initial fusion value at each index time and a preset constant is calculated to determine the local real-time average torque at each index time. The local real-time average torque at all index times is arranged to determine the real-time torque fusion sequence.

[0006] Furthermore, the method for calculating the response lag time includes: In the real-time torque fusion sequence, the local time interval value at each index time is determined by the product between the index value and the torque value at each index time; the local time interval values ​​at all index times are accumulated to obtain the cumulative time interval value; the torque values ​​at all index times are accumulated to determine the cumulative torque value in the real-time torque fusion sequence; the cumulative time interval value is used as the numerator and the cumulative torque value is used as the denominator to obtain the time centroid of the real-time torque fusion sequence. Based on the principle of obtaining the time centroid of the real-time torque fusion sequence, the time centroid of the reference torque fusion sequence is determined; the difference between the time centroid of the real-time torque fusion sequence and the time centroid of the reference torque fusion sequence is taken as the response lag time.

[0007] Furthermore, the method for calculating the torque roughness index includes: In the real-time torque fusion sequence, the second-order difference value of the torque value at each index time is calculated, the second-order difference value at all index times is arranged to determine the corresponding second-order difference sequence, the L1 norm of the second-order difference sequence is calculated, and the L1 norm is used as the periodic feature value of the real-time torque fusion sequence on the torque value. Based on the principle of obtaining the periodic characteristic value of the real-time torque fusion sequence on the torque value, the periodic characteristic value of the reference torque fusion sequence on the torque value is determined; the difference between the periodic characteristic value of the real-time torque fusion sequence on the torque value and the periodic characteristic value of the reference torque fusion sequence on the torque value is used as the torque roughness index.

[0008] Furthermore, the process of obtaining the dynamic first threshold includes: In the initial sequence of the reference torque, the time centroid of the initial sequence of the reference torque is determined based on the principle of obtaining the time centroid of the real-time torque fusion sequence. The time fluctuation value is determined based on the relative deviation between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence. The real-time torque mean and the reference torque mean of the real-time torque fusion sequence and the reference torque fusion sequence are obtained respectively. Based on the real-time torque mean and the reference torque mean, the comparison feature value between the real-time torque mean and the reference torque mean is determined. The dynamic first threshold is obtained by using the product of the time fluctuation value and the comparison feature value as a basis, combined with the first sensitivity coefficient and the first empirical compensation index for correction.

[0009] Furthermore, the method for calculating the time fluctuation value includes: Calculate the difference between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence to determine the response lag time corresponding to each initial reference torque sequence. Based on the standard deviation of the response lag times corresponding to all initial reference torque sequences, determine the time fluctuation value.

[0010] Furthermore, the process of obtaining the dynamic second threshold includes: In the initial sequence of the reference torque, the periodic characteristic value of the initial sequence of the reference torque is determined based on the principle of obtaining the periodic characteristic value of the torque value in the real-time torque fusion sequence. Calculate the difference between the periodic characteristic value of each initial reference torque sequence and the periodic characteristic value of the fused reference torque sequence, determine the torque roughness index corresponding to each initial reference torque sequence, and determine the roughness fluctuation value based on the standard deviation of the torque roughness index corresponding to all initial reference torque sequences. The dynamic second threshold is determined based on the roughness fluctuation value and the comparison feature value.

[0011] Furthermore, the method for determining the dynamic second threshold based on the roughness fluctuation value and the comparison feature value includes: The dynamic second threshold is obtained by using the product of the roughness fluctuation value and the contrast feature value as a basis, combined with the second sensitivity coefficient and the second empirical compensation index for correction.

[0012] Furthermore, the method for determining the stirring state of the silver paste includes: If the torque roughness index is greater than the dynamic second threshold, the silver paste is determined to be in the hard cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is greater than the dynamic first threshold, then the silver paste is determined to be in the soft cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is less than or equal to the dynamic first threshold, then the silver paste is determined to be in an ideal stable state.

[0013] Furthermore, the method for controlling the preparation of silver powder includes: If the silver paste is in the hard cluster formation stage, physical crushing of the silver paste is used for regulation. If the silver paste is in the soft cluster formation stage, a chemical regulator is added to the silver paste to control it. If the silver paste is in an ideal and stable state, then maintain the current working state without adjustment.

[0014] The present invention has the following beneficial effects: This invention addresses the difficulty in accurately identifying dispersant failure and hard agglomeration in silver paste during silver powder preparation, which leads to poor stability and low efficiency in the preparation of highly dispersed silver powder. First, this invention performs deep decoupling analysis on the motor torque signal to obtain the response hysteresis time and torque roughness index, achieving accurate quantification of the different microscopic characteristics of soft and hard clusters. This eliminates the masking of microscopic features by macroscopic viscosity in high-solids systems. Second, by setting different dynamic thresholds for soft and hard clusters, it reduces background noise fluctuations caused by equipment wear and viscosity changes during the reaction phase, improving the accuracy of real-time differentiation between soft and hard clusters and benefiting long-term stable operation in industrial settings. Finally, by differentiating the different states of the silver paste, different control strategies are formulated for soft and hard clusters, avoiding the traditional one-size-fits-all approach and enabling targeted treatment for different states, significantly improving the stability and efficiency of highly dispersed silver powder preparation. Attached Figure Description

[0015] Figure 1 The flowchart illustrates a method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste, as provided in one embodiment of the present invention. Figure 2A waveform diagram of a reference torque fusion sequence under normal conditions provided in an embodiment of the present invention; Figure 3 This is a waveform comparison diagram of a real-time torque fusion sequence and a reference torque fusion sequence in a soft cluster state, provided as an embodiment of the present invention. Figure 4 This is a waveform comparison diagram of a real-time torque fusion sequence and a reference torque fusion sequence in a hard cluster state, provided as an embodiment of the present invention. Detailed Implementation

[0016] The following describes in detail, with reference to the accompanying drawings, a specific scheme for the preparation and control method of highly dispersed ultrafine silver powder for photovoltaic silver paste provided by the present invention.

[0017] Please see Figure 1 The diagram illustrates a flowchart of a method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to an embodiment of the present invention. The method includes: Step S101: During the silver paste stirring process, acquire all real-time torque initial sequences of the motor stirring; merge and preprocess all real-time torque initial sequences to determine the real-time torque fusion sequence; acquire all reference torque initial sequences and reference torque fusion sequences.

[0018] The preparation principle of highly dispersible ultrafine silver powder is to add silver-containing compounds (such as silver nitrate) to a good solvent (such as deionized water or ethylene glycol) and stir. The good solvent includes a reducing agent (such as hydrazine hydrate) to reduce elemental silver, and a high molecular weight dispersant (such as polyvinylpyrrolidone, abbreviated as PVP) is used to assist in the process of fully mixing and uniformly dispersing the silver-containing compounds and reducing agents. In this process, the solution in the reactor needs to be continuously stirred by a motor-controlled stirring blade to accelerate the reaction rate. When the silver-containing solution (such as silver nitrate added to deionized water) is added to the reactor, each drop will produce a small but strictly periodic pressure (or flow) pulsation in the stirring slurry. By utilizing the nonlinear amplification effect of the local shear stress on the microstructure of the high non-Newtonian fluid (silver paste), the rheological response characteristics at the moment of addition can be captured. The influence of this rheological response characteristics on the stirring blade is reflected in the torque of the stirring blade. That is, each drop of silver-containing solution will change the torque of the stirring blade. Therefore, by analyzing the torque of the stirring blade during the stirring process, the process of each drop of silver-containing solution and its impact can be intuitively displayed. Therefore, it is necessary to obtain the torque sequence of the motor stirring during the silver paste stirring process to monitor the state of the solution in the reactor.

[0019] As an example, in one specific implementation of this invention, the stirring motor is a 2.2kW variable frequency motor, and the stirring blade is a frame-type stirring blade. A torque sensor is installed on the stirring shaft of the stirring motor. The variable frequency motor controls the frame-type stirring blade to stir in the reactor. In addition, the silver-containing solution is delivered by a pulsating volumetric pump (such as a precision plunger metering pump or a diaphragm metering pump). This type of pump generates strictly periodic flow / pressure pulsations during operation. To ensure that the minute feeding pulses can cause flow field changes detectable by the motor torque sensor, the hardware layout is clearly defined: the outlet of the silver-containing solution feeding pipe is located in the high-shear mixing zone of the stirring blade sweep area, i.e., the radial distance between the feeding pipe outlet and the outer edge tip of the stirring blade is controlled between 10mm and 50mm. This is because in this region, the fluid is directly coupled to the power output end of the stirring blade, and the local viscosity or rheological state changes caused by the feeding pulses can be transmitted to the stirring shaft with minimal attenuation, thus being captured by the motor torque sensor.

[0020] During the chemical reaction of adding a silver-containing solution, on the one hand, the rheological characteristics of the solution in the reactor change with the formation of solid silver particles, which in turn changes the torque of the stirring blades. If only the torque sequence of the motor stirring at a certain moment is used to reflect the torque sequence of the motor stirring throughout the entire stirring process, the monitoring signal will be insufficiently representative and unable to capture the dynamic changes throughout the stirring process. Therefore, it is necessary to collect the torque sequence of the stirring blades in real time. On the other hand, with the formation of solid silver particles, the random turbulent pulsation inside the solution in the reactor is significantly enhanced, and the signal-to-noise ratio of the rheological response signal caused by a single drop is greatly reduced. Since the pushing action of the feed pump has a strictly periodic coherent characteristic, while the stirring turbulence noise is a random incoherent signal, in order to suppress the background noise caused by stirring turbulence, multiple real-time initial torque sequences are collected, fused, and preprocessed. This can eliminate the background noise caused by stirring turbulence to the greatest extent, extract the most coherent and stable rheological response feature (i.e., torque) with respect to the dropping pulse, and improve the accuracy of subsequent calculations.

[0021] First, due to the interference of individual equipment differences (such as motor harmonics, shaft eccentricity, and mechanical clearance) on rheological characteristic analysis, and the different equipment having different individual differences, in order to eliminate the influence of these equipment differences and to obtain the torque change that can be shown only by adding silver solution, it is necessary to quantify the torque sequence of the equipment itself as the reference torque sequence for subsequent calculations. Based on this reference torque sequence, the real-time torque sequence of the silver solution addition process can be collected, which can eliminate the influence of individual equipment differences, significantly improve the consistency and comparability of the signal, and make the rheological characteristic analysis of different equipment and different batches have a unified benchmark. Second, due to the influence of real-world errors, in order to obtain an accurate and stable reference torque sequence, it is necessary to collect multiple initial reference torque sequences. The reference torque fusion sequence obtained by the initial reference torque sequences can effectively suppress random noise, instantaneous disturbances, and equipment start-up transients in a single acquisition process, thereby constructing an idealized and highly stable torque sequence that represents the inherent mechanical and rheological characteristics of the system. Therefore, before the reaction system is formally added with silver-containing solution for chemical reduction reaction, a one-time benchmark calibration is performed during the bottom liquid circulation stage (at which time the reactor contains only good solvent, reducing agent and dispersant). This is because during the bottom liquid stage, only liquid exists in the reactor, and the fluid medium is a homogeneous Newtonian fluid with no particle interference. The data collected at this time can truly reflect the inherent mechanical characteristics of the system.

[0022] As an example, in a specific implementation of this invention, the method for obtaining multiple real-time torque initial sequences is as follows: S1: The system starts the stirring motor to the constant speed set by the process (e.g., 300 rpm) and controls the feed pump to run in internal circulation (at this time, the feed pump delivers the silver-containing solution). The system continuously monitors the drive current of the feed pump motor or the encoder signal through a torque sensor. When the rising edge of the drive signal is detected to exceed 80% of the peak value, the system marks this moment as the pump push trigger moment. This moment physically corresponds to the starting point when the feed pump piston or diaphragm begins to apply pressure to the pipeline; S2: During the reduction reaction, with At time zero, the system synchronously captures the torque signal of the stirring shaft at a preset high-frequency sampling rate (e.g., 200Hz), with a capture duration of [time value missing]. ,in Multiple real-time torque initial signals are acquired corresponding to the effective duration of a single complete stroke (e.g., 0.5 seconds).

[0023] It should be noted that the method for obtaining multiple initial reference torque sequences is based on the same principle as the method for obtaining multiple initial real-time torque sequences. Simply replace the silver-containing solution delivered by the feed pump with a good solvent, and replace "during the reduction reaction" with "before the reduction reaction," then repeat steps S1 to S2 to obtain multiple initial reference torque sequences. Replacing the silver-containing solution delivered by the feed pump with a good solvent aims to avoid the influence of solid silver particles generated during the reduction reaction. Simultaneously, process parameters such as the motor stirring speed and the feed pump frequency are strictly maintained throughout the entire process, and the system also includes signals generated by the feed pump's internal circulation operation. It should be further explained that existing methods typically control the reaction process by monitoring macroscopic stirring torque. That is, after acquiring the reference torque and real-time torque, a fixed macroscopic torque threshold is set to monitor whether the stirring speed needs to be adjusted. If the real-time torque exceeds the macroscopic torque threshold (e.g., set to 1.5 times the reference torque), the stirring speed is uniformly increased (e.g., increasing the stirring speed from 400 rpm to 500 rpm and maintaining it for 30 seconds). This method uses the same control measures regardless of the failure mode, resulting in poor control effect. However, the method proposed in this invention, after acquiring torque data, can effectively distinguish between two different failure modes by analyzing the time domain and frequency domain characteristics respectively. Then, based on the different failure modes, targeted control strategies can be designed. Compared with existing methods, it can significantly improve the stability of silver powder preparation control.

[0024] Step S102: Determine the response lag time based on the time distribution benchmark deviation of the real-time torque fusion sequence; in the frequency domain, determine the corresponding torque roughness index based on the time-series fluctuation benchmark deviation of the torque value in the real-time torque fusion sequence.

[0025] During the reaction process of adding a silver-containing solution dropwise into the reactor, there are two states: normal and abnormal. In the normal state, the PVP dispersant molecular chains are fully extended in the good solvent. Under the action of the dispersant, the solid silver particles generated during the reduction reaction have not yet agglomerated into clusters. At this time, the fluid mainly exhibits viscous loss behavior. The resistance change felt by the stirring blades and the push pressure waveform of the feed pump are highly synchronized in time. That is, the real-time torque fusion sequence is highly synchronized with the reference torque sequence under no particle interference. In the abnormal state, when the local PVP concentration is too high or charge shielding causes the molecular chains to coil and entangle, the Weissenberg number of the system increases, and the fluid exhibits a significant elastic memory effect. This causes the stress transmission process to relax and lag in time relative to historical moments. That is, the stress that should have acted on the stirring blades will be delayed, resulting in a delay in the change of torque. As stirring continues, the coiling and entanglement of molecular chains will intensify, and this delay phenomenon will become more and more obvious, resulting in a significant time delay in the real-time torque fusion sequence compared to the reference torque sequence under no particle interference. Please see Figure 2 It shows a waveform diagram of a reference torque fusion sequence under normal conditions provided by an embodiment of the present invention; please refer to Figure 3 This diagram illustrates a waveform comparison between a real-time torque fusion sequence in a soft cluster state and a reference torque fusion sequence provided by an embodiment of the present invention. By comparing 2 and... Figure 3 It can be seen that the torque value (peak) in the soft cluster state lags behind the torque value (peak) in the normal state in terms of time. Therefore, in reality, there may be situations where the real-time torque fusion sequence lags behind the reference sequence in time. Thus, it is necessary to analyze the deviation of the time distribution of the real-time torque fusion sequence from the reference sequence, that is, to analyze the deviation of the stated time distribution from the reference sequence. The larger the deviation, the greater the time lag, and vice versa. This allows us to obtain the lag time of the real-time torque fusion sequence from the reference torque sequence, and to intuitively quantify the impact of uneven dispersant distribution on the stirring process.

[0026] In addition to the aforementioned abnormal state caused by uneven distribution (insufficient spreading) of PVP dispersant, another abnormal state exists due to the formation of micron-sized rigid silver powder agglomerates near the feed nozzle caused by insufficient mixing efficiency. These rigid heterogeneous particles alter the local microscopic turbulent structure and dissipation rate of the fluid. This change in the flow field microstructure leads to discontinuous high-frequency pulsations in the fluid's shear response. The impact on the agitator blade torque manifests as significant fluctuations in the torque variation range. Specifically, compared to the baseline torque sequence without particle interference, the real-time torque sequence exhibits larger and more drastic fluctuations in torque values ​​(in the frequency domain) over time. Please refer to [link to relevant documentation]. Figure 4It shows a waveform comparison diagram of a real-time torque fusion sequence and a reference torque fusion sequence in a hard cluster state provided by an embodiment of the present invention; comparison Figure 2 and Figure 4 It can be seen that the waveform of the real-time torque fusion sequence in the hard cluster state is less smooth than that of the reference torque sequence, meaning that the torque value fluctuates more violently. Therefore, it is necessary to analyze the deviation of the torque value fluctuation of the real-time torque sequence from the reference torque sequence, that is, to analyze the deviation of the torque value time-series fluctuation from the reference. The larger the deviation, the greater the torque fluctuation (the rougher the torque sequence change), and vice versa. The smaller the deviation, the smaller the torque fluctuation (the smoother the torque sequence change). This allows us to obtain the roughness index of the real-time torque fusion sequence compared to the reference torque sequence, which can accurately show the extent to which insufficient mixing efficiency near the feeding nozzle affects the stirring process.

[0027] Step S103: Based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in response lag time, determine the dynamic first threshold; in the frequency domain, based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in torque roughness index, determine the dynamic second threshold.

[0028] In step S102, the response lag time and torque roughness index were obtained. Both can quantify the abnormal state of the solution in the reactor. However, in reality, the production process allows for a certain degree of error (tolerance). That is, their influence on the reaction process is permissible before the corresponding threshold is reached. Therefore, it is necessary to obtain the thresholds corresponding to the response lag time and torque roughness index respectively to determine the minimum quantitative standard for the occurrence of abnormal states. This ensures production tolerance while avoiding excessive or delayed intervention, thus ensuring the stability of the production process.

[0029] As the silver-containing solution is continuously added during the stirring process, the content of solid silver particles in the reactor continuously increases, which in turn increases the impact on the stirring blades. Consequently, the response lag time and torque roughness index change continuously over time. Therefore, it is necessary to set dynamic thresholds for the response lag time and torque roughness index to more accurately determine the abnormal state of the solution and improve the control precision and stability of the silver powder preparation process.

[0030] It should be noted that since the response lag time and torque roughness index are obtained by comparing them with the benchmark torque sequence, the influence of individual equipment differences and actual errors has been eliminated. Therefore, when obtaining the dynamic threshold, the influence of individual equipment differences and actual errors also needs to be eliminated. Thus, when setting the dynamic threshold, it is necessary to obtain the dynamic threshold corresponding to the response lag time (torque roughness index) based on the overall deviation fluctuation of the benchmark initial torque sequence and the benchmark fused torque sequence in response lag time (or torque roughness index). This ensures that the dynamic threshold is decoupled from specific equipment characteristics and random errors, thereby establishing an objective and stable state judgment benchmark.

[0031] Step S104: Analyze the numerical relationships between the dynamic first threshold and the response lag time, and between the dynamic second threshold and the torque roughness index, respectively, to determine the silver paste stirring state, and perform silver powder preparation control according to the silver paste stirring state.

[0032] On the one hand, in step S102, the causes of the abnormal state of the silver paste (solution in the reactor) have been analyzed, namely, uneven distribution of the dispersant and insufficient mixing efficiency near the feeding nozzle, and the abnormal state has been quantified (response hysteresis time and torque roughness index). It should be noted that: uneven distribution of the dispersant will lead to excessively high local concentration of PVP molecular chains, resulting in entanglement and bridging, forming soft clusters with relatively weak binding force due to the binding of polymer chains; insufficient mixing efficiency near the feeding nozzle will lead to local oversaturation of the reducing agent, causing explosive nucleation and growth of silver nuclei, and silver particles will be directly bonded together by solid-state sintering or strong van der Waals forces, forming tightly bound hard clusters; therefore, response hysteresis time and torque roughness index can respectively represent the two states of soft clusters and hard clusters. On the other hand, the dynamic first threshold and dynamic second threshold obtained in step S103 correspond to the response hysteresis time and torque roughness index, respectively. By comparing their respective numerical relationships, the quantitative characteristics can be transformed into key steps for executable control decisions, realizing the accurate distinction and independent monitoring of the two micro-failure modes, and providing clear physical basis and criterion input for subsequent targeted regulation.

[0033] Since silver paste has two origins and two drastically different abnormal states (soft clusters and hard clusters), in order to accurately and effectively regulate these two states, different execution control strategies need to be set for soft clusters and hard clusters. This can achieve targeted regulation, avoid regulation failure, resource waste, or even state deterioration caused by mismatch between the control strategy and the failure mode, and significantly improve the stability and efficiency of the silver powder preparation process.

[0034] In summary: First, this invention performs deep decoupling analysis on the motor torque signal to obtain the response lag time and torque roughness index, achieving accurate quantification of the different microscopic characteristics of soft and hard clusters. This eliminates the masking of microscopic characteristics by macroscopic viscosity in high-solids-content systems. Second, by setting different dynamic thresholds for soft and hard clusters, it reduces the background noise fluctuations caused by wear of different batches of equipment and viscosity changes during the reaction stage, improving the accuracy of real-time differentiation between soft and hard clusters and benefiting long-term stable operation in industrial settings. Finally, by differentiating the different states of silver paste, different control strategies are formulated for soft and hard clusters, avoiding the traditional one-size-fits-all control and achieving targeted treatment for different states, significantly improving the stability and efficiency of high-dispersion silver powder preparation.

[0035] Preferably, in some possible implementations of the embodiments of the present invention, the method for calculating the real-time torque fusion sequence includes: In all real-time torque initial sequences, calculate the average value of the torque values ​​at the same index time, determine the initial fusion value at each index time, arrange all the initial fusion values, and determine the reference initial fusion sequence; The reference initial fusion sequence is standardized to determine the standard initial fusion value at each index time. The difference between the standard initial fusion value at each index time and a preset constant is calculated to determine the local real-time average torque at each index time. The local real-time average torques at all index times are arranged to determine the real-time torque fusion sequence. The process of standardizing the reference initial fusion sequence and then subtracting the standard initial fusion value from the preset constant constitutes the preprocessing process.

[0036] After entering the chemical reaction stage, as solid silver particles are generated, the background noise caused by stirring and turbulence gradually increases. In order to suppress this noise, it is necessary to calculate the average value of the torque values ​​at the same index time of all real-time torque initial sequences, that is, the initial fusion value at each index time. Since the torque sensor collects torque signals in a time sequence, the real-time torque initial sequence is also sorted by time. Therefore, all the initial fusion values ​​can be arranged in time sequence to obtain an initial fusion sequence under a reference. This can align and fuse multiple torque signals with the same index time, thereby providing a clear and stable analysis object for subsequent feature extraction (response hysteresis time and torque coarsening index).

[0037] It should be noted that the index time refers to the relative time of the actual sampling time in the sequence. That is, the index time and index value at the same position in the sequence are fixed, while the torque value is variable.

[0038] After obtaining the initial reference fusion sequence, it needs to be standardized. This is because during the chemical reduction process of photovoltaic silver paste, the solid content of the reaction system continuously increases, causing the overall macroscopic apparent viscosity of the paste to exhibit a non-linear drifting upward trend. This change in macroscopic viscosity causes an overall scaling of the stirring torque baseline value and fluctuation amplitude, thus masking the fine rheological characteristics caused by micro-feeding pulses. By standardizing to determine the standard initial fusion value at each index time, the dimension of rheological monitoring can be transformed from the absolute value of the amplitude to the shape of the amplitude (amplitude fluctuation). To provide a unified benchmark for subsequent calculations, the sequence is shifted as a whole by subtracting the standard initial fusion value at each index time from a preset constant, resulting in the local real-time average torque at each index time. This ensures that the minimum value in all sequences remains the same, eliminating the interference of large numerical ranges caused by standardization and establishing a unified and comparable benchmark data space for subsequent calculations. By rearranging the local real-time average torque at all index times, the real-time torque fusion sequence can be determined.

[0039] In a specific implementation of this invention, the specific calculation steps for the real-time torque fusion sequence are as follows: S1: The system creates a memory with a length of... The first-in-first-out buffer (e.g., set to 5) is defined as a stroke data sliding window queue. Each time the system starts or resets, it first enters the pre-filling stage, during which the system continuously collects the initial data. The initial real-time torque sequence for each stroke cycle is sequentially pushed into the queue. Until the queue is full, the system only performs data filling during this period; S2: After pre-filling is completed, the real-time monitoring cycle begins, and the system continuously monitors the pump injection trigger moment. Whenever the first This trigger, the current period duration is [duration value]. Real-time torque initial sequence The latest Push into stroke data sliding window queue At the tail of the queue, the oldest data at the head of the queue is removed. Keeping the queue length constant, the queue now stores the most recent items arranged in chronological order. A real-time torque initial sequence; S3: Perform ensemble averaging on the data in the queue and output the reference initial fusion sequence. ,for The first sampling points ( From 1 to the total number of samples per stroke The iterative calculation process for its numerical value is as follows: ,in, Indicates the number of elements in the queue. The initial real-time torque sequence at the ... Torque values ​​at each sampling point; S4: For those containing Reference initial fusion sequence of each sampling point Calculate the arithmetic mean of all its sampled values. with standard deviation Subsequently, on Each sampling point on ( Perform standardization operations to generate a standardized fusion sequence. Among these, to prevent the standard deviation from being measured under ideal laminar flow or extremely stable signal conditions. Approaching zero can lead to computational overflow or program crashes. This embodiment introduces a numerical stability factor into the denominator. Then the first The standardization formula for each sampling point is: ,in, Represent a very small positive integer (e.g., a value of 1000). ), used to prevent the denominator from being 0; S5: In the standardized fusion sequence Find the minimum value among all the sampled points. And generate a non-negative real-time torque fusion sequence. The calculation for each sampling point is as follows: ,in, Represents the standardized fusion sequence The i-th sampling point in the sample.

[0040] Preferably, in some possible implementations of the embodiments of the present invention, the method for calculating the response lag time includes: In the real-time torque fusion sequence, the local time interval value at each index time is determined by the product between the index value and the torque value at each index time; the local time interval values ​​at all index times are accumulated to obtain the cumulative time interval value; the torque values ​​at all index times are accumulated to determine the cumulative torque value in the real-time torque fusion sequence; the cumulative time interval value is used as the numerator and the cumulative torque value is used as the denominator to obtain the time centroid of the real-time torque fusion sequence. Based on the principle of obtaining the time centroid of the real-time torque fusion sequence, the time centroid of the reference torque fusion sequence is determined; the difference between the time centroid of the real-time torque fusion sequence and the time centroid of the reference torque fusion sequence is taken as the response lag time.

[0041] If soft clusters appear in the solution, it indicates that the real-time torque fusion sequence lags behind the baseline torque fusion sequence. A torque value that appears in the baseline torque fusion sequence will appear later in the real-time torque fusion sequence due to the lag, meaning it will appear at a later position in the real-time torque fusion sequence. Therefore, the product between this torque value and its corresponding index value will be larger, i.e., the time centroid will be larger. Thus, a larger time centroid indicates a higher probability of response lag, i.e., a higher probability of the presence of soft clusters. Conversely, a smaller time centroid indicates a lower probability of response lag, i.e., a lower probability of the presence of soft clusters.

[0042] In one specific implementation of this invention, the formula for calculating the time centroid of the real-time torque fusion sequence is: in, Indicates the time centroid of the real-time torque fusion sequence; This represents the index value at the i-th index time. Represents the real-time torque fusion sequence The torque value at the i-th index; L represents the real-time torque fusion sequence. The number of torque values.

[0043] It should be noted that index values ​​typically start from 0; in this embodiment, the index value... The sequence is incremented by 1 sequentially from the positive integer 1 until the last value in the sequence, in order to preserve the influence of the first torque value in the sequence on the time centroid; the numerator and denominator are added separately because all elements in the sequence are used to reflect the time centroid of the sequence. This section aims to combine time and torque information. The accumulation process is a weighted summation process. This is because, in the reduction reaction of silver paste, if the polymer chains become entangled (soft agglomerate), the elasticity of the fluid will cause the torque response sequence to be delayed along the time axis. Simply comparing the peak value of the sequence or the time position of a certain point is easily affected by noise and is not robust. The denominator uses... The reason is that the weighting process on the numerator is normalized to eliminate the influence of the absolute intensity of the torque signal on the calculation results, thus making the time centroid a dimensionless relative characteristic value that purely characterizes the time distribution of torque. It should be noted that when the denominator is 0, it indicates that the reaction system is in an ideal state of extreme homogeneity and perfectly indeterminate stirring flow field. In this case, the denominator can be replaced with a very small constant, such as... This is to ensure computational stability.

[0044] Based on the calculation principle of the time centroid of the real-time torque fusion sequence, the time centroid of the reference torque fusion sequence can be obtained by simply replacing the real-time torque fusion sequence with the reference torque fusion sequence and repeating the calculation formula of the time centroid of the real-time torque fusion sequence.

[0045] Since the real-time torque fusion sequence may lag behind the reference torque fusion sequence in response, the time centroid of the real-time torque fusion sequence will be larger than that of the reference torque fusion sequence when a response delay occurs. The difference between the two reflects the degree of lag in the response; the larger the difference, the more delayed the response, and the smaller the difference, the closer the response is to the reference response. Therefore, in a specific implementation of this invention, the response lag time can be expressed by the following formula: in, Indicates the response lag time; Indicates the time centroid of the real-time torque fusion sequence; This indicates the time centroid of the reference torque fusion sequence. It can intuitively represent the degree of elastic relaxation of the current fluid medium relative to the pure solvent reference. A significantly positive value indicates that the dispersant molecular chain network has undergone an elastic energy storage and release process in response to the pulse energy, suggesting that the molecular chain conformation may be coiled; conversely, a significantly negative value indicates that the dispersant molecular chain is normally stretched and no coiling phenomenon has occurred.

[0046] Furthermore, in a specific implementation of this invention, the method for obtaining the reference torque fusion sequence is as follows: according to the principle of obtaining the real-time torque fusion sequence, simply replace the real-time torque initial sequence with the reference torque initial sequence, and repeat steps S3 to S5 of the calculation steps for the real-time torque fusion sequence to obtain the reference torque fusion sequence; wherein, the number of reference torque initial sequences obtained is 20.

[0047] Preferably, in some possible implementations of the embodiments of the present invention, the method for calculating the torque roughness index includes: In the real-time torque fusion sequence, the second-order difference value of the torque value at each index time is calculated, the second-order difference value at all index times is arranged to determine the corresponding second-order difference sequence, the L1 norm of the second-order difference sequence is calculated, and the L1 norm is used as the periodic feature value of the real-time torque fusion sequence on the torque value. Based on the principle of obtaining the periodic characteristic value of the real-time torque fusion sequence on the torque value, the periodic characteristic value of the reference torque fusion sequence on the torque value is determined; the difference between the periodic characteristic value of the real-time torque fusion sequence on the torque value and the periodic characteristic value of the reference torque fusion sequence on the torque value is used as the torque roughness index.

[0048] When micron-sized rigid silver powder agglomerates are generated near the feeding nozzle due to insufficient mixing efficiency, the change period of the torque value in the original state will be disrupted. Therefore, the generation state of the hard agglomerates can be reflected by the difference between the periodic characteristics of the torque value in the real-time torque fusion sequence and the reference torque fusion sequence. Since the second-order difference can quantify the periodic characteristics, the second-order difference value is used to represent the periodic characteristics. By calculating the L1 norm, the cumulative intensity of all high-frequency disturbances caused by rigid particles in the real-time torque fusion sequence can be represented, which can more comprehensively and stably reflect the overall severity of the hard agglomerates.

[0049] In a specific implementation of this invention, the formula for calculating the periodic characteristic value of the real-time torque fusion sequence on the torque value is as follows: in, This represents the periodic characteristic value of the real-time torque fusion sequence on the torque value; This represents the second-order difference value at the i-th point in the real-time torque fusion sequence; , and These represent the (i+1)th, ith, and (i-1)th torque values ​​in the real-time torque fusion sequence, respectively. represents the absolute value function; L represents the number of torque values ​​in the real-time torque fusion sequence.

[0050] It should be noted that the reason why i takes values ​​from 2 to L-1 is that the first and last elements in the sequence cannot be used to calculate the second-order difference value, so these two elements are removed.

[0051] Based on the calculation principle of the formula for calculating the periodic characteristic value of the real-time torque fusion sequence on the torque value, the periodic characteristic value of the reference torque fusion sequence can be obtained by simply replacing the real-time torque fusion sequence with the reference torque fusion sequence and repeating the formula for calculating the periodic characteristic value of the real-time torque fusion sequence.

[0052] Since the real-time torque fusion sequence may lag behind the reference torque fusion sequence, the periodic characteristic value of the real-time torque fusion sequence will be greater than that of the reference torque fusion sequence when a response delay occurs. The difference between the two can reflect the roughness of the torque; the larger the difference, the rougher the torque, and the smaller the difference, the closer the torque is to the reference torque. Therefore, in a specific implementation of this invention, the torque roughness index can be expressed by the following formula: in, Indicates the torque roughness index; This represents the periodic characteristic value of the real-time torque fusion sequence on the torque value; This represents the periodic characteristic value of the reference torque fusion sequence on the torque value; The high-frequency "spurs" in the rheological response waveform can be physically quantified. A sudden increase in the value indicates that the microscopic continuity of the flow field is disrupted by rigid entities, suggesting an extremely high risk of hard agglomeration.

[0053] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the dynamic first threshold includes: In the initial sequence of the reference torque, the time centroid of the initial sequence of the reference torque is determined based on the principle of obtaining the time centroid of the real-time torque fusion sequence. The time fluctuation value is determined based on the relative deviation between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence. The real-time torque mean and the reference torque mean of the real-time torque fusion sequence and the reference torque fusion sequence are obtained respectively. Based on the real-time torque mean and the reference torque mean, the comparison feature value between the real-time torque mean and the reference torque mean is determined. The dynamic first threshold is obtained by using the product of the time fluctuation value and the comparison feature value as a basis, combined with the first sensitivity coefficient and the first empirical compensation index for correction.

[0054] As the chemical reaction proceeds, the solid content and viscosity of the slurry continuously increase. According to the principles of fluid mechanics, the fluid resistance torque on the stirring blades increases, and the micro-vibration of the mechanical transmission chain (gearbox, bearings) and the intensity of turbulent pulsation of the fluid itself are also amplified. If the time fluctuation value under low viscosity solvent is directly used as the judgment threshold under high viscosity conditions, it is very easy to cause "false positive" misjudgment (that is, normal high viscosity background noise is misjudged as abnormal). Therefore, it is necessary to use the real-time viscosity of silver paste at different stirring times (represented by comparative characteristic values) to dynamically compensate the judgment threshold, which improves the adaptive ability and full-cycle robustness of the system state judgment, and ensures that under complex conditions of high viscosity and strong noise, the control decision can still accurately focus on the real micro-failure signal, rather than being misled by macro-process changes.

[0055] In one specific implementation of this invention, the formula for calculating the dynamic first threshold is: in, Indicates the dynamic first threshold; This represents the statistical sensitivity coefficient (e.g., a value of 3 corresponds to a normal distribution). Confidence interval); Indicates the time fluctuation value; This represents the average real-time torque in the real-time torque fusion sequence; This represents the mean of the reference torque in the reference torque fusion sequence; This represents the experience compensation index (ranging from 0.5 to 1.0, depending on the mechanical characteristics of the equipment; in this embodiment, it is taken as a positive integer of 1). It should be noted that, This reflects the dispersion of multiple time centroid measurements under ideal reference conditions of pure solvent and no particles, due to inherent random noise (such as electrical noise and micromechanical vibration). The existence of this allows for the establishment of a correlation between the dynamic first threshold and the response lag time; and The torque values ​​were obtained by calculating the average values ​​from the real-time torque fusion sequence and the reference torque fusion sequence, respectively. This term aims to achieve adaptive scaling of the dynamic first threshold, representing the real-time viscosity of the current reaction system relative to a macroscopic relative viscosity of a reference, which is due to... It is unchanging, and The ratio changes as the reaction proceeds, and is positively correlated with the real-time viscosity. The larger the ratio, the larger the real-time viscosity and the larger the dynamic first threshold. Conversely, the smaller the ratio, the smaller the real-time viscosity and the smaller the dynamic first threshold. This allows for the construction of a judgment threshold that can be dynamically adjusted to follow the reaction process. The aim is to reduce the probability of normal data being misclassified as abnormal data. Taking 3 can reduce the false positive rate to ; The aim is to change the impact of increased real-time viscosity on system noise (i.e. The degree of amplified nonlinear effects, Take the positive integer 1 to represent and They are directly proportional. Taking a value less than a positive integer 1 indicates a pair The amplification speed varies with The growth rate is slower than linear growth, which may be due to the damping effect of the mechanical system, causing vibrations to be partially absorbed. Taking a value greater than a positive integer 1 indicates a pair The amplification speed varies with The growth rate is faster than linear growth, which may be due to resonance or nonlinear rheological behavior, and can be adjusted according to the specific implementation scenario.

[0056] Preferably, in some possible implementations of the embodiments of the present invention, the method for calculating the time fluctuation value includes: Calculate the difference between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence to determine the response lag time corresponding to each initial reference torque sequence. Based on the standard deviation of the response lag times corresponding to all initial reference torque sequences, determine the time fluctuation value.

[0057] When setting the threshold, physical equipment and real-world factors must also be considered. The influence of physical equipment and real-world factors must be eliminated. That is, the threshold must be set under the premise of a benchmark. Therefore, a time fluctuation value can be obtained by the relative deviation (i.e., standard deviation) between the time centroid of multiple benchmark torque initial sequences and the time centroid of the benchmark torque fusion sequence. This can minimize the influence of physical equipment and real-world factors, improve the purity and representativeness of the threshold base, and make it contain only the inherent, uneliminable random noise information of the system, thus establishing a judgment standard that is independent of specific equipment and random errors.

[0058] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the dynamic second threshold includes: In the initial sequence of the reference torque, the periodic characteristic value of the initial sequence of the reference torque is determined based on the principle of obtaining the periodic characteristic value of the torque value in the real-time torque fusion sequence. Calculate the difference between the periodic characteristic value of each initial reference torque sequence and the periodic characteristic value of the fused reference torque sequence, determine the torque roughness index corresponding to each initial reference torque sequence, and determine the roughness fluctuation value based on the standard deviation of the torque roughness index corresponding to all initial reference torque sequences. The dynamic second threshold is determined based on the roughness fluctuation value and the comparison feature value.

[0059] The principle for obtaining the dynamic second threshold is the same as that for obtaining the dynamic first threshold. Therefore, based on the principle for obtaining the dynamic first threshold, the dynamic second threshold can be obtained simply by replacing the time fluctuation value with the roughness fluctuation value.

[0060] In this embodiment of the invention, the formula for calculating the dynamic second threshold is: in, Indicates a dynamic second threshold; This represents the statistical sensitivity coefficient (e.g., a value of 3 corresponds to a normal distribution). Confidence interval); This represents the roughness fluctuation value; This represents the real-time average torque. This represents the average baseline torque. This represents the experience compensation index (ranging from 0.5 to 1.0, depending on the mechanical characteristics of the equipment; in this embodiment, it is a positive integer of 0.5).

[0061] Preferably, in some possible implementations of the embodiments of the present invention, the method for determining the dynamic second threshold based on the roughness fluctuation value and the comparison feature value includes: The dynamic second threshold is obtained by using the product of the roughness fluctuation value and the contrast feature value as a basis, combined with the second sensitivity coefficient and the second empirical compensation index for correction.

[0062] Due to the differences in real-world equipment, there are both linear and nonlinear relationships between roughness fluctuation values ​​and contrast feature values. If the product of the two is simply used as the dynamic second threshold, it cannot adapt to the mechanical-fluid coupling characteristics and complex noise transmission laws of all equipment types. This may result in the dynamic second threshold being too lenient (missed judgment) or too sensitive (false judgment) on other equipment. Therefore, it is necessary to introduce a second sensitivity coefficient and a second empirical compensation index for correction in order to achieve the adaptability of the dynamic second threshold setting.

[0063] In one specific implementation of this invention, the second sensitivity coefficient Setting it to 3 can reduce the false positive rate to If Setting it to 2 will reduce the false positive rate to 2. The first is designed to reduce the probability of misclassifying normal data as abnormal data; the second is the empirical compensation index. The aim is to alter the degree of nonlinear influence of increasing the contrast eigenvalue on amplifying the roughness fluctuation value. Take the positive integer 1 to represent and They are directly proportional. Taking a value less than a positive integer 1 indicates a pair The amplification speed varies with The growth rate is slower than linear growth because the damping effect of the mechanical system partially absorbs the vibration. Taking a value greater than a positive integer 1 indicates a pair The amplification speed varies with The growth rate is faster than linear growth, which is due to resonant or nonlinear rheological behavior; the second sensitivity coefficient and the second empirical compensation index can be adjusted automatically according to the specific implementation scenario.

[0064] Preferably, in some possible implementations of the embodiments of the present invention, the method for determining the stirring state of the silver paste includes: If the torque roughness index is greater than the dynamic second threshold, the silver paste is determined to be in the hard cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is greater than the dynamic first threshold, then the silver paste is determined to be in the soft cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is less than or equal to the dynamic first threshold, then the silver paste is determined to be in an ideal stable state.

[0065] Since molecular chain entanglement mainly causes an overall shift in the waveform in the time domain (low-frequency characteristics), while rigid hard aggregation mainly causes local high-frequency oscillations in the frequency domain (high-frequency characteristics), these two physical phenomena have natural orthogonality in signal processing. Therefore, this embodiment of the invention adopts a dual-determination logic. In a specific implementation of this embodiment, the determination method is as follows: First priority determination: If Then regardless Based on the numerical value, the system determines that the current system is in the stage dominated by hard agglomeration formation; this is because rigid agglomerates have irreversible destructive effects on product quality and will rapidly deteriorate the flow field once formed, thus having the highest judgment priority.

[0066] Second priority determination: If and At this point, hard agglomeration was not triggered, but the system determined that the current system was in the dispersant conformation failure (soft agglomeration) stage. At this point, the fluid medium remained microscopically continuous and no large rigid particles were generated, but the polymer chains coiled and entangled, and the elastic effect was significant.

[0067] Third priority determination: If and At this point, the system is in an ideally dispersed state, and its microstructure is stable.

[0068] By employing a triple-judgment system, the accuracy, reliability, and security of system decisions are improved, enabling targeted identification and tiered management of different failure modes.

[0069] Preferably, in some possible implementations of the embodiments of the present invention, the method for controlling the preparation of silver powder includes: If the silver paste is in the hard cluster formation stage, physical crushing of the silver paste is used for regulation. If the silver paste is in the soft cluster formation stage, a chemical regulator is added to the silver paste to control it. If the silver paste is in an ideal and stable state, then maintain the current working state without adjustment.

[0070] Since the underlying mechanisms for the formation of soft and hard clusters in silver paste are different, a single regulatory measure cannot adequately adapt to all states. Therefore, different regulatory strategies need to be set for soft and hard clusters, achieving mechanism-based targeted regulation and significantly improving the accuracy and effectiveness of regulation.

[0071] The formation of soft clusters is due to the physical curling of dispersant molecular chains caused by local concentration polarization or charge shielding effect. Therefore, the pH value of silver paste can be changed by chemical regulation to change the charge distribution state of dispersant molecular chains and particle surfaces, thereby increasing the electrostatic repulsion force of molecular chain segments and promoting the molecular chains to unfold again through electrostatic repulsion.

[0072] The formation of hard clusters is due to the generation of micron-sized rigid silver powder agglomerates near the feed nozzle. At this time, the fluid shear force provided by the conventional continuous and stable feeding mode is insufficient to break these tightly bound agglomerates. Therefore, a larger shear force is required to physically break the hard agglomerates and ensure that the hard clusters can be fully physically broken.

[0073] In one specific implementation of this invention, the method for controlling the preparation of silver powder is as follows: If the silver paste is in the hard cluster formation stage, the system sends a command to the feed pump controller to switch the pump's operating mode from "continuous flow mode" to "intermittent high-frequency pulse mode." This mode switching can be achieved by changing the motor drive waveform. In this mode, the feed pump performs a "transient impact-emergency stop" cycle. During the transient impact phase, the pump is controlled to operate at maximum rated power (or a higher rate, depending on equipment requirements) at extremely high speed. During the emergency stop phase (e.g., 0.2 seconds), the rapid acceleration of the pump rotor drives the fluid to instantly reach a high velocity within the pipeline; during the emergency stop phase, the pump is immediately stopped. Time (e.g., 1 second). Through this rapid start-stop operation, utilizing the water hammer effect in fluid dynamics, an extremely high-intensity instantaneous pressure wave and microscopic turbulent shear field are generated at the feed pipe outlet. This instantaneous impact force is far higher than the shear force of conventional continuous flow, effectively breaking up the nascent hard clusters generated at the feed nozzle and preventing their diffusion into the entire reactor. The system continuously executes the above pulse action a preset number of times (e.g., 10 cycles) to ensure sufficient physical fragmentation of the local high-concentration area.

[0074] During the physical crushing of hard clusters, the operating frequency, flow characteristics, and pulse waveform of the feed pump are completely different from those of the conventional continuous flow mode. Therefore, the torque response data collected at this time no longer conforms to the baseline torque characteristics. If this data continues to be included in the stroke data sliding window queue... Performing calculations in this way can lead to data corruption and misjudgments. Therefore, when performing physical fragmentation of hard clusters, it is necessary to protect the data simultaneously, as follows: Upon triggering a physical high-frequency fracture, the system immediately executes a monitoring and fracturing procedure: forcibly pausing torque sequence acquisition and status determination. Simultaneously, the system sends a command to immediately clear the current stroke data sliding window queue. Clear all historical data from the old state; after the physical high-frequency crushing cycle ends, the system controls the feed pump to automatically switch back to the conventional continuous flow mode. At this time, the system does not immediately resume monitoring, but enters a recovery waiting phase, waiting for the feed pump to complete. The next (i.e., queue length) consecutive injection action, pending stroke data sliding window queue After the valid data under the current continuous flow state is refilled, the monitoring circuit breaker is automatically deactivated, and the real-time decoupled analysis of the rheological characteristics is restored. This ensures that the closed-loop control system always makes decisions based on isomorphic and reliable data, avoiding transient interference during mode switching.

[0075] If the silver paste is in the soft cluster formation stage, the system controls the start of the pH adjuster micro-metering pump connected to the reactor. The type of adjuster is pre-selected according to the characteristics of the dispersant (e.g., dilute ammonia for PVP systems). The system performs a single pulse-type droplet addition, with the addition amount set to a very small proportion of the total mass of the reaction system (e.g., 0.005%). After this micro-adjuster enters the system, it can rapidly change the charge distribution state of the dispersant molecular chains and particle surfaces, increasing the electrostatic repulsion between molecular chain segments. After the droplet addition is completed, the system enters a mixing lock state, the duration of which is set to... (e.g., 30 seconds) During this period, the system maintains regular stirring and feeding, but temporarily disables the function of judging the state of the silver paste. This is to allow time for the regulator to disperse evenly and for the molecular chains to physically expand, avoiding excessive addition due to regulation lag; after the lockout ends, the system resumes real-time monitoring. If the next cycle... If the temperature falls back to within the threshold range, it indicates that the regulation is effective; if it still exceeds the limit, the system will automatically trigger the next dripping cycle.

[0076] If the silver paste is in an ideal and stable state, then maintain the current working state without making any adjustments.

Claims

1. A method for preparing and controlling highly dispersible ultrafine silver powder for photovoltaic silver paste, characterized in that, The method includes: During the silver paste stirring process, all real-time torque initial sequences of the motor stirring were obtained from the reactor; all real-time torque initial sequences were fused and preprocessed to determine the real-time torque fused sequence; all reference torque initial sequences and reference torque fused sequences were obtained. The response lag time is determined based on the time distribution benchmark deviation of the real-time torque fusion sequence; in the frequency domain, the corresponding torque roughness index is determined based on the time-series fluctuation benchmark deviation of the torque value in the real-time torque fusion sequence. Based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in response lag time, a dynamic first threshold is determined; in the frequency domain, based on the overall deviation fluctuation of the initial reference torque sequence and the fused reference torque sequence in torque roughness index, a dynamic second threshold is determined. The numerical relationships between the dynamic first threshold and the response lag time, and between the dynamic second threshold and the torque roughness index are analyzed respectively to determine the silver paste stirring state. Based on the silver paste stirring state, silver powder preparation control is implemented. The method for determining the stirring state of the silver paste includes: If the torque roughness index is greater than the dynamic second threshold, the silver paste is determined to be in the hard cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is greater than the dynamic first threshold, then the silver paste is determined to be in the soft cluster formation stage. If the torque roughness index is less than or equal to the dynamic second threshold and the response hysteresis time is less than or equal to the dynamic first threshold, then the silver paste is determined to be in an ideal stable state.

2. The method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to claim 1, characterized in that, The calculation method for the real-time torque fusion sequence includes: In all real-time torque initial sequences, calculate the average value of the torque values ​​at the same index time, determine the initial fusion value at each index time, arrange all the initial fusion values, and determine the reference initial fusion sequence; The reference initial fusion sequence is standardized to determine the standard initial fusion value at each index time. The difference between the standard initial fusion value at each index time and a preset constant is calculated to determine the local real-time average torque at each index time. The local real-time average torque at all index times is arranged to determine the real-time torque fusion sequence.

3. The method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to claim 1, characterized in that, The method for calculating the response lag time includes: In the real-time torque fusion sequence, the local time interval value at each index time is determined by the product between the index value and the torque value at each index time; the local time interval values ​​at all index times are accumulated to obtain the cumulative time interval value; the torque values ​​at all index times are accumulated to determine the cumulative torque value in the real-time torque fusion sequence; the cumulative time interval value is used as the numerator and the cumulative torque value is used as the denominator to obtain the time centroid of the real-time torque fusion sequence. Based on the principle of obtaining the time centroid of the real-time torque fusion sequence, the time centroid of the reference torque fusion sequence is determined; the difference between the time centroid of the real-time torque fusion sequence and the time centroid of the reference torque fusion sequence is taken as the response lag time.

4. The method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to claim 3, characterized in that, The method for calculating the torque roughness index includes: In the real-time torque fusion sequence, the second-order difference value of the torque value at each index time is calculated, the second-order difference value at all index times is arranged to determine the corresponding second-order difference sequence, the L1 norm of the second-order difference sequence is calculated, and the L1 norm is used as the periodic feature value of the real-time torque fusion sequence on the torque value. Based on the principle of obtaining the periodic characteristic value of the real-time torque fusion sequence on the torque value, the periodic characteristic value of the reference torque fusion sequence on the torque value is determined; the difference between the periodic characteristic value of the real-time torque fusion sequence on the torque value and the periodic characteristic value of the reference torque fusion sequence on the torque value is used as the torque roughness index.

5. The method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to claim 4, characterized in that, The process of obtaining the dynamic first threshold includes: In the initial sequence of the reference torque, the time centroid of the initial sequence of the reference torque is determined based on the principle of obtaining the time centroid of the real-time torque fusion sequence. The time fluctuation value is determined based on the relative deviation between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence. The real-time torque mean and the reference torque mean of the real-time torque fusion sequence and the reference torque fusion sequence are obtained respectively. Based on the real-time torque mean and the reference torque mean, the comparison feature value between the real-time torque mean and the reference torque mean is determined. The dynamic first threshold is obtained by using the product of the time fluctuation value and the comparison feature value as a basis, combined with the first sensitivity coefficient and the first empirical compensation index for correction.

6. The method for preparing and controlling highly dispersible ultrafine silver powder for photovoltaic silver paste according to claim 5, characterized in that, The method for determining the time fluctuation value includes: Calculate the difference between the time centroid of each initial reference torque sequence and the time centroid of the fused reference torque sequence to determine the response lag time corresponding to each initial reference torque sequence. Based on the standard deviation of the response lag times corresponding to all initial reference torque sequences, determine the time fluctuation value.

7. The method for preparing and controlling highly dispersible ultrafine silver powder for photovoltaic silver paste according to claim 5, characterized in that, The process of obtaining the dynamic second threshold includes: In the initial sequence of the reference torque, the periodic characteristic value of the initial sequence of the reference torque is determined based on the principle of obtaining the periodic characteristic value of the torque value in the real-time torque fusion sequence. Calculate the difference between the periodic characteristic value of each initial reference torque sequence and the periodic characteristic value of the fused reference torque sequence, determine the torque roughness index corresponding to each initial reference torque sequence, and determine the roughness fluctuation value based on the standard deviation of the torque roughness index corresponding to all initial reference torque sequences. The dynamic second threshold is determined based on the roughness fluctuation value and the comparison feature value.

8. The method for preparing and controlling highly dispersible ultrafine silver powder for photovoltaic silver paste according to claim 7, characterized in that, The method for determining the dynamic second threshold based on the roughness fluctuation value and the comparison feature value includes: The dynamic second threshold is obtained by using the product of the roughness fluctuation value and the contrast feature value as a basis, combined with the second sensitivity coefficient and the second empirical compensation index for correction.

9. The method for preparing and controlling highly dispersed ultrafine silver powder for photovoltaic silver paste according to claim 1, characterized in that, The method for controlling the preparation of silver powder includes: If the silver paste is in the hard cluster formation stage, physical crushing of the silver paste is used for regulation. If the silver paste is in the soft cluster formation stage, a chemical regulator is added to the silver paste to control it. If the silver paste is in an ideal and stable state, then maintain the current working state without adjustment.

Images