Veterinary drug powder production abnormal intelligent diagnosis method and system

Abstract

The application discloses an abnormal intelligent diagnosis method and system for veterinary powder production, and particularly relates to the field of veterinary powder production condition monitoring and abnormal diagnosis, and is used for solving the problem that micro-flow state deterioration in powder production is difficult to be found early and actively corrected. First, the electroacoustic and temperature and humidity signals of a mixer are synchronously collected, torque current is extracted based on a park transformation, and a condition fusion matrix aligned in time sequence is constructed. Then, an acoustic frequency band envelope and torque current are cross-mapped to construct an acoustic-electric impedance ratio sequence, a distortion node is extracted to reconstruct a flow state impedance evolution sequence to highlight powder micro-phase change. The evolution sequence is nonlinearly aligned with a reference track through a dynamic time warping algorithm, an anchor point deviating from time is locked, and a distortion feature is extracted to form a diagnosis mark. Finally, a condensation gradient is calculated based on anchor point backtracking temperature and humidity, a control instruction is issued under a linkage strategy library, and a closed-loop system from state perception to adaptive repair is constructed, thereby providing scientific guarantee for high-quality veterinary production.

Description

Technical Field

[0001] This invention relates to the field of monitoring and diagnosing abnormalities in veterinary drug powder production, specifically to an intelligent diagnostic method and system for abnormalities in veterinary drug powder production. Background Technology

[0002] As a key material for ensuring livestock and poultry health and maintaining the stable development of animal husbandry, the uniformity of mixing and the physicochemical stability during the production process of veterinary drug powders directly determine their efficacy and safety. In the continuous production process of powder feeding, crushing, mixing, and packaging, materials are highly susceptible to changes in the workshop's microenvironment and fluctuations in equipment operating conditions. With the advancement of intelligent manufacturing, industrial sites place extremely high demands on the continuity, stability, and real-time intervention of anomalies in the powder production process. Accurately capturing and promptly intervening in minor deteriorations during production has become a core requirement for ensuring the quality of veterinary drug powders.

[0003] However, existing monitoring methods for powder production processes mainly rely on setting threshold alarms for macroscopic electrical parameters such as spindle motor current and stirring speed. The fundamental flaw of this method is that changes in macroscopic electrical parameters lag significantly behind changes in the microscopic state of the powder. When fluctuations in ambient temperature and humidity cause powder particles to absorb moisture and become sticky, leading to early capillary bridging and micro-agglomeration, the overall work load of the motor changes extremely slightly, and conventional thresholds fail to trigger alarms. Only when powder agglomerates to form large-scale dead zones and stagnant material will the current exhibit sudden, excessive changes, but by then the entire batch of material has already undergone irreversible mixing and must be scrapped. Furthermore, fluctuations in a single electrical parameter cannot decouple mechanical load from material load. The system cannot distinguish whether the increased resistance is due to density differences in the material itself, mechanical wear of the shaft bearings, or powder adhering to the walls due to environmental humidity. This information aliasing results in a high false alarm rate for existing methods and completely lacks the ability to trace physical root causes and for equipment to proactively compensate and correct deviations. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent diagnostic method and system for abnormalities in veterinary drug powder production, thus solving the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent diagnostic method for abnormalities in veterinary drug powder production, comprising the following steps: S1, synchronously acquiring the three-phase transient electrical parameter sequence of the powder mixer spindle, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port; performing Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence; and aligning the torque current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence according to the timestamp sequence to construct a mixing condition fusion matrix; S2, extracting the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence from the mixing condition fusion matrix, and performing time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; and extracting the trough distortion nodes of the acoustic impedance ratio sequence within a time sliding window. S3. Construct a powder flow impedance evolution sequence; S4. Call the preset benchmark batch flow impedance sequence, and use the dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence, and calculate the time warping offset; when the cumulative mutation rate of the time warping offset exceeds the preset threshold range, extract the acoustic and electrical distortion feature vector corresponding to the deviation from the starting time anchor point, and construct a deterioration diagnosis label; S5. Based on the deviation from the starting time anchor point in the deterioration diagnosis label, extract the temperature and humidity sequence in the corresponding traceability window in the mixing condition fusion matrix, and calculate the microenvironment humid heat condensation gradient; input the microenvironment humid heat condensation gradient and the acoustic and electrical distortion feature vector into the preset flow phase change compensation strategy library for matching, generate the equipment control instruction set and send it to the programmable logic controller.

[0006] Furthermore, the process of simultaneously acquiring the three-phase transient electrical parameter sequence of the powder mixer main shaft, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port, and performing Parker transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence is as follows: The three-phase transient electrical parameter sequence is acquired through a Hall sensor array; the high-frequency acoustic emission waveform sequence is acquired through a piezo-acoustic sensor; and the temperature and humidity sequence of the feeding port is acquired through a high-frequency response temperature and humidity sensor. Phase-locked loop frequency tracking is performed on the three-phase transient electrical parameter sequence to extract the fundamental phase angle of the stator voltage. Based on the fundamental phase angle of the stator voltage, a Clarke transform is performed on the three-phase transient electrical parameter sequence to generate a two-phase stationary orthogonal stator current sequence. A rotating coordinate system transformation is performed on the two-phase stationary orthogonal stator current sequence to separate the excitation current component sequence and the torque current component sequence.

[0007] Furthermore, the specific process of aligning the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence according to the timestamp sequence to construct the mixing condition fusion matrix is ​​as follows: extract the hardware system timestamps of the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence, and set the sampling frequency of the torque current component sequence as the reference resampling frequency; perform downsampling anti-aliasing filtering on the high-frequency acoustic emission waveform sequence based on the reference resampling frequency, perform spline interpolation fitting on the temperature and humidity sequence, output the synchronous time sequence with the same frequency, perform tensor splicing operation on the synchronous time sequence with the same frequency according to the signal dimension features, and output the mixing condition fusion matrix.

[0008] Furthermore, the specific process of extracting the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the fusion matrix of the mixing condition and constructing the acoustic impedance ratio sequence by performing time-frequency domain cross-mapping with the torque current component sequence is as follows: Perform wavelet packet decomposition on the high-frequency acoustic emission waveform sequence to extract the high-frequency detail coefficients characterizing the frequency band of powder particle collision and friction response; perform Hilbert transform demodulation on the high-frequency detail coefficients to extract the energy envelope and generate the characteristic frequency band envelope sequence; perform maximum-minimum normalized scaling mapping on the characteristic frequency band envelope sequence and the torque current component sequence; use the scale-mapped torque current component sequence as the load response variable and the characteristic frequency band envelope sequence as the acoustic dissipation variable; perform dissipation load ratio feature mapping extraction to output the acoustic impedance ratio sequence.

[0009] Furthermore, the specific process of extracting the trough distortion nodes of the acoustic impedance ratio sequence within the time sliding window and constructing the powder flow impedance evolution sequence is as follows: a moving data sliding window is constructed along the time axis, and morphological cap transformation processing is performed on the acoustic impedance ratio sequence within the moving data sliding window to extract the abnormal concave trough nodes of the morphological cap transformation processed sequence; the waveform deformation curvature feature vectors before and after the abnormal concave trough nodes are extracted, and free fluctuation nodes are filtered according to the preset curvature gradient reference interval to extract the trough distortion nodes; the trough distortion nodes are arranged in order along the time axis, and dynamic impedance trend envelope reconstruction is performed to output the powder flow impedance evolution sequence.

[0010] Furthermore, a preset benchmark batch flow impedance sequence is invoked, and a dynamic time warping algorithm is used to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence. The specific process for calculating the time warping offset is as follows: Construct a state evolution difference mapping matrix between the powder flow impedance evolution sequence and the benchmark batch flow impedance sequence; apply a dynamic programming algorithm to the state evolution difference mapping matrix to search for the optimal state matching path that satisfies the endpoint binding and continuity constraints, and extract the multidimensional state space mapping feature vectors of each aligned node pair on the optimal state matching path; perform cumulative feature extraction of evolution trajectory deviation based on the multidimensional state space mapping feature vectors, and output the time warping offset.

[0011] Furthermore, when the cumulative mutation rate of the time-order regularization offset exceeds the preset threshold range, the specific process of extracting the acoustic-electric distortion feature vector corresponding to the deviation from the starting time anchor point and constructing the degradation diagnosis label is as follows: Extract the local variation rate feature sequence of the time-order regularization offset along the time axis, construct the mutation rate trend curve, perform boundary cross-judgment on the mutation rate trend curve, and mark the first jump node that breaks through the upper limit of the preset threshold range as the deviation from the starting time anchor point; extract the neighborhood time truncation window corresponding to the deviation from the starting time anchor point, extract the working condition evolution time slice in the powder flow impedance evolution sequence, perform flow feature decoupling and separation on the working condition evolution time slice to extract the core deformation component, construct the acoustic-electric distortion feature vector, and package and reassemble the acoustic-electric distortion feature vector with the deviation from the starting time anchor point to construct the degradation diagnosis label.

[0012] Furthermore, based on the deviation from the starting time anchor point in the deterioration diagnosis identifier, the temperature and humidity sequence within the corresponding traceability window is extracted from the fusion matrix of the mixing condition, and the specific process for calculating the microenvironmental humid heat condensation gradient is as follows: The deterioration diagnosis identifier is parsed to obtain the deviation from the starting time anchor point; the temperature and humidity sequence with a preset span preceding the deviation from the starting time anchor point is extracted in reverse along the time axis to form a traceability data slice; the temperature waveform and relative humidity waveform in the traceability data slice are separated, and dew point temperature nonlinear fitting is performed in combination with the saturated water vapor pressure parameter to generate a near-wall condensation critical characteristic curve; the condensation phase change approximation rate characteristics of the near-wall condensation critical characteristic curve and the actual temperature waveform are extracted, and the microenvironmental humid heat condensation gradient is output.

[0013] Furthermore, the specific process of matching the microenvironmental humid heat condensation gradient with the acoustic and electrical distortion feature vector into a preset flow phase change compensation strategy library to generate a set of equipment control instructions and send them to the programmable logic controller is as follows: Perform multi-dimensional operating condition feature fusion reconstruction on the microenvironmental humid heat condensation gradient and the acoustic and electrical distortion feature vector to construct multi-dimensional state addressing encoding. Perform operating condition similarity feature optimization matching of multi-dimensional state addressing encoding in the flow phase change compensation strategy library to match the target compensation control logic. Parse the target compensation control logic into a set of equipment control instructions containing spindle speed adaptive load reduction parameters and high-speed bridge breaking parameters of the flying knife. Encapsulate the set of equipment control instructions into an industrial fieldbus protocol message and send it to the programmable logic controller to perform dynamic intervention.

[0014] A smart diagnostic system for abnormal veterinary drug powder production, used to execute the aforementioned smart diagnostic method for abnormal veterinary drug powder production, includes the following modules: a multi-source data fusion module, used to synchronously acquire the three-phase transient electrical parameter sequence of the powder mixer spindle, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port; perform Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence; and align the torque current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence according to the timestamp sequence to construct a mixing condition fusion matrix; a flow impedance reconstruction module, used to extract the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the mixing condition fusion matrix, and perform time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; and extract the trough distortion nodes of the acoustic impedance ratio sequence within a time sliding window. A powder flow impedance evolution sequence is constructed; an abnormal deviation diagnosis module is used to call a preset benchmark batch flow impedance sequence, and use a dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence, and calculate the time warping offset; when the cumulative mutation rate of the time warping offset exceeds a preset threshold range, the acoustic and electrical distortion feature vector corresponding to the deviation start time anchor point is extracted to construct a deterioration diagnosis label; a source tracing compensation control module is used to extract the temperature and humidity sequence within the corresponding source tracing window in the mixing condition fusion matrix according to the deviation start time anchor point in the deterioration diagnosis label, and calculate the microenvironment humid heat condensation gradient; the microenvironment humid heat condensation gradient and the acoustic and electrical distortion feature vector are input into a preset flow phase change compensation strategy library for matching, generate equipment control instruction set and send it to the programmable logic controller.

[0015] The present invention has the following beneficial effects: (1) An intelligent diagnostic method for abnormalities in veterinary drug powder production overcomes the shortcomings of traditional single-parameter monitoring, which suffers from physical information aliasing. By synchronously fusing the main shaft electrical parameters, high-frequency acoustic signals from the cylinder wall, and temperature and humidity at the feeding port, and using Park transform to extract the pure torque current component after eliminating reactive interference, high signal-to-noise ratio underlying data is obtained. Furthermore, this method creatively performs cross-mapping calculations of the acoustic envelope features characterizing micro-particle collision friction and the electrical features characterizing the macro-load of the equipment, constructing an acoustic impedance ratio sequence. This processing successfully removes the interference of pure mechanical transmission resistance fluctuations in the equipment, accurately amplifies the early physical distortion features of the powder when it transitions from a free-flowing state to a bonded state, and achieves highly sensitive quantitative characterization of the micro-deterioration process of the powder.

[0016] (2) An intelligent diagnostic method for abnormalities in veterinary drug powder production has been developed, overcoming the technical bottlenecks of traditional static threshold alarms, which are delayed and lack traceability. This method introduces a dynamic time warping algorithm to perform nonlinear time-series matching on the evolution trajectory of powder flow patterns. This eliminates reasonable fluctuations in feeding speed and process cycle between different batches, and accurately pinpoints the starting time anchor point of minute deviations, achieving early warning before large-scale powder agglomeration. More importantly, this method uses the extracted deviation anchor point to trace back historical temperature and humidity characteristics, calculates the microenvironmental humid heat condensation gradient, accurately identifies the physical root cause of powder phase change, and autonomously issues control commands in conjunction with the compensation strategy library. This establishes a complete control loop from abnormal state perception and root cause tracing to equipment adaptive repair, significantly reducing the scrap rate of defective products.

[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] Figure 1 This is a flowchart of an intelligent diagnostic method for abnormalities in the production of veterinary drug powder according to the present invention.

[0019] Figure 2 Flowchart for constructing the torque-current decoupling and mixing condition fusion matrix.

[0020] Figure 3 A flowchart for generating acoustic impedance ratio reconstruction and powder flow evolution trajectory.

[0021] Figure 4 A flowchart for dynamic intervention in microenvironment tracing and fluid state compensation.

[0022] Figure 5 This is a flowchart of an intelligent diagnostic system for abnormalities in the production of veterinary drug powder according to the present invention. Detailed Implementation

[0023] This application provides an intelligent diagnostic method and system for veterinary drug powder production anomalies, which solves the problem that existing technologies rely solely on static threshold alarms based on macroscopic electrical parameters, resulting in the inability to detect early microscopic deterioration of powder in a timely manner and making it difficult to trace the root cause for intervention.

[0024] The overall approach of the scheme in this application embodiment is as follows: First, multi-phase electrical parameters of the mixer spindle, high-frequency acoustic waveforms outside the cylinder, and ambient temperature and humidity in the material inlet area are collected synchronously. After coordinate transformation, torque current, which accurately represents mechanical work, is extracted, and various heterogeneous signals are aligned and fused along the time axis to form a basic operating condition matrix. Second, acoustic frequency band signals reflecting particle friction characteristics are extracted from the fused matrix and combined with torque current feature mapping to eliminate mechanical noise interference and reconstruct an evolution sequence that can truly reflect the changes in powder flow resistance. Next, historical golden batch trajectories are used as standard references, and a nonlinear time warp algorithm is used to elastically match the current resistance evolution sequence to identify subtle trajectory deviation time points and assemble diagnostic identification features. Finally, based on the locked deviation time points, the corresponding historical ambient temperature and humidity gradient changes are traced back to find the physical condensation root cause that leads to material abnormalities. Then, the corresponding compensatory control parameters are matched in conjunction with the equipment strategy library, and the automatically generated control commands are sent to the underlying controller for real-time correction.

[0025] Example 1; please refer to Figure 1 This invention provides a technical solution: an intelligent diagnostic method for abnormalities in veterinary drug powder production, comprising the following steps: S1, synchronously acquiring the three-phase transient electrical parameter sequence of the powder mixer spindle, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port; performing Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence; and aligning the torque current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence according to the timestamp sequence to construct a mixing condition fusion matrix; S2, extracting the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the mixing condition fusion matrix, and performing time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; extracting the trough distortion nodes of the acoustic impedance ratio sequence within a time sliding window to construct the powder... S3. Call the preset benchmark batch flow impedance sequence, and use the dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence, and calculate the time warping offset; when the cumulative mutation rate of the time warping offset exceeds the preset threshold range, extract the acoustic and electrical distortion feature vector corresponding to the deviation from the starting time anchor point, and construct the deterioration diagnosis label; S4. According to the deviation from the starting time anchor point in the deterioration diagnosis label, extract the temperature and humidity sequence in the corresponding traceability window in the mixing condition fusion matrix, and calculate the microenvironment humid heat condensation gradient; input the microenvironment humid heat condensation gradient and the acoustic and electrical distortion feature vector into the preset flow phase change compensation strategy library for matching, generate the equipment control instruction set and send it to the programmable logic controller.

[0026] In this implementation plan, step S1 is mainly used to acquire multi-dimensional raw operating condition data of the mixing equipment and achieve standardized alignment of multi-source data in space and time. In this step, the Parker transformation is a professional decoupling calculation method that transforms the stator three-phase AC static coordinate system to a coordinate system that rotates synchronously with the rotor. The torque current component extracted using this transformation represents the current portion of the main shaft motor used to overcome the internal material stirring resistance, removing reactive power and spurious signals from grid fluctuations. The mixing condition fusion matrix is ​​a high-dimensional data set formed by uniformly binding electrical work data, cylinder wall acoustic characteristic data, and environmental temperature and humidity data from different physical dimensions onto a completely consistent time reference axis. Through this step, the system not only eliminates monitoring errors caused by unstable power supply in the workshop and extracts accurate load data that truly reflects the macroscopic mechanical load of the stirring shaft, but also eliminates the time delay difference between heterogeneous sensors, providing a precise underlying data source with accurate time nodes for subsequent in-depth cross-analysis of multi-dimensional materialized signals.

[0027] Step S2 is mainly used to decouple and extract characteristic evolution laws that can truly reflect the microscopic flow and mixing state of veterinary drug powder from heterogeneous fused data. In this step, the characteristic frequency band envelope sequence is the energy intensity distribution shape extracted after filtering the captured high-frequency acoustic signal, which is used to accurately characterize the microscopic intensity of collision and friction between powder particles and between particles and the cylinder wall; the acoustic impedance ratio sequence is a set of dynamic ratios obtained by cross-comparing and mapping the acoustic parameters representing microscopic frictional energy with the torque and current parameters representing the overall macroscopic stirring resistance; the trough distortion node refers to the sudden abnormal minimum value point of the impedance ratio sequence on the time axis, representing the sudden local stagnation of the originally continuous and uniform powder flow. By establishing a mapping relationship between the acoustic response of particle microscopic collision and friction and the macroscopic driving electrical load of the equipment, this step successfully eliminates the load fluctuation interference caused by the wear of the mixer's own bearings and purely mechanical factors, and can extremely sensitively capture the unique flow resistance change generated when veterinary drug powder changes from a loose and free-flowing state to a hygroscopic and sticky state, and restore the evolution process of material mixing characteristics with high fidelity.

[0028] Step S3 is primarily used to accurately identify gradual deterioration and early-stage, concealed production anomalies in the current batch. In this step, the baseline batch flow impedance sequence refers to the standard trajectory data retained from the best historical batches that have undergone rigorous laboratory testing and achieved perfect mixing uniformity in previous production. The dynamic time warping algorithm is a nonlinear elastic matching algorithm that measures the similarity between two time series with different time lengths or inconsistent action rhythms. The time warping offset refers to the cumulative spatial error value calculated after the current batch production data and historical baseline data are elastically stretched or compressed to align with the time axis, deviating from the normal standard trajectory. This step and algorithm completely overcome the false alarm problem caused by differences in feeding speed among different operators or minor fluctuations in the process when comparing fixed time axes. It can capture minute trajectory deformations at the initial stage when the current material mixing state just begins to deviate from the golden production standard, accurately locating the exact time point of anomaly occurrence, greatly improving the sensitivity and early warning capability for identifying early signs of powder agglomeration and clumping.

[0029] Step S4 is mainly used to trace the physical root causes of powder deterioration and to implement active adaptive correction and repair of the equipment at the control level. In this step, the traceability window refers to a specific historical review time range intercepted before the time of the anomaly occurrence locked in step S3; the microenvironment humid heat condensation gradient characterizes the physical reaction rate at which water vapor near the feeding port condenses on the surface of material particles in a short time and causes changes in the surface tension of the powder, resulting in capillary bridging phase change; the flow state phase change compensation strategy library is a pre-set set of optimal equipment intervention actions corresponding to various abnormal powder adhesion conditions. This step breaks through the limitations of traditional diagnostics that only issue alarms but do not solve problems. It can not only clearly identify that the current resistance abnormality is caused by the physical condensation, stickiness, and agglomeration of powder due to the rapid changes in workshop temperature and humidity, but also link this root cause analysis results with the strategy library to automatically generate control instructions that match the current degree of deterioration. For example, temporarily reduce the speed of the main agitator to slow down the powder compression, and at the same time start the high-speed flying knife device at full speed to cut up the early agglomerates. These instructions are directly sent to the field logic controller for intervention, realizing a fully intelligent closed loop from abnormality detection and physical tracing to equipment autonomous repair, fundamentally saving high-value veterinary drug products that are about to be scrapped.

[0030] Please see Figure 2Specifically, the process of synchronously acquiring the three-phase transient electrical parameter sequence of the powder mixer main shaft, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port, and performing Parker transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence is as follows: The three-phase transient electrical parameter sequence is acquired through a Hall sensor array; the high-frequency acoustic emission waveform sequence is acquired through a piezo-acoustic sensor; and the temperature and humidity sequence of the feeding port is acquired through a high-frequency response temperature and humidity sensor. Phase-locked loop frequency tracking is performed on the three-phase transient electrical parameter sequence to extract the fundamental phase angle of the stator voltage. Based on the fundamental phase angle of the stator voltage, a Clarke transform is performed on the three-phase transient electrical parameter sequence to generate a two-phase stationary orthogonal stator current sequence. A rotating coordinate system transformation is performed on the two-phase stationary orthogonal stator current sequence to separate the excitation current component sequence and the torque current component sequence.

[0031] In this implementation scheme, a three-dimensional physical signal acquisition hardware architecture covering the motor drive end, equipment mechanical end, and material environment end is constructed through the collaborative operation of a Hall sensor array, piezo-acoustic sensor, and high-frequency response temperature and humidity probe. This ensures that the acquired data can accurately reflect the transient physical field changes during the mixing of veterinary drug powder. Because the workshop power supply network frequently experiences harmonic interference and voltage fluctuations, directly using the acquired three-phase transient electrical parameter sequence is insufficient to accurately extract the effective work component used to overcome material stirring resistance. Therefore, coordinate system decoupling calculation is necessary. First, the periodic fluctuations of the power grid frequency are locked using phase-locked loop frequency tracking technology to extract the precise stator voltage fundamental phase angle, providing an absolute phase reference for subsequent coordinate mapping. Then, the three-phase transient electrical parameter sequence is projected onto a two-phase stationary orthogonal coordinate system using the Clarke transform to generate a two-phase stationary orthogonal stator current sequence. Finally, a rotating coordinate system transformation calculation, i.e., a Parker transform, is performed based on the stator voltage fundamental phase angle. In the Park transform solution, to eliminate the calculation distortion caused by the local voltage drop in the power grid at the moment of veterinary drug feeding, a power grid fluctuation compensation coefficient is introduced for adaptive correction. This completely separates the torque current component sequence that purely characterizes the mixing equipment overcoming the internal frictional and viscous resistance of the powder. The specific calculation expression is as follows: ,in, : Indicates a time node The torque current component below; : Indicates a time node The power grid fluctuation compensation coefficient under the following conditions; : Represents the alpha axis current component in a two-phase stationary orthogonal stator current sequence; : Indicates a time node The fundamental phase angle of the stator voltage extracted below; : Represents the beta-axis current component in a two-phase static orthogonal stator current sequence; for power grid fluctuation compensation coefficient The method for determining this is to obtain the current time node. The transient voltage drop deviation of the equipment bus is calculated, and the product of the voltage drop deviation and the preset penalty factor is calculated. The negative exponent of the natural constant is then added to the constant to obtain the result. This dynamically corrects the torque distortion caused by external power grid fluctuations, so that the separated torque current component sequence can fully map the actual adhesion force change of the veterinary drug powder in the mixing drum.

[0032] Specifically, the process of aligning the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence according to their timestamps to construct the mixing condition fusion matrix is ​​as follows: Extract the hardware system timestamps of the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence; set the sampling frequency of the torque current component sequence as the reference resampling frequency; perform downsampling anti-aliasing filtering on the high-frequency acoustic emission waveform sequence based on the reference resampling frequency; perform spline interpolation fitting on the temperature and humidity sequence; output the synchronous time sequence with the same frequency; perform tensor splicing operation on the synchronous time sequence with the same frequency according to the signal dimension features; and output the mixing condition fusion matrix.

[0033] In this implementation scheme, since the original hardware sampling frequencies of electrical signals, acoustic signals, and ambient temperature and humidity signals span a huge range from Hertz to Megahertz, direct combination would lead to severe phase misalignment and feature collapse. Therefore, it is necessary to extract the hardware system timestamps of each type of signal and perform cross-domain reconstruction using the time axis as the absolute reference. Considering that the sampling cycle of the torque current component sequence best matches the macroscopic mechanical rotation dynamic characteristics of the powder mixer spindle, the system sets it as the reference resampling frequency. For high-frequency acoustic emission waveform sequences with extremely high sampling rates, the system performs downsampling anti-aliasing filtering, using a finite-length unit impulse response filter to compress redundant data while fully preserving the high-frequency energy envelope characteristics of powder particles impacting the metal cylinder wall. For temperature and humidity sequences with extremely low sampling rates and slow changes, high-order spline interpolation fitting is performed to smoothly transition and fill in the low-frequency discrete points to the reference resampling frequency. After acquiring the synchronous time sequence at the same frequency, the system performs tensor splicing operations according to the signal dimensional characteristics. To prevent feature annihilation due to different physical dimensions in tensor space, cross-modal alignment confidence weights are incorporated into the tensor concatenation operation for adaptive balancing of the feature space, outputting a structured fusion matrix for the mixing process. The specific tensor concatenation calculation expression is as follows: ,in, : Indicates the alignment index The eigenvectors of the fusion matrix under the mixing condition; : Indicates the alignment index Cross-modal alignment confidence weights; : Represents the torque-current frequency sequence components after extreme value normalization. : Represents the tensor dimension orthogonal splicing operator; : Represents the acoustic emission frequency sequence components after downsampling and anti-aliasing filtering; : Represents the same-frequency sequence components of temperature and humidity after spline interpolation fitting; for cross-modal alignment confidence weights The method for determining this is to extract the alignment index from various original sensors. The jitter variance of the clock signal within the corresponding physical time capture window is calculated by summing the reciprocals of the jitter variances of each clock signal, and then normalizing the ratio of the reciprocal of the local variance to the sum of these reciprocals. The smaller the hardware clock jitter, the higher the timing alignment accuracy at that moment, and the greater the feature weight assigned to it. This ensures the dual rigor of the output mixing condition fusion matrix in terms of time alignment and physical representation, laying a solid foundation for the underlying data of subsequent cross-analysis of powder flow acoustic impedance.

[0034] Please see Figure 3 Specifically, the process of extracting the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence from the fusion matrix of the mixing condition and constructing the acoustic impedance ratio sequence by performing time-frequency domain cross-mapping with the torque current component sequence is as follows: Wavelet packet decomposition is performed on the high-frequency acoustic emission waveform sequence to extract high-frequency detail coefficients characterizing the collision and friction response frequency band of powder particles. Hilbert transform demodulation is performed on the high-frequency detail coefficients to extract the energy envelope, generating the characteristic frequency band envelope sequence. Max-min normalized scaling is performed on the characteristic frequency band envelope sequence and the torque current component sequence. The scale-scaled torque current component sequence is used as the load response variable, and the characteristic frequency band envelope sequence is used as the acoustic dissipation variable. Dissipative load ratio feature mapping is performed to extract the acoustic impedance ratio sequence.

[0035] In this implementation scheme, since the original high-frequency acoustic emission waveform sequence contains not only the effective sound of powder particle impact and friction, but also wind noise from the agitator blades cutting the air and background broadband noise from bearing operation, the system employs wavelet packet decomposition technology to perform multi-scale frequency domain refinement analysis on the original waveform in order to accurately extract the pure powder friction acoustic characteristics. Wavelet packet decomposition can provide more refined frequency band division than traditional wavelet transform, especially with higher resolution in the high-frequency part. The system uses this to lock onto specific high-frequency bands characterizing the collision and friction response of powder particles and extract the high-frequency detail coefficients in this band, which are then demodulated using Hilbert transform. Hilbert transform can perfectly extract the low-frequency energy envelope of the high-frequency oscillation waveform by constructing an analytical signal, thereby generating a characteristic frequency band envelope sequence. At this point, the characteristic frequency band envelope sequence represents the intensity of the microscopic motion inside the powder, while the torque current component sequence extracted in the previous step represents the mechanical work done by the main shaft driving the macroscopic tumbling of the powder. To eliminate the significant difference in absolute magnitude between acoustic and current amplitudes, the system employs a max-min normalization algorithm to compress them into a unified dimensionless calculation interval. Finally, the system treats the scale-unified torque-current component sequence as the load response variable at the equipment end and the characteristic frequency band envelope sequence as the acoustic dissipation variable generated at the material end. It then performs dissipative load ratio feature mapping extraction to construct an acoustic impedance ratio sequence that sensitively reflects the precursors of material adhesion and stickiness. The specific calculation expression is as follows: ,in, : Represents the acoustic impedance ratio sequence component at sampling node k; : Represents the frequency band response correction coefficient at sampling node k; : Represents the characteristic frequency band envelope component at sampling node k; : Represents the load response variable component at sampling node k; : Represents the minimum bias constant used to prevent the denominator from being zero; for frequency band response correction coefficients The method for determining the acoustic impedance ratio is to extract the characteristic energy value of the wavelet packet decomposition node corresponding to the current sampling node and calculate the ratio of this characteristic energy value to the original full-band acoustic emission total energy value. Using the acoustic impedance ratio model constructed through this formula, when the powder exhibits localized moisture absorption, stickiness, and agglomeration due to increased microenvironment humidity, the friction and collision between particles become dull, leading to a decrease in the characteristic frequency band envelope amplitude. Simultaneously, the increased bonding resistance causes an increase in torque current. The numerator decreases while the denominator increases, resulting in a drastic drop in the calculated acoustic impedance ratio, thus amplifying the phase transition of microscopic materials.

[0036] Specifically, the process of extracting the trough distortion nodes of the acoustic impedance ratio sequence within a time sliding window and constructing the powder flow impedance evolution sequence is as follows: a moving data sliding window is constructed along the time axis, and morphological cap transformation is performed on the acoustic impedance ratio sequence within the moving data sliding window to extract the abnormal concave trough nodes of the morphological cap transformation sequence; the waveform deformation curvature feature vectors before and after the abnormal concave trough nodes are extracted, and free fluctuation nodes are filtered according to the preset curvature gradient reference interval to extract the trough distortion nodes; the trough distortion nodes are arranged in order along the time axis, and dynamic impedance trend envelope reconstruction is performed to output the powder flow impedance evolution sequence.

[0037] In this implementation scheme, since the movement of veterinary drug powder within the mixer is a multidimensional chaotic nonlinear dynamic process, the generated acoustic impedance ratio sequence is often accompanied by a large number of normal fluctuation spikes. Directly setting a fixed drop threshold for diagnosis can easily lead to malfunctions in the control system. Therefore, the system constructs a moving data sliding window along the time axis, dividing continuous time-series data into multiple overlapping local dynamic segments, and introducing morphological cap transformation processing within each segment. Morphological cap transformation is a nonlinear filtering tool based on mathematical morphology. It can effectively remove slowly changing background signals and specifically highlight abnormally low-amplitude trough nodes that are darker than the surrounding background, i.e., nodes with sudden amplitude drops. These nodes physically correspond to the moment when the powder flow rate in a certain array area inside the mixer suddenly slows down. To further eliminate pseudo-distortion nodes caused by transient inrush currents in the power grid or feeding drop differences, the system extracts the waveform deformation curvature feature vectors before and after the abnormally low-amplitude trough nodes, and calculates a numerical representation of the steepness of the drop based on the data gradient of the surrounding neighborhood. The specific calculation expression is as follows: ,in, : Represents the waveform deformation curvature characteristic at distortion evaluation node p; : Represents the local smoothing penalty coefficient at distortion evaluation node p; : Represents the amplitude of the abnormal concave valley extracted at the distortion assessment node p; : Represents the trough amplitude corresponding to the preceding node of the distortion evaluation node p; : Represents the trough amplitude corresponding to the subsequent node of the distortion evaluation node p; : Indicates a scalar value for attenuation adjustment; : Represents the time dimension span between distortion evaluation node p and its predecessor node; for the local smoothing penalty coefficient The determination method involves calculating the local fluctuation variance of the acoustic impedance ratio data within a preset time window before and after the distortion assessment node p, and outputting the reciprocal of this local fluctuation variance as a penalty coefficient. This achieves adaptive weight reduction and suppression of pseudo-valleys caused by random, severe high-frequency noise. After calculation, the system compares the obtained curvature feature with the preset curvature gradient reference interval to filter out free fluctuation nodes, retaining only the valley distortion nodes truly caused by the deterioration of the material flow regime. Finally, the system rearranges and connects the verified qualified valley distortion nodes in time-axis order, performs dynamic impedance trend envelope reconstruction, and uses a smooth curve to reconstruct an overall deterioration trajectory through these key nodes, outputting a sequence of powder flow regime impedance evolution. This plays a key technical role in transforming discrete microscopic distortion points into characteristic lines that can continuously reflect the deterioration trend of the entire batch of veterinary drug mixture flow regime, providing high-purity diagnostic evidence for subsequent traceability.

[0038] Specifically, the process involves calling a preset benchmark batch flow impedance sequence and using a dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence. The specific steps for calculating the time warping offset are as follows: Constructing a state evolution difference mapping matrix between the powder flow impedance evolution sequence and the benchmark batch flow impedance sequence; applying a dynamic programming algorithm to the state evolution difference mapping matrix to search for the optimal state matching path that satisfies endpoint binding and continuity constraints; extracting the multidimensional state space mapping feature vectors of each aligned node pair on the optimal state matching path; and performing cumulative feature extraction of the evolution trajectory deviation based on the multidimensional state space mapping feature vectors to output the time warping offset.

[0039] In this implementation scheme, due to objective and reasonable differences in feeding speed, manual operation rhythm, and initial material viscosity among different batches of veterinary drug powder, even during normal mixing processes, the flow regime of the powder will undergo stretching or compressing deformation over time. If a direct point-to-point hard time comparison is performed between the current batch's powder flow regime impedance evolution sequence and the baseline batch's flow regime impedance sequence, severe misalignment and false alarms are highly likely. Therefore, the system constructs a state evolution difference mapping matrix covering the state nodes of the current production batch and historical golden batches, and applies a dynamic programming algorithm to this matrix. The dynamic programming algorithm searches within the matrix for an optimal state matching path that best matches the overall shape of the two sequences, based on the endpoint binding condition that must start from the process start point and end point, and the continuity constraint that time can only advance and not regress. After locking the optimal state matching path, the system extracts the multidimensional state space mapping feature vector of each aligned node pair on the optimal state matching path to characterize the actual material impedance deviation after eliminating the process time difference. Based on the multidimensional state space mapping feature vector, the system performs cumulative feature extraction of evolution trajectory deviation and outputs the time-ordered offset. The specific calculation expression is as follows: ,in, : Represents the timing warp offset component under the alignment path node c; : Represents the material phase change sensitivity weighting coefficient under the alignment path node c; : Indicates the index of the powder flow impedance evolution sequence at the current time. Eigenvalues ​​at; : Indicates the index of the reference batch flow impedance sequence at the matching time. Eigenvalues ​​at; : Represents the cardinality of the time warp penalty; : Represents the time span of nonlinear stretching or compression along the time axis occurring at node c in the alignment path; weighting coefficient for material phase change sensitivity. The determination method is to extract the flow impedance sequence of the reference batch at the matching time index. The local data variance in the preceding and following neighborhood intervals is calculated, and the ratio of the local data variance to the global maximum data variance is obtained by subtracting the calculated ratio from a constant. Through this calculation model, the system effectively absorbs the time redundancy and cycle fluctuations brought about by normal production operations, and applies a logarithmic decay penalty to the matching situation of excessively distorted time axis. It only accumulates and amplifies the resistance deviations caused by physical deterioration such as powder stickiness and micro-agglomeration, which plays a technical role in accurately filtering time axis pseudo-scatter distortion and extracting pure material phase transformation anomalies.

[0040] Specifically, when the cumulative mutation rate of the time-order regularization offset exceeds the preset threshold range, the specific process of extracting the acoustic-electric distortion feature vector corresponding to the deviation from the starting time anchor point and constructing the degradation diagnosis label is as follows: Extract the local variation rate feature sequence of the time-order regularization offset along the time axis, construct the mutation rate trend curve, perform boundary cross-judgment on the mutation rate trend curve, and mark the first jump node that breaks through the upper limit of the preset threshold range as the deviation from the starting time anchor point; extract the neighborhood time truncation window corresponding to the deviation from the starting time anchor point, extract the working condition evolution time slice in the powder flow impedance evolution sequence, perform flow feature decoupling and separation on the working condition evolution time slice to extract the core deformation component, construct the acoustic-electric distortion feature vector, and package and reassemble the acoustic-electric distortion feature vector with the deviation from the starting time anchor point to construct the degradation diagnosis label.

[0041] In this implementation scheme, after extracting the time-ordered offset after removing time errors, it is necessary to further determine whether the small offset has evolved into a substantial mixed anomaly. The system extracts the local variation rate feature sequence of the time-ordered offset along the time axis, and then constructs a mutation rate trend curve reflecting the accelerated deterioration. A boundary cross-judgment is performed on the mutation rate trend curve. When the curve crosses the upper limit of the preset safety threshold range from bottom to top, it indicates that the agglomeration and clumping inside the powder has broken the critical equilibrium of the equipment's self-healing capability. The system immediately marks the first jump node that breaks the upper limit of the preset threshold range as the deviation from the initial time anchor point. The deviation from the initial time anchor point accurately records the initial onset time of the powder anomaly. To diagnose the specific physical properties of the anomaly, the system extracts the neighborhood time cutoff window corresponding to the deviation from the initial time anchor point and extends it forward and backward to obtain a time slice of the operating condition evolution that includes the anomaly's incubation period and outbreak period. Since the operating condition evolution time slice still contains normal macroscopic stirring resistance fluctuations, the system performs flow characteristic decoupling and separation on the operating condition evolution time slice to extract the core deformation component. The specific calculation expression is as follows: ,in, : Indicates the core deformation component under the truncated window index w; : Indicates the amplified weight of abnormal focusing under the truncated window index w; : Represents the actual impedance amplitude of the time slice under the truncated window index w; : Represents the baseline amplitude of the macro trend extracted after applying low-pass smoothing filtering to the time slice of the operating condition evolution; : Represents the overall discrete variance of impedance data within the entire operating condition evolution time slice; : Represents a minimal constant to prevent denominator overflow; amplifies weights for anomaly focusing. The method for determining the impedance amplitude is to calculate the absolute difference between the actual impedance amplitude of the time slice of the operating condition evolution and the macroscopic trend benchmark amplitude, and then divide the absolute difference by the arithmetic square root of the overall discrete variance. By extracting the core deformation components, the system eliminates the influence of background baseline drift, constructs a pure acoustic-electric distortion feature vector representing the contour of the abnormal features, and packages and recombines the acoustic-electric distortion feature vector with the deviation from the starting time anchor point to form a degradation diagnostic label. This series of core operations enables the system not only to grasp the exact time point of the abnormality, but also to obtain the specific impedance distortion morphology when powder agglomerates. This provides a high-dimensional and high-fidelity diagnostic record for subsequent matching of specific physical condensation compensation strategies, playing a decisive technical role in transforming abstract fluctuation data into concrete diagnostic labels.

[0042] Please see Figure 4Specifically, based on the deviation from the starting time anchor point in the deterioration diagnosis identifier, the temperature and humidity sequence within the corresponding traceability window is extracted from the fusion matrix of the mixing condition, and the specific process for calculating the microenvironmental humid heat condensation gradient is as follows: The deterioration diagnosis identifier is parsed to obtain the deviation from the starting time anchor point; the temperature and humidity sequence with a preset span preceding the deviation from the starting time anchor point is extracted in reverse along the time axis to form a traceability data slice; the temperature waveform and relative humidity waveform in the traceability data slice are separated, and dew point temperature nonlinear fitting is performed in combination with the saturated water vapor pressure parameter to generate a near-wall condensation critical characteristic curve; the condensation phase change approximation rate characteristics of the near-wall condensation critical characteristic curve and the actual temperature waveform are extracted, and the microenvironmental humid heat condensation gradient is output.

[0043] In this implementation scheme, since the stickiness and agglomeration of powder inside the mixer is often a delayed physical reaction caused by changes in the temperature and humidity of the workshop environment, direct environmental data analysis at the moment of the anomaly cannot find the true cause. Therefore, the system must analyze the degradation diagnostic markers to obtain the anchor point of deviation from the start time, and backtrack along the time axis to extract a source data slice containing the anomaly latency period. In physical space, when the actual temperature of the mixing drum wall or the powder surface is lower than the dew point temperature of the air, water vapor in the air will precipitate and condense. After the moisture is absorbed by the powder, it will cause the surface tension of the particles to increase, thus forming capillary bridging and micro-agglomeration. Based on this physical mechanism, the system separates the temperature waveform and the relative humidity waveform, and performs nonlinear fitting with the saturated vapor pressure parameter to deduce the near-wall condensation critical characteristic curve, i.e., the dynamic baseline of the dew point temperature evolution over time. In order to quantify the severity of the current environment causing the powder to undergo a condensation phase change, the system extracts the approximation correlation between the near-wall condensation critical characteristic curve and the actual temperature waveform, and outputs the microenvironmental humid heat condensation gradient. The specific calculation expression is as follows: ,in, : Represents the microenvironmental humid-thermal condensation gradient at time node n in the source tracing process; : Represents the phase transition sensitivity penalty coefficient at the tracing time node n; : Represents the near-wall condensation critical characteristic value generated by nonlinear fitting at the source tracing time node n; : Represents the actual temperature waveform amplitude at time node n in the source tracing process; : Represents the absolute temperature constant offset to prevent the denominator from being zero; : Represents the actual relative humidity waveform amplitude at time node n in the source tracing process; penalty coefficient for phase change sensitivity. The method for determining the slope is to extract the actual temperature decrease slope within the preceding local time window of the traceability time node n, and calculate the ratio of the absolute value of this actual temperature decrease slope to the preset environmental buffer heat capacity factor. Through the microenvironmental humid heat condensation gradient constructed in this step, the system not only grasps the absolute water molecule content in the air, but also accurately captures the rate at which the actual working conditions are approaching the physical critical point that triggers large-area powder agglomeration. This plays a decisive technical role in understanding the causes of powder state deterioration from the root of the environment.

[0044] Specifically, the process of matching the microenvironmental humid heat condensation gradient with the acoustic and electrical distortion feature vector into a preset flow phase change compensation strategy library to generate a set of equipment control instructions and send them to the programmable logic controller is as follows: Multi-dimensional operating condition feature fusion and reconstruction is performed on the microenvironmental humid heat condensation gradient and the acoustic and electrical distortion feature vector to construct a multi-dimensional state addressing code. Operating condition similarity feature optimization matching of the multi-dimensional state addressing code is performed in the flow phase change compensation strategy library to match the target compensation control logic. The target compensation control logic is parsed into a set of equipment control instructions containing spindle speed adaptive load reduction parameters and high-speed bridge breaking parameters of the flying knife. The set of equipment control instructions is encapsulated into an industrial fieldbus protocol message and sent to the programmable logic controller for dynamic intervention.

[0045] In this implementation plan, to achieve non-stop self-repair of veterinary drug powder that is about to clump and become unusable, the system performs multi-dimensional operating condition feature fusion and reconstruction by fusing the microenvironmental humid heat condensation gradient (representing the cause of abnormal outbreaks) and the acoustic and electrical distortion feature vector (representing the symptoms of material mixing deterioration). This is then spliced ​​to generate a multi-dimensional state-addressing code that comprehensively includes causal relationships. The flow phase change compensation strategy library pre-stores a large number of expert correction plans to cope with various extreme humid heat conditions and different degrees of adhesion. The system performs operating condition similarity feature optimization matching in this strategy library to find the target compensation control logic that best matches the current production deterioration state. The specific matching calculation expression is as follows: ,in, : Represents the condition similarity evaluation value between the current encoding and the x-th strategy template in the flow state phase change compensation strategy library; : Represents the historical application reliability weight coefficient for the x-th strategy template; : Represents the addressing preference coefficient of the dimension of the acoustic-electric distortion feature vector; : Represents the core scalar aggregate value of the currently constructed acoustic-electric distortion feature vector; : Represents the standard acoustic-electric distortion scalar boundary value pre-stored in the x-th strategy template; : Represents the addressing preference coefficient of the humid heat condensation gradient; : Represents the current calculated output value of the microenvironment humid heat condensation gradient polymerization value; : Represents the standard condensation gradient scalar boundary value pre-stored in the x-th strategy template; reliability weighting coefficient for historical applications. The determination method involves extracting the number of successful blocking events of the x-th strategy template in the fluid phase change compensation strategy library from historical abnormal intervention records, and calculating the ratio of this successful blocking event to the total number of trigger events. After the system locks the target compensation control logic with the highest operating condition similarity evaluation value, it parses it into specific mechanical intervention parameters, namely, reducing the rotation speed of the mixer spindle to slow down the extrusion and heating of the agglomerated powder, and simultaneously starting the high-frequency full-speed sidewall flying knife device to forcibly break up the forming micro-bridging agglomerates. Finally, these equipment control instruction sets are encapsulated into industrial fieldbus protocol messages and sent to the programmable logic controller to drive the field equipment to complete millisecond-level dynamic intervention actions. This complete closed loop completely changes the backward situation of traditional systems that can only passively alarm and shut down, giving the production line the ability to autonomously correct and compensate for temperature and humidity disturbances and material phase changes, and significantly improving the finished product qualification rate of continuous production of veterinary drug powder.

[0046] Example 2; please refer to Figure 5 A smart diagnostic system for abnormal veterinary drug powder production, used to execute the smart diagnostic method for abnormal veterinary drug powder production described in the embodiments, includes the following modules: a multi-source data fusion module, used to synchronously collect the three-phase transient electrical parameter sequence of the powder mixer spindle, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port; perform Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence; and align the torque current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence according to the timestamp sequence to construct a mixing condition fusion matrix; a flow impedance reconstruction module, used to extract the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the mixing condition fusion matrix, and perform time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; and extract the trough distortion of the acoustic impedance ratio sequence within a time sliding window. The system comprises several modules: a node for constructing a powder flow impedance evolution sequence; an anomaly deviation diagnosis module for calling a preset benchmark batch flow impedance sequence and using a dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the benchmark batch flow impedance sequence, calculating the time warping offset; when the cumulative mutation rate of the time warping offset exceeds a preset threshold range, extracting the acoustic and electrical distortion feature vector corresponding to the deviation start time anchor point to construct a deterioration diagnosis identifier; and a source-tracing compensation control module for extracting the temperature and humidity sequence within the corresponding source-tracing window from the mixing condition fusion matrix based on the deviation start time anchor point in the deterioration diagnosis identifier, calculating the microenvironmental humid heat condensation gradient; and inputting the microenvironmental humid heat condensation gradient and the acoustic and electrical distortion feature vector into a preset flow phase change compensation strategy library for matching, generating an equipment control instruction set and sending it to the programmable logic controller.

[0047] In this implementation scheme, the multi-source data fusion module serves as the underlying sensing and data preprocessing hub of the entire system. Its core function is to break down data silos between different physical quantity acquisition devices. This module is not only responsible for comprehensively acquiring the driving electrical, mechatronic, and material environment status of the mixing equipment, but more importantly, it eliminates reactive power interference caused by power grid fluctuations through Parker transformation and strictly locks these heterogeneous data with different sampling frequencies and physical dimensions onto a unified time reference. This ensures that the data relied upon for subsequent analysis has absolute spatiotemporal consistency, laying a solid foundation of high-quality and highly synchronous underlying data for accurately analyzing the powder mixing state. The flow impedance reconstruction module acts as the computational engine for decoupling deep process feature extraction and physical principles. The role of this module is to cross-integrate the microscopic acoustic characteristics of powder particle collisions with the macroscopic electrical characteristics of equipment mechanical work, thereby successfully separating pure mechanical transmission interference from complex production noise. By extracting waveform distortion nodes within a specific sliding window and reconstructing the evolution trajectory, this module transforms massive and chaotic raw underlying signals into characteristic curves that intuitively reflect the phase transition of powder from free flow to hygroscopic and sticky state, achieving visualization and continuous quantitative characterization of the physical deterioration state of veterinary drug powder mixtures. The abnormal deviation diagnosis module constitutes the core of the system's intelligent discrimination for early warning of production deterioration. This module aims to overcome time axis alignment errors caused by differences in feeding speed or worker operation rhythm in different batches of production. By introducing a dynamic time warping algorithm to elastically match and compare the current flow evolution trajectory with the historical golden batch trajectory, this module can extremely sensitively capture the minute resistance deviations generated by powder during the agglomeration bud stage. Its mechanism of accurately locking the onset time of abnormality and extracting corresponding acoustic and electrical distortion features enables the system to accurately diagnose and characterize the disease before irreversible large-area dead material agglomeration occurs. The traceability compensation control module is a closed-loop execution terminal that enables the production line to transition from passive monitoring and alarm to proactive adaptive repair. The key function of this module is to use the diagnosed anomaly initiation time point to retrospectively trace historical environmental data, accurately quantifying the microenvironmental physical condensation phase transition causes that lead to powder stickiness and agglomeration. Furthermore, the module combines environmental causes with material deterioration characterization, optimizing the dynamic correction scheme most suitable for the current degree of deterioration from an expert strategy library, and directly translating the scheme into low-level execution instructions for the programmable logic controller (PLC). This completely establishes a fully automated closed loop of state perception, physical root cause tracing, and proactive equipment intervention, effectively saving veterinary drug powder products from potential quality incidents.

[0048] In summary, this application has at least the following effects: A method and system for intelligent diagnosis of anomalies in veterinary drug powder production is proposed. This method integrates pure mechanical torque current extracted via Parker transform decoupling, high-frequency acoustic energy envelope from the cylinder wall, and temperature and humidity signals from the feeding port to construct an acoustic impedance ratio characteristic mapping sequence. This successfully eliminates pseudo-load fluctuation interference caused by the equipment's own mechanical transmission and amplifies the microscopic phase transition characteristics of early moisture absorption and stickiness in the powder with high sensitivity. Simultaneously, combined with a dynamic time warping algorithm, it effectively absorbs time cycle misalignment errors between normal process batches, enabling precise diagnosis before large-scale irreversible agglomeration of the powder occurs. By identifying the initial lesion anchor point of minute resistance deterioration, and then using this abnormal onset anchor point to reverse-engineer the microenvironment's humid and thermal condensation gradient, the root cause of powder agglomeration can be seen from a deep physical and thermodynamic perspective. In conjunction with the flow state phase change compensation strategy library, dynamic correction commands such as load reduction and bridge breaking are autonomously issued to the underlying logic controller. This completely opens up the control closed loop from hidden early warning, physical root cause tracing to equipment non-stop adaptive repair, significantly reducing the veterinary drug scrap rate caused by microenvironmental disturbances, and greatly improving the robustness and finished product qualification rate of the continuous intelligent powder production line.

[0049] 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 the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent diagnosis of abnormalities in veterinary drug powder production, characterized in that, Includes the following steps: S1. Synchronously collect the three-phase transient electrical parameter sequence of the main shaft of the powder mixer, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port. Perform Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence, and align the torque current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence according to the timestamp sequence to construct a mixing condition fusion matrix. S2. Extract the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the fusion matrix of the mixing condition, and perform time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; extract the trough distortion nodes of the acoustic impedance ratio sequence within the time sliding window to construct the powder flow state impedance evolution sequence. S3. Call the preset reference batch flow impedance sequence, and use the dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the reference batch flow impedance sequence, and calculate the time warping offset; when the cumulative mutation rate of the time warping offset exceeds the preset threshold range, extract the acoustic and electrical distortion feature vector corresponding to the deviation from the starting time anchor point, and construct the deterioration diagnosis label. S4. Based on the deviation from the starting time anchor point in the deterioration diagnosis identifier, extract the temperature and humidity sequence in the corresponding traceability window from the mixing condition fusion matrix, and calculate the microenvironmental humid heat condensation gradient; input the microenvironmental humid heat condensation gradient and the acoustic and electrical distortion feature vector into the preset flow state phase change compensation strategy library for matching, generate the equipment control instruction set and send it to the programmable logic controller.

2. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 1, characterized in that: The specific process of simultaneously acquiring the three-phase transient electrical parameter sequence of the powder mixer main shaft, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port, and performing Park transform on the three-phase transient electrical parameter sequence to extract the torque current component sequence is as follows: Three-phase transient electrical parameter sequences are acquired using a Hall sensor array, high-frequency acoustic emission waveform sequences are acquired using a piezo-acoustic sensor, and temperature and humidity sequences at the feeding port are acquired using a high-frequency response temperature and humidity sensor. A phase-locked loop frequency tracking is performed on the three-phase transient electrical parameter sequence to extract the fundamental phase angle of the stator voltage. Based on the fundamental phase angle of the stator voltage, a Clarke transform is performed on the three-phase transient electrical parameter sequence to generate a two-phase static orthogonal stator current sequence. A rotating coordinate system transformation is performed on the two-phase stationary orthogonal stator current sequence to separate the excitation current component sequence and the torque current component sequence.

3. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 2, characterized in that: The specific process of aligning the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence according to timestamps to construct the fusion matrix for the mixing condition is as follows: Extract the hardware system timestamps of the torque current component sequence, high-frequency acoustic emission waveform sequence, and temperature and humidity sequence, and set the sampling frequency of the torque current component sequence as the reference resampling frequency. Based on the reference resampling frequency, downsampling and anti-aliasing filtering are performed on the high-frequency acoustic emission waveform sequence, spline interpolation fitting is performed on the temperature and humidity sequence, and the synchronous time sequence with the same frequency is output. Tensor splicing operation is performed on the synchronous time sequence with the same frequency according to the signal dimension characteristics, and the fusion matrix of mixing conditions is output.

4. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 1, characterized in that: The specific process of extracting the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence from the fusion matrix of the mixing condition, and constructing the acoustic impedance ratio sequence by performing time-frequency domain cross-mapping with the torque current component sequence is as follows: Wavelet packet decomposition is performed on the high-frequency acoustic emission waveform sequence to extract high-frequency detail coefficients that characterize the frequency band of powder particle collision and friction response. Hilbert transform is then performed on the high-frequency detail coefficients to demodulate and extract the energy envelope, generating a characteristic frequency band envelope sequence. The maximum-minimum normalization scaling is applied to the characteristic frequency band envelope sequence and the torque current component sequence. The scale-scaled torque current component sequence is used as the load response variable, and the characteristic frequency band envelope sequence is used as the acoustic dissipation variable. The dissipation load ratio feature mapping is extracted, and the acoustic impedance ratio sequence is output.

5. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 4, characterized in that: The specific process of extracting the trough distortion nodes of the acoustic impedance ratio sequence within a time sliding window and constructing the powder flow impedance evolution sequence is as follows: A moving data sliding window is constructed along the time axis. Morphological bottom-hat transformation is performed on the acoustic impedance ratio sequence within the moving data sliding window to extract the abnormal concave valley nodes of the morphological bottom-hat transformation sequence. Extract the waveform deformation curvature feature vectors before and after the abnormal concave valley nodes, filter free fluctuation nodes according to the preset curvature gradient reference interval, and cut off the valley distortion nodes. Arrange the trough distortion nodes in time order, perform dynamic impedance trend envelope reconstruction, and output the powder flow impedance evolution sequence.

6. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 1, characterized in that: The preset reference batch flow impedance sequence is invoked, and a dynamic time warping algorithm is used to nonlinearly align the powder flow impedance evolution sequence with the reference batch flow impedance sequence. The specific process for calculating the time warping offset is as follows: Construct a state evolution difference mapping matrix between the powder flow impedance evolution sequence and the baseline batch flow impedance sequence; In the state evolution difference mapping matrix, a dynamic programming algorithm is applied to search for the optimal state matching path that satisfies the endpoint binding and continuity constraints, and the multidimensional state space mapping feature vector of each aligned node pair on the optimal state matching path is extracted. Based on the multidimensional state space mapping feature vector, the cumulative feature extraction of evolution trajectory deviation is performed, and the temporal regularization offset is output.

7. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 6, characterized in that: When the cumulative mutation rate of the time-order regularization offset exceeds the preset threshold range, the specific process of extracting the acoustic-electric distortion feature vector corresponding to the deviation from the initial time anchor point and constructing the degradation diagnosis label is as follows: Extract the local variation rate feature sequence of temporal regularization offset along the time axis, construct the mutation rate trend curve, perform boundary cross-judgment on the mutation rate trend curve, and mark the first jump node that breaks through the upper limit of the preset threshold interval as the deviation from the starting time anchor point. Extract the neighborhood time cutoff window corresponding to the deviation from the starting time anchor point, extract the working condition evolution time slice in the powder flow impedance evolution sequence, perform flow feature decoupling and separation on the working condition evolution time slice to extract the core deformation component, construct the acoustic and electrical distortion feature vector, and package and reassemble the acoustic and electrical distortion feature vector with the deviation from the starting time anchor point to form a deterioration diagnosis label.

8. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 1, characterized in that: Based on the deviation from the starting time anchor point in the deterioration diagnosis identifier, the temperature and humidity sequence within the corresponding traceability window is extracted from the mixing condition fusion matrix, and the specific process for calculating the microenvironmental humid heat condensation gradient is as follows: The degradation diagnosis markers are analyzed to obtain the deviation from the starting time anchor point. The temperature and humidity sequence with a preset span preceding the deviation from the starting time anchor point is extracted in reverse along the time axis to form a source traceability data slice. The temperature and relative humidity waveforms in the source data slices are separated, and the dew point temperature is nonlinearly fitted by combining the saturated water vapor pressure parameter to generate the near-wall condensation critical characteristic curve. Extract the condensation phase transition approximation rate characteristics of the near-wall condensation critical characteristic curve and the actual temperature waveform, and output the microenvironment humid heat condensation gradient.

9. The intelligent diagnostic method for abnormalities in veterinary drug powder production according to claim 8, characterized in that: The specific process of matching the microenvironmental humid-thermal condensation gradient with the acoustic-electric distortion feature vector by inputting it into a preset flow phase change compensation strategy library, generating a set of equipment control instructions, and sending them to the programmable logic controller is as follows: The microenvironment humid heat condensation gradient and acoustic-electric distortion feature vector are fused and reconstructed using multi-dimensional operating condition features to build a multi-dimensional state addressing code. The operating condition similarity feature optimization and matching of the multi-dimensional state addressing code is performed in the flow phase change compensation strategy library to match the target compensation control logic. The target compensation control logic is parsed into a set of equipment control instructions that includes spindle speed adaptive load reduction parameters and high-speed bridge breaking parameters of the fly knife. The set of equipment control instructions is encapsulated into industrial fieldbus protocol messages and sent to the programmable logic controller to perform dynamic intervention.

10. An intelligent diagnostic system for abnormalities in veterinary drug powder production, used to execute the intelligent diagnostic method for abnormalities in veterinary drug powder production as described in any one of claims 1-9, characterized in that, Includes the following modules: The multi-source data fusion module is used to synchronously collect the three-phase transient electrical parameter sequence of the powder mixer spindle, the high-frequency acoustic emission waveform sequence of the cylinder wall, and the temperature and humidity sequence of the feeding port. The Park transform is performed on the three-phase transient electrical parameter sequence to extract the torque and current component sequence, and the torque and current component sequence, the high-frequency acoustic emission waveform sequence, and the temperature and humidity sequence are aligned according to the timestamp sequence to construct the mixing condition fusion matrix. The flow impedance reconstruction module is used to extract the characteristic frequency band envelope sequence of the high-frequency acoustic emission waveform sequence in the mixing condition fusion matrix, and perform time-frequency domain cross-mapping with the torque current component sequence to construct the acoustic impedance ratio sequence; within the time sliding window, the trough distortion nodes of the acoustic impedance ratio sequence are extracted to construct the powder flow impedance evolution sequence. The abnormal deviation diagnosis module is used to call the preset reference batch flow impedance sequence, and use the dynamic time warping algorithm to nonlinearly align the powder flow impedance evolution sequence with the reference batch flow impedance sequence to calculate the time warping offset; when the cumulative mutation rate of the time warping offset exceeds the preset threshold range, the acoustic and electrical distortion feature vector corresponding to the deviation from the starting time anchor point is extracted to form a deterioration diagnosis label. The traceability and compensation control module is used to extract the temperature and humidity sequence within the corresponding traceability window from the mixing condition fusion matrix based on the deviation from the starting time anchor point in the deterioration diagnosis identifier, and calculate the microenvironmental humid heat condensation gradient; the microenvironmental humid heat condensation gradient is matched with the acoustic and electrical distortion feature vector by inputting it into the preset flow phase change compensation strategy library, generating the equipment control instruction set and sending it to the programmable logic controller.

Images