Centrifugal parameter self-adjusting method of centrifugal machine for hulling oats

Abstract

The invention relates to the field of centrifugal parameter self-adjustment of centrifugal hulling equipment, in particular to a centrifugal parameter self-adjustment method of a centrifugal machine for hulling oats. The method comprises the following steps: acquiring a centrifugal shelling equipment identifier, a rotor assembly model and a variable frequency driving parameter, and constructing a session configuration packet; synchronously acquiring rotating speed feedback, feeding position data and quality agent data based on the session configuration packet, and generating an alignment feature packet; executing heuristic sequence arrangement and segmented operation sampling, and calculating sensitivity and a tradeoff index to form a fingerprint packet; and performing safety gating, quality gating and convergence gating judgment on the basis of the fingerprint packet, generating and selecting a candidate centrifugal parameter group under gating constraint, completing abnormal typing and processing action sequence generation, and outputting an updated centrifugal parameter group and an archiving record packet. According to the method, self-adjustment and process traceability of the centrifugal parameters in the operation process are achieved through unified collection, calculation and judgment links, and the stability and consistency of shelling operation are improved.

Description

Technical Field

[0001] This invention relates to the field of self-adjustment of centrifugal parameters in centrifugal dehulling equipment, and more particularly to a method for self-adjustment of centrifugal parameters in a centrifuge for dehulling oats. Background Technology

[0002] In the field of self-adjustment of centrifugal parameters in centrifugal dehulling equipment, existing solutions for self-adjustment of centrifugal parameters in centrifuges for dehulling oats typically revolve around the preset and switching of centrifugal parameter sets. These solutions rely on operational experience or fixed process windows to configure rotation speed and feeding cycle time, and obtain quality-related information through external observation or offline sampling during operation. However, these methods suffer from limitations such as difficulty in characterizing the coupling relationship between the centrifugal parameter set and actual operating conditions, inconsistent triggering conditions for parameter switching, and incomplete recording links during operation. Existing methods often collect or process rotation speed feedback, feeding position data, and quality proxy data in a dispersed manner, lacking a unified sampling caliber and field mapping driven by session configuration packages. Furthermore, they fail to adequately constrain the temporal consistency of collected data. In scenarios where different batches of materials fluctuate and operational disturbances coexist, issues such as asynchronous rotation speed feedback and feeding position data, and delayed or distorted quality proxy data easily arise. This makes it difficult to achieve stable implementation of generating fingerprint packages based on aligned feature packages and outputting updated centrifugal parameter sets and archived record packages under gating constraints. For the joint processing of session configuration packages, alignment feature packages, and fingerprint packages, existing technologies generally lack engineering constraints that form a consistent link between trial sequence arrangement, segmented operation sampling, sensitivity calculation, and trade-off index calculation. They also lack a process management mechanism that integrates safety gating, quality gating, and convergence gating into the same judgment caliber and connects them with candidate centrifugation parameter group generation, comprehensive scoring and sorting, perturbation distance constraint selection, anomaly fingerprint construction, anomaly classification, and the generation of disposal action sequences. This makes it difficult to form a consistent process of collection-alignment-calculation-judgment-selection-disposal-archiving during the continuous operation of centrifuges for dehulling oats. As a result, the basis for updating centrifugation parameter groups lacks traceability, the field caliber and temporal context of archived record packages are incomplete, and problems such as parameter rollback, inconsistent disposal actions, and difficulties in process review occur during production operation. Summary of the Invention

[0003] To solve the above-mentioned technical problems, the present invention provides a method for self-adjusting centrifugation parameters of a centrifuge for dehulling oats, comprising:

[0004] S100: Obtain the centrifugal dehulling equipment identifier, rotor assembly model, and frequency converter drive parameters; perform centrifugal parameter group definition, candidate parameter set library construction, and state machine template registration processing to obtain the session configuration package.

[0005] S200. Based on the session configuration package, synchronously collect and process rotation speed feedback, feeding position data and quality proxy data to generate an alignment feature package;

[0006] S300. Based on the alignment feature package, perform trial sequence arrangement and segmented sampling processing, and perform sensitivity calculation and trade-off index calculation processing to generate a fingerprint package;

[0007] S400. Based on the fingerprint package, perform security gating, quality gating, and convergence gating processing. Under the constraints of the gating results, generate candidate centrifuge parameter groups, sort comprehensive scores, select disturbance distance constraints, construct abnormal fingerprints, classify abnormalities, and generate action sequences for handling actions. Generate updated centrifuge parameter groups and archived record packages. The updated centrifuge parameter groups include speed setting values, speed ramps, feeding trajectories, and gating threshold groups encapsulated from the target candidate centrifuge parameter groups corresponding to the fields selected from the candidate centrifuge parameter groups. Add an updated centrifuge parameter group version identifier to generate archived record packages. The archived record packages include session configuration package identifiers, alignment feature package summaries, fingerprint package summaries, gating result field sets, updated centrifuge parameter groups, abnormal classification results, and action sequence summaries. Perform partitioned statistical processing to update the parameter application tags and gating threshold group version identifiers in the candidate parameter set library.

[0008] Furthermore, the process of defining the centrifugation parameter set includes:

[0009] The centrifugal parameter group definition process includes generating a speed reachable domain constraint field based on the available ramp gears in the frequency converter drive parameters and the rated speed level in the rotor assembly model; generating a feeding reachable domain constraint field based on the reachable domain description field in the feeding actuator parameters; performing a consistency comparison process between the three subsets of the gated threshold group and the above-mentioned reachable domain constraint fields to exclude cases where the action range required by the threshold group exceeds the reachable domain; and writing an exception reason code in the session registration area and triggering the confirmation process of the human-machine interaction unit when the consistency comparison fails.

[0010] Furthermore, the process of constructing the candidate parameter set library and registering the state machine template includes:

[0011] The candidate parameter set construction process includes generating a basic candidate set based on rotor assembly model, frequency converter drive parameters, and feed actuator parameters. This set consists of conservative candidate items, conventional candidate items, impact-enhanced candidate items, and crushing-suppressed candidate items. For each candidate centrifugal parameter group item, a parameter distance benchmark field is calculated for subsequent comprehensive scoring and ranking, used to calculate the switching cost penalty. Simultaneously, an item index table is constructed and indexed according to applicable condition tags, including raw grain fluctuation level tags, equipment wear level tags, vibration sensitivity level tags, and feed response level tags. A state machine template registration process is then performed, which includes reading three subsets of the gating threshold group. It generates a gating event mapping table to map load fluctuation index threshold out-of-bounds, vibration frequency band energy index threshold out-of-bounds, and temperature rise gradient index threshold out-of-bounds as gating out-of-bounds events; to map shelled particle ratio threshold out-of-bounds and fragmented particle ratio threshold out-of-bounds as subclass reason codes of gating out-of-bounds events; to map continuous period consistency threshold satisfaction as convergence achievement event; and to map continuous period consistency threshold violation accompanied by abnormal fingerprint triggering condition as convergence violation event. It also writes the library version identifier of the candidate parameter set library into the context reference field of the state machine template and writes the health status field of the sensor list into the precondition field of the state machine template to constrain the entry conditions of the trial state and the stable state, and generates a session configuration package.

[0012] Furthermore, the process of synchronously collecting and processing rotation speed feedback, feed position data, and quality agent data includes:

[0013] Speed ​​feedback is obtained by the variable frequency drive interface unit from the variable frequency drive's internal speed estimate or encoder feedback value. Load signal is obtained by the variable frequency drive interface unit from the variable frequency drive's output current, output power, or torque estimate. Vibration signal is obtained by the accelerometer located near the bearing housing of the centrifugal dehulling equipment, which collects a triaxial or single-axis acceleration sequence. Temperature rise signal is obtained by the bearing temperature sensor or housing temperature sensor. Feeding position data is obtained by the feeding actuator drive unit from the current position, target position, following error, and interlock status. Quality proxy data includes the separated grain image frames acquired by the image acquisition device, which are then processed by the edge computing unit to perform target segmentation and counting, outputting the proportion of shelled grains and the proportion of broken grains.

[0014] Furthermore, the process of exploratory sequence arrangement and segmented sampling includes:

[0015] The trial sequence arrangement process includes extracting a set of candidate centrifugal parameter groups that satisfy the perturbation distance constraint from the candidate parameter set library in the session configuration package. The perturbation distance constraint is calculated by the parameter distance between the candidate centrifugal parameter group and the current centrifugal parameter group. The parameter distance is based on the difference measure of the speed set value, speed ramp, feeding trajectory and gate threshold group, and is organized into a sequence structure of two or more running segments. For each running segment, a running segment description field is generated, including the running segment number, target centrifugal parameter group reference number, segment length configuration, sampling window configuration and interlock condition code. Segmented running sampling processing is performed. The segmented running sampling processing includes sending the target centrifugal parameter group to the frequency conversion drive interface unit and the feeding execution mechanism drive unit segment by segment according to the trial sequence structure. In each running segment, a pre-stabilization window, an effective sampling window and a post-buffer window are set to extract vector fragments from the aligned feature package to form running segment sample pairs, and a sample consistency mark field is generated.

[0016] Furthermore, the process of performing sensitivity calculation and trade-off index calculation includes:

[0017] The sensitivity calculation process includes extracting a set of sub-features sensitive to the speed ramp from the state feature vector fragments corresponding to the two operating segments. These sub-features include speed fluctuation features, load fluctuation features, and vibration frequency band energy features. The response slope vector is calculated based on the speed ramp difference and used as the sensitivity field. The trade-off index calculation process is then performed. This process includes reading the fragmentation ratio from the quality proxy vector fragments corresponding to the two operating segments and calculating the quality response slope as the trade-off index field. Simultaneously, consistency checks are performed, and low confidence is marked when the shelled statistics show abnormal jumps. A fingerprint packet is generated, which includes a session configuration packet identifier, an alignment feature packet digest reference number, a trial sequence structure reference number, a set of operating segment description fields, a sensitivity field, a trade-off index field, a state feature vector digest, a quality proxy vector digest, a quality tag digest field, and a parameter association field.

[0018] Furthermore, the processes for safety gating, quality gating, and convergence gating include:

[0019] The security gating decision process includes extracting load fluctuation indicators, vibration frequency band energy indicators, and temperature rise gradient indicators from the state feature vector summary of the fingerprint packet, and performing outlier removal processing based on the quality tag summary field. Subsequently, it reads the security gating threshold group from the gating threshold group and performs a combined strategy of same-direction comparison and hysteresis comparison for threshold comparison, generating a security gating decision result field and recording the trigger cause code and time window index range. This is followed by quality gating decision processing, which includes extracting a shelled particle ratio summary and a fragmented particle ratio summary from the quality proxy vector summary of the fingerprint packet, and performing a setting based on the quality tag summary field. The signal correction process generates shell-gated input values ​​and fragmented gated input values ​​through weighted aggregation. Then, it reads the quality gating threshold group from the gating threshold group and performs threshold comparison to generate a quality gating judgment result field. When the image acquisition is abnormal, a weight reduction judgment flag is written. Convergence gating judgment processing is then performed. The convergence gating judgment processing includes extracting the gating result sequence of the most recent period under the same session configuration package identifier from the operation log area, performing consistency statistics on the consistency interval length and the number of state flips, and comparing it with the convergence gating threshold group in the gating threshold group to generate a convergence gating judgment result field. When the number of flips exceeds the limit, a gating jitter flag is written.

[0020] Furthermore, the process of generating candidate centrifugation parameter sets and ranking them by comprehensive score includes:

[0021] The candidate centrifugation parameter group generation process includes parsing the initial selection set of candidate centrifugation parameter groups from the parameter association field of the fingerprint package, performing gating filtering to remove entries rejected by the safety gating judgment result field and entries rejected by the quality gating judgment result field with reduced weight, and enabling a conservative generation mode when the convergence gating judgment result field is rejected to retain only entries with small parameter distances and lock the feeding trajectory unchanged, and performing comprehensive scoring and sorting processing. The comprehensive scoring and sorting processing includes calculating a comprehensive score for each candidate centrifugation parameter group set. The comprehensive score includes a quality deviation penalty calculated by the trade-off index field and the quality gating judgment result field, a risk penalty calculated by the sensitivity field and the safety gating judgment result field, and a switching cost penalty calculated by the parameter distance between the candidate centrifugation parameter group and the current centrifugation parameter group. The three types of penalties are weighted and synthesized to output the comprehensive scoring and sorting result field.

[0022] Furthermore, the processes of perturbation distance constraint selection, anomaly fingerprint construction, and anomaly classification include:

[0023] The disturbance distance constraint selection process includes traversing candidate entries under the constraint of the comprehensive score ranking result field, calculating the parameter distance and comparing it with a preset threshold, selecting the first candidate entry that meets the distance constraint as the target candidate centrifugal parameter group and recording the candidate centrifugal parameter group selection result field; otherwise, triggering conservative backoff processing to maintain the current centrifugal parameter group, and performing abnormal fingerprint construction and abnormal classification processing. The abnormal fingerprint construction process includes parsing vibration spectrum peak features, load mutation morphology features and feed position fluctuation features from the state feature vector summary of the fingerprint package to construct an abnormal fingerprint field. The abnormal classification processing includes identifying blockage anomalies, slippage anomalies, imbalance anomalies and feed fluctuation anomalies based on the classification rule engine, and outputting the abnormal classification result and classification rule version number.

[0024] Furthermore, the process of generating and processing the action sequence includes:

[0025] The process of generating the action sequence includes generating a summary field of the action sequence based on the anomaly classification result, the set of gating result fields, and the candidate centrifuge parameter group selection result fields. This summary field includes actions such as reducing material, slowing down, resetting and probing, slowing down and locking, or maintenance prompts. The process also generates and updates the centrifuge parameter group.

[0026] The key innovations of this invention include:

[0027] (1) Based on the centrifugal desizing equipment identifier, the rotor assembly model and the frequency conversion drive parameters, the session configuration package is formed in S100 by defining the centrifugal parameter group, constructing the candidate parameter set library and registering the state machine template, so that each subsequent step takes the session configuration package as the only operating baseline and organizes the acquisition and calculation link under the same terminology.

[0028] (2) In S200, based on the session configuration package, the rotation speed feedback, the feeding position data and the quality agent data are synchronously acquired and processed to generate the alignment feature package, so that the alignment feature package carries multi-source operating status and quality agent information under the same acquisition session and the same field caliber, and serves as the direct input object of S300.

[0029] (3) In the continuous link between S300 and S400, the alignment feature packet first drives the arrangement of the trial sequence and the segmented sampling to complete the sensitivity calculation and trade-off index calculation to generate the fingerprint packet. Then, based on the fingerprint packet, security gating judgment, quality gating judgment and convergence gating judgment are performed. Under the constraints of the gating results, candidate centrifugation parameter group generation, comprehensive score sorting, disturbance distance constraint selection, abnormal fingerprint construction, abnormal classification and disposal action sequence generation are completed. Finally, the updated centrifugation parameter group and the archived record packet are output, thus forming a closed-loop data organization and control judgment mechanism of "fingerprint-gating-selection-disposal-archiving".

[0030] The following are its main beneficial effects:

[0031] (1) In view of the common problems in the background technology that the coupling relationship between centrifugal parameter group and actual working conditions is difficult to characterize, the parameter switching trigger condition is inconsistent and the operation process recording link is incomplete, by establishing the session configuration package in S100 and incorporating the centrifugal parameter group definition, candidate parameter set library construction and state machine template registration into the same session configuration, the centrifugal parameter group, the candidate parameter set library and the state machine template are referenced and updated under the same operating baseline. The session configuration package provides a consistent data organization entry for the subsequent synchronous acquisition processing and alignment feature package generation in S200, and provides a verifiable configuration context for the subsequent formation of the archived record package.

[0032] (2) In view of the common problems of insufficient time consistency, lag or distortion of quality proxy data caused by the scattered collection or step-by-step processing of speed feedback, feeding position data and quality proxy data in the background technology, by performing synchronous acquisition processing based on the session configuration package in S200 and generating the alignment feature package, the speed feedback and feeding position data are established in correspondence with the quality proxy data under the same acquisition session. The alignment feature package is used as the input of S300 so that the state information and quality proxy information on which the trial sequence arrangement and segmented operation sampling depend have a consistent organizational caliber, thereby reducing the drift of the judgment and control link caused by the inability to align cross-source information.

[0033] (3) In response to the common problems in the background technology that there is a lack of a consistent link between the arrangement of trial sequences, segmented operation sampling, sensitivity calculation and trade-off index calculation, and a lack of a process management mechanism that connects safety gate judgment, quality gate judgment and convergence gate judgment with the generation of candidate centrifuge parameter groups and the generation of disposal action sequences, the following continuous link is established: "The alignment feature package generates the fingerprint package - the fingerprint package triggers the gate judgment - under the gate result constraint, the candidate centrifuge parameter group is generated, the comprehensive score is sorted, the disturbance distance constraint is selected, the abnormal fingerprint is constructed, the abnormal classification and disposal action sequence is generated - the updated centrifuge parameter group and the archived record package are output". This makes the parameter update no longer dependent on external observation or offline sampling to form a traceable judgment basis and disposal path in the same operation link. The archived record package provides a structured record for subsequent operation review and consistency verification, thereby forming a consistent process of collection-alignment-calculation-judgment-selection-disposal-archiving in the continuous operation scenario of centrifuge for dehulling oats. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating a method for self-adjusting centrifugation parameters in a centrifuge for dehulling oats, as provided in an embodiment of this application. Detailed Implementation

[0035] Reference Figure 1 This is a flowchart illustrating a self-adjusting method for centrifugation parameters in a centrifuge for dehulling oats, provided in an embodiment of the present invention. The process may include at least steps S100-S400:

[0036] S100: Obtain the centrifugal dehulling equipment identifier, rotor assembly model, and frequency converter drive parameters; perform centrifugal parameter group definition, candidate parameter set library construction, and state machine template registration processing to obtain the session configuration package.

[0037] S200. Based on the session configuration package, synchronously collect and process rotation speed feedback, feeding position data and quality proxy data to generate an alignment feature package;

[0038] S300. Based on the aligned feature package, perform trial sequence arrangement and segmented sampling operations, and perform sensitivity calculation and trade-off index calculation to generate a fingerprint package;

[0039] S400. Based on the fingerprint package, perform security gating, quality gating and convergence gating processing, and under the constraints of the gating results, perform candidate centrifugation parameter group generation, comprehensive score sorting, disturbance distance constraint selection, abnormal fingerprint construction, abnormal classification and handling action sequence generation processing, and generate updated centrifugation parameter group and archived record package.

[0040] S100: Obtain the centrifugal dehulling equipment identifier, rotor assembly model, and frequency converter drive parameters; perform centrifugal parameter group definition, candidate parameter set library construction, and state machine template registration processing to obtain the session configuration package.

[0041] This step is executed by the control unit of the centrifugal dehulling equipment under trigger conditions such as power-on initialization, batch feeding, maintenance reset, or convergence gate failure. The control unit consists of an industrial controller, a frequency converter drive interface unit, a feeding actuator drive unit, a data acquisition and time service unit, and a configuration management unit, and maintains a configuration read / write channel with the human-machine interface unit. Specifically, the centrifugal dehulling equipment identifier comes from the equipment nameplate code or the equipment registration record embedded in the controller. The equipment identifier includes four types of fields: equipment serial number, model code, installation point number, and controller firmware version number. After reading the equipment identifier, the configuration management unit generates a session configuration package identifier and writes it to the local, unmodifiable session registration area. The session registration area is stored separately from the operation log area to maintain the auditability of the session traceability link. The rotor assembly model is obtained from the rotor assembly assembly record or the identification code on the rotor assembly. The rotor assembly model includes the rotor structure type, rotor quality grade, rated speed grade, rotor radius setting, shell ring structure setting, fastener torque setting, and balance calibration batch number. The configuration management unit establishes a binding relationship between the rotor assembly model and the equipment identifier and generates an assembly snapshot field, which is written into the session configuration package. This is used for subsequent steps to solidify the acquisition aperture of speed feedback, load signal, and vibration signal under the same time reference. The variable frequency drive parameters are derived from the inverter parameter table and the operation license table. These parameters include the speed setpoint channel type, speed feedback channel type, acceleration / deceleration ramp position, torque limiting position, overcurrent protection position, communication refresh cycle, and fault reset strategy code. The variable frequency drive interface unit performs a validity check on these parameters. This validity check includes three types of actions: parameter range check, parameter mutual exclusion check, and drive capability check. The parameter mutual exclusion check is used to exclude conflicting control channel combinations from being enabled simultaneously within the same session. The drive capability check is used to map the achievable speed limit to the available ramp position by comparing the rated speed level of the rotor assembly model. The mapping result is written into the session configuration package as a boundary condition when defining the centrifugal parameter group. The parameters of the feeding actuator are obtained from the parameter table on the driver nameplate of the feeding actuator or the actuator registry of the controller. The parameters of the feeding actuator include the actuator type, position feedback resolution, stroke endpoint calibration value, drive response delay, maximum step rate, minimum resolvable step size, and action interlock condition code. After reading the above parameters, the feeding actuator drive unit performs consistency verification processing on the position endpoint. The consistency verification includes the integer division relationship verification between the stroke endpoint calibration value and the position feedback resolution, the consistency verification of the limit signal of the endpoint homing action, and the coupling relationship verification between the interlock condition code and the safety gating input. After the verification is passed, the feeding actuator parameters are solidified into the reachable domain description field of the feeding trajectory and written into the session configuration package.The sensor list is derived from the sensor registry and wiring topology table. It includes six channel entries: speed feedback channel, load signal channel, vibration signal channel, temperature rise signal channel, feed position data channel, and quality proxy data channel. Each channel entry includes a channel number, installation location description, sampling rate upper limit, range setting, calibration coefficient version number, timestamp source type, and health status field. The data acquisition and time service unit performs channel self-checks on the sensor list and generates a self-check result snapshot field. Self-check actions include wire breakage detection, zero-point drift detection, range exceedance detection, and timestamp monotonicity detection. The self-check result snapshot field is written into the session configuration package and serves as a prerequisite source record for subsequent gating threshold groups. The gated threshold groups are derived from the equipment process database or threshold configuration tables authorized by maintenance personnel. These groups comprise three subsets: safety gated threshold groups, quality gated threshold groups, and convergence gated threshold groups. The safety gated threshold groups include load fluctuation index thresholds, vibration frequency band energy index thresholds, and temperature rise gradient index thresholds. The quality gated threshold groups include shelled particle ratio thresholds and fragmented particle ratio thresholds. The convergence gated threshold groups include continuous period consistency thresholds and convergence freeze release thresholds. The configuration management unit performs version verification and authorization verification on the gated threshold groups. Version verification checks the compatibility between the threshold group version identifier and the firmware version number and process library version number in the equipment identifier. Authorization verification checks the matching relationship between the signature field of the threshold configuration table and the controller's authorized whitelist. Upon successful verification, the gated threshold group version identifier is written into the session configuration package and locked as the only effective version for this session. This locking action remains unchangeable until the session ends. If a maintenance request for a change occurs, it triggers session termination and the opening of a new session, thus maintaining clear audit boundaries.

[0042] After completing the above acquisition actions, the control unit enters the centrifugal parameter group definition process. The centrifugal parameter group is the minimum core parameter set of this invention, which includes four types of parameter fields: speed setpoint, speed ramp, feeding trajectory, and gate threshold group. Among them, the speed setpoint is used to describe the target speed given by the variable frequency drive interface unit, the speed ramp is used to describe the ramp gear selection during acceleration and deceleration, the feeding trajectory is used to describe the position command sequence of the feeding actuator on the time axis, and the gate threshold group is used to describe the comparison boundaries of safety gate judgment, quality gate judgment, and convergence gate judgment. Specifically, the configuration management unit first generates a speed reachability constraint field based on the available ramp gears in the frequency converter drive parameters and the rated speed level in the rotor assembly model. Then, it generates a feeding reachability constraint field based on the reachability description field in the feeding actuator parameters. Subsequently, it performs a consistency comparison process between the three subsets of the gated threshold group and the aforementioned reachability constraint fields. The consistency comparison is used to exclude cases where the required action range of the threshold group exceeds the reachability range. If the consistency comparison fails, an exception reason code is written to the session registration area and the confirmation process of the human-machine interaction unit is triggered. After the confirmation process is passed, the threshold group is still locked, but the corresponding out-of-bounds item is marked as a disabled item. The disabled item is written to the threshold effectiveness mask field in the session configuration package. Subsequently, the configuration management unit generates a centrifugal parameter group structure and writes it into the session configuration package. The structure contains five types of fields: parameter group identifier, parameter item list, parameter item value range, parameter item source record, and parameter item effective condition code. The parameter item source record records the source type from the equipment process database, the operation and maintenance system, or the factory default. The parameter item effective condition code records the dependency of the sensor self-test result snapshot field on the parameter item. The dependency is used to constrain the parameter update action in subsequent steps when the sensor channel is in an unhealthy state.

[0043] Further, the process of constructing a candidate parameter set library is performed. The candidate parameter set library is a core logic component of this invention. Its structure consists of a set of candidate parameter entries and an entry index table. The candidate parameter entry set comprises multiple candidate centrifugal parameter group entries. Each candidate centrifugal parameter group entry includes four types of fields: speed setpoint, speed ramp, feeding trajectory, and gate threshold group reference number, along with applicable condition labels and disabled condition labels. Specifically, the configuration management unit generates a basic candidate set based on the rotor assembly model, frequency converter drive parameters, and feeding actuator parameters. The basic candidate set consists of conservative candidate entries, conventional candidate entries, impact-enhanced candidate entries, and crushing-suppressed candidate entries. Conservative and crushing-suppressed candidate entries are formed using lower speed setpoints or gentler speed ramps combined with smoother feeding trajectories. Impact-enhanced candidate entries are formed using higher speed setpoints or steeper speed ramps combined with restricted feeding trajectories. Conventional candidate entries fall between these two categories. The above-mentioned entry generation action is completed by the candidate generation submodule. During the generation process, the candidate generation submodule is constrained by the speed reachable domain constraint field and the feed reachable domain constraint field, and calculates the parameter distance benchmark field for each candidate centrifugal parameter group entry. The parameter distance benchmark field is used for the calculation of switching cost penalty in the subsequent comprehensive scoring and ranking. The candidate parameter set library also contains an entry index table, which is indexed according to applicable condition tags. The applicable condition tags include four types of tags: raw grain fluctuation level tag, equipment wear level tag, vibration sensitivity level tag, and feed response level tag. Among them, the raw grain fluctuation level tag is consistent with the sensitivity calculation caliber in the fingerprint package generated in the subsequent steps, the equipment wear level tag is linked to the temperature rise gradient index threshold, the vibration sensitivity level tag is linked to the vibration frequency band energy index threshold, and the feed response level tag is linked to the drive response delay and maximum step rate of the feed actuator parameters. Understandably, the minimum set of the candidate parameter set library consists of conservative candidate entries and regular candidate entries. This minimum set satisfies the integrity of the four types of fields defined in the centrifugal parameter group and meets the locking conditions of the gating threshold group. Impact enhancement candidate entries and fragmentation suppression candidate entries are preferred extended entries, used to provide a more granular candidate space when the threshold for the proportion of shelled particles or the threshold for the proportion of broken particles exceeds the limit in the quality gating judgment. After the candidate parameter set library is constructed, the configuration management unit writes the candidate parameter set library into the session configuration package and generates a library version identifier. The library version identifier, together with the gating threshold group version identifier and the sensor list calibration coefficient version number, constitutes the configuration triplet version field of this session. The configuration triplet version field is written into the session registration area and simultaneously written into the archive reservation area. The archive reservation area is directly referenced when generating the archive record package in subsequent steps, thereby forming an auditable version link across sessions.

[0044] Furthermore, the state machine template registration process is executed. The state machine template is the scheduling skeleton used by this invention to drive the fully automated operation of the entire process. Its structure consists of a state set, an event set, a transition rule set, an entry action set, and an exit action set. The state set includes startup state, trial state, stable state, anomaly handling state, archive state, and shutdown state. The event set includes power-on event, batch change event, gating out-of-bounds event, convergence achievement event, convergence failure event, and maintenance reset event. The transition rule set describes the combined triggering relationship between states and events and includes interlocking condition codes. Specifically, when registering the state machine template, the configuration management unit first reads three subsets of the gating threshold group and generates a gating event mapping table. This table maps out-of-bounds thresholds for load fluctuation, vibration frequency band energy, and temperature rise gradient to gating out-of-bounds events. It also maps out-of-bounds thresholds for the proportion of shelled particles and the proportion of fragmented particles to subclass cause codes for these events. Furthermore, it maps continuous periodic consistency threshold satisfaction to convergence achievement events and continuous periodic consistency threshold violation accompanied by abnormal fingerprint triggering conditions to convergence failure events. Subsequently, the state machine template registration submodule writes the library version identifier of the candidate parameter set library into the context reference field of the state machine template and writes the health status field of the sensor list into the precondition field. The precondition field constrains the entry conditions for the trial state and the stable state. If any critical channel in the sensor list is in an unhealthy state, the state machine template will jump to the shutdown state and record the cause code. The minimum set of critical channels here includes the speed feedback channel, load signal channel, vibration signal channel, and feed position data channel. The temperature rise signal channel and quality proxy data channel are preferred channels. When a preferred channel is missing, the state machine template is still allowed to enter the trial state, but the gating threshold group reference number is written into the threshold effectiveness mask field and the corresponding gating comparison item is disabled, thus keeping the process runnable while retaining complete limitation records. After the state machine template is registered, the configuration management unit generates a state machine template version identifier and establishes an association record with the configuration triplet version field. The association record is written to the session configuration package and synchronously written to the operation log area. The operation log area records the timestamp, trigger event, interlock condition code, entry action number, and exit action number of each state transition, thereby meeting the requirements for stability and auditability in industrial automation scenarios.

[0045] In one engineering implementation scenario, a centrifugal dehulling device is located in the oat dehulling section of a grain processing workshop. The control unit is connected to the frequency converter via an industrial bus and to the feeding actuator via pulse or bus. The vibration signal channel in the sensor list is provided by an accelerometer installed near the bearing housing on the machine casing. The load signal channel is provided by the current or power feedback of the frequency converter. The speed feedback channel is provided by the frequency converter speed estimation or encoder feedback. The feeding position data channel is provided by the displacement feedback of the feeding actuator. The quality proxy data channel is provided by the image frame acquisition device of the separated grains. The temperature rise signal channel is provided by the bearing temperature sensor. Before starting the machine, the workshop operator selects the batch number of the raw grain and enters the basic process file number through the human-machine interface unit. After receiving the start-up event, the control unit enters this step, reads the equipment identifier and completes the session registration, reads the rotor component model and completes the assembly snapshot fixation, reads the frequency converter drive parameters and completes the drive capability mapping, reads the feeding actuator parameters and completes the stroke consistency verification, reads the sensor list and completes the channel self-test, and reads the gate threshold group and completes the version lock. Subsequently, the configuration management unit defines centrifugal parameter groups and fixes the value range of parameter items. The candidate generation submodule constructs a candidate parameter set library and generates a library version identifier. The state machine template registration submodule registers the state machine template and generates a template version identifier, ultimately forming a session configuration package. At the end of this step, the session configuration package is encapsulated and written to the control unit's session buffer, and simultaneously written to an index pointer in the archive reserved area. This index pointer contains the session configuration package identifier, a configuration triplet version field, and a state machine template version identifier. Understandably, the session buffer serves as the configuration source for subsequent data acquisition and gating decisions. Its writing action is constrained by interlock condition codes. The interlock condition codes require that the inverter communication refresh cycle be within a valid range and that the feeding actuator be in a zero-return completed state. If the interlock condition code is not met, the session configuration package remains in a pending state and records the waiting reason code until the interlock condition code is met, at which point it transitions to an effective state and triggers the next step.

[0046] The output of this step is a session configuration package. This package is explicitly recorded as an output field name in the main text and serves as the input location for the next step, corresponding to the "session configuration package" in S200. Subsequent steps extract the sampling rate and field mapping from the session configuration package and perform synchronous acquisition, time alignment, and noise reduction processing to form an aligned feature package, which is then advanced to the fingerprint package and gating decision link. Understandably, this step maintains cross-step consistency with S200, S300, and S400 through the session configuration package identifier and the configuration triplet version field. The runtime log records the reference relationships between each step, thus forming a complete closed-loop connection.

[0047] In summary, this step achieves the following technical results: It unifies the registration and solidifies the centrifugation parameter group, candidate parameter set library, and state machine template into a session configuration package. This session configuration package provides a unique configuration source for subsequent synchronous data acquisition and gating decisions. The gating threshold group, sensor list, and version identifier are locked and written to the session registration area in this step, ensuring configuration consistency within the same session boundary in subsequent steps. The candidate parameter set library and state machine template are registered for auditability in this step, allowing subsequent steps to call the corresponding entries and write them to the archive link under gating trigger conditions.

[0048] S200. Based on the session configuration package, synchronously collect and process rotation speed feedback, feeding position data and quality proxy data to generate an alignment feature package;

[0049] This step is triggered by the state machine template when the control unit enters a trial or stable state. The triggering is based on the entry action number and interlock condition code from the session configuration package. The interlock condition code satisfies the following conditions: the frequency converter drive interface unit is in an enabled state, the feeding actuator drive unit is in a controllable state, and the data acquisition and time service unit is in a clock-locked state. Specifically, the session configuration package is a configuration object output by S100 and written to the session buffer of the control unit. It includes fields such as sampling rate and field mapping, sensor list, self-test result snapshot field, gate threshold group version identifier, candidate parameter set library version identifier, and state machine template version identifier. The sampling rate is a set of sampling periods or frame rates defined for speed feedback, load signal, vibration signal, temperature rise signal, feeding position data, and quality proxy data, respectively. The field mapping is a set of mapping relationships from the original acquisition fields of each channel to fields with a unified caliber. The field mapping includes channel number, original field name, target field name, scaling factor, offset correction item, timestamp source type, and quality mark bit width. At the start of this step, the data acquisition and time service unit reads the sampling rate and field mapping from the session configuration package and generates an acquisition plan table. The acquisition plan table schedules the sampling points of each channel according to a unified time base and binds the channel number and target field name to each sampling point, thereby converting multi-source asynchronous acquisition into an alignable sampling task sequence.

[0050] In the synchronous acquisition link, the speed feedback is read from the inverter by the variable frequency drive interface unit, either through bus register read or analog sampling. The original fields of the speed feedback include the internal speed estimate or encoder feedback value of the drive and its corresponding drive-side timestamp. The data acquisition and time service unit converts the original fields into a unified target field according to the field mapping and adds a session configuration packet identifier and channel number. The load signal is formed by the variable frequency drive interface unit reading the inverter's output current, output power, or torque estimate. The original fields of the load signal include the instantaneous value, the sliding window average, and the drive alarm bit. The data acquisition and time service unit performs alarm bit decoding on the original field and writes it into the quality flag field corresponding to the quality flag bit width. The vibration signal is acquired by an accelerometer located near the bearing housing on the casing of the centrifugal dehulling equipment. Acquisition is performed via analog sampling or digital sensor bus sampling. The raw fields of the vibration signal include a triaxial or single-axis acceleration sequence and a sensor-defined health status field. The data acquisition and time service unit performs a consistency comparison of the health status field during acquisition. If the health status field indicates an anomaly, a failure flag is written into the quality flag field, and the sampling point is recorded as an anomaly cause code in the operation log area. The temperature rise signal is acquired by a bearing temperature sensor or a casing temperature sensor. The raw fields of the temperature rise signal include the temperature value and a sensor timestamp. The data acquisition and time service unit performs range out-of-range detection on the temperature value and writes it into the quality flag field when an out-of-range event occurs. The feeding position data is read by the feeding actuator drive unit. The raw fields of the feeding position data include the current position, target position, following error, and interlock status bit. The data acquisition and time service unit writes the following error and interlock status bit into the quality flag field and triggers a gated out-of-range event in the state machine template when the interlock status bit indicates uncontrollable behavior. The quality proxy data includes separated grain image frames, which are acquired by an image acquisition device and transmitted to the control unit or edge computing unit via a data bus. The original fields of the image acquisition device include the image frame sequence, exposure status field, frame number, and acquisition timestamp. Further, in this step, the quality proxy data uses a two-stage processing link to form a unified field: the first stage involves the image acquisition device performing frame integrity verification on the image frame sequence and appending a frame number; the second stage involves the edge computing unit performing target segmentation and counting processing on the image frame sequence, outputting the proportion of shelled grains and the proportion of broken grains, and writing these as unified target fields into the quality proxy data channel.The target segmentation and counting process comprises four sub-processes: image preprocessing, region candidate generation, particle contour extraction, and category counting. Image preprocessing performs brightness normalization and noise suppression on the image frame sequence to form a stable input for segmentation. Region candidate generation performs threshold segmentation on the image frame sequence and generates connected component candidates. Particle contour extraction performs morphological closing operations and edge tracking on the connected component candidates to obtain a set of particle contours. Category counting classifies the particle contour set based on appearance and fragmentation features and counts the proportion of shelled particles and the proportion of fragmented particles. Understandably, the proportion of shelled particles and the proportion of fragmented particles are the minimum set fields of the quality proxy data in this invention, while the image frame sequence, exposure state field, and frame number are preferred extended fields used to provide traceable evidence when quality gating judgments exceed limits in subsequent steps.

[0051] After completing the synchronous acquisition of all channels, the data acquisition and time service unit enters the time alignment processing link. This time alignment processing uses a unified time reference, provided by the time service module within the control unit, and locked as the sole time anchor source for this session during the state machine template registration process in S100. Specifically, the time alignment processing first parses the timestamp source type of each channel. For channels with timestamps from the inverter or sensor itself, the time service module reads the device-side timestamp and establishes a master-slave association record with the control unit clock. This master-slave association record includes offset, drift, and update timestamp. For channels with timestamps sampled locally by the control unit, the time service module directly writes the local sample timestamp into the unified timestamp field. Subsequently, time alignment processing performs resampling alignment on channels with inconsistent sampling rates. Resampling alignment is performed under the constraint of sampling rate and field mapping: For vibration signals with high sampling rates, multiple sampling points are aggregated into a unified time window index using window aggregation and a window summary field is generated. The window summary field includes the window mean square value, peak-to-peak value, and frequency band energy calculation input segment index. For quality proxy data with low sampling rates, the nearest frame strategy is used to bind the proportion of shelled particles and the proportion of broken particles to the nearest unified time window index, and a hold flag is written in the quality flag field. For speed feedback, load signal, temperature rise signal, and feed position data with medium sampling rates, linear interpolation or zero-order hold is used to align to the unified time window index. The applicable condition for linear interpolation is that the time interval between two adjacent points does not exceed a preset alignment threshold, and the applicable condition for zero-order hold is that the time interval between two adjacent points exceeds the alignment threshold but the channel health status field does not indicate an abnormality. After time alignment is completed, the data acquisition and time service unit generates an alignment index table. The alignment index table includes a unified time window index, reference pointers for sampling points in each channel, reference numbers for master-slave related records, and a quality tag summary field. The alignment index table is then written into the metadata area of ​​the subsequent alignment feature package.

[0052] In the denoising process, the denoising process employs differentiated paths for various signals and is constrained by the scaling factor and bias correction term in the field mapping. Specifically, for speed feedback and load signals, the denoising process uses a combination of sliding window mid-range filtering and abrupt change suppression. Sliding window mid-range filtering suppresses short-time spikes, while abrupt change suppression limits jumps that do not meet physical reachability constraints and records the limiting flag. For vibration signals, the denoising process uses a combination of bandpass filtering and pre-cleaning for frequency band energy calculation. The passband of the bandpass filter is determined by the rotor component model and vibration sensitivity level label in the session configuration package. Pre-cleaning for frequency band energy calculation removes sampling segments containing failure markers and records the removal ratio field. For temperature rise signals, the denoising process uses a combination of low-pass smoothing and drift detection. The drift detection method compares the temperature rise signal with the ambient temperature baseline or the temperature baseline of the previous operating segment and generates a drift marker. For the feeding position data, the denoising process uses a combination of error limiting and interlock status consistency verification. The interlock status consistency verification writes a jitter marker for cases where the interlock status position switches frequently in a short period of time. For the quality proxy data, the denoising process uses a combination of frame-level anomaly removal and proportional smoothing. Frame-level anomaly removal removes abnormal frames based on the exposure status field and frame integrity verification results and writes a frame removal marker. Proportional smoothing performs window smoothing on the proportion of shelled particles and the proportion of broken particles and retains the original proportion field for subsequent traceability. Understandably, the minimum set in the above denoising process is the denoising path for speed feedback, load signal, vibration signal, and feeding position data. This minimum set has a direct correspondence with the input fields for subsequent sensitivity calculation and anomaly fingerprint construction. The denoising path for the temperature rise signal and quality proxy data is a preferred extended path linked with the gating threshold group, used to provide more complete available fields in safety gating judgment and quality gating judgment, while retaining anomaly processing traces through the quality marker field.

[0053] In the engineering implementation scenario, after the control unit receives the power-on event and completes S100, it enters the probing state. This step, under the indication of the entry action number, initiates the acquisition plan and begins synchronous acquisition. The frequency converter drive interface unit reads the speed feedback and load signal according to the sampling rate and writes them into the acquisition buffer. The accelerometer outputs the vibration signal according to the sampling rate, which is written into the vibration buffer by the data acquisition and time service unit. The temperature sensor outputs the temperature rise signal according to the sampling rate and writes it into the temperature rise buffer. The feeding actuator drive unit outputs the feeding position data according to the sampling rate and writes it into the feeding buffer. The image acquisition device outputs the separated grain image frames according to the frame rate, and the edge computing unit outputs the proportion of shelled particles and the proportion of broken particles, which are then written into the quality proxy buffer. The data acquisition and time service unit periodically calls the time service module to update the master-slave association record, and performs time alignment on the data in each buffer to generate a unified time window index sequence when the alignment threshold is met. Subsequently, the data acquisition and time service unit performs noise reduction processing by channel and writes the processed unified fields into the aligned data area, while simultaneously writing the quality mark summary field and the alignment index table into the metadata area. If a failure marker or uncontrollable interlock status bit is detected in a critical channel during the acquisition process, the data acquisition and time service unit will write the corresponding abnormal cause code into the operation log area and trigger the gating out-of-bounds event of the state machine template, causing the control unit to enter the abnormal handling state or shutdown state in the next state transition cycle. This triggering path is consistent with the S400 safety gating decision link under the same gating threshold group version identifier.

[0054] Furthermore, the data acquisition and time service unit encapsulates the alignment index table, the denoised unified field sequence, and the quality marker field sequence into an alignment feature package. The alignment feature package structure includes a session configuration package identifier, a unified time window index sequence, a speed feedback sequence, a load signal sequence, a vibration signal summary sequence, a temperature rise signal sequence, a feed position data sequence, a quality proxy data sequence, a quality marker field sequence, and a source record field. This alignment feature package, as the output product of this step, is explicitly recorded as an output field name in the text and written to the step cache area of ​​the control unit. Simultaneously, an alignment feature package summary is registered in the operation log area. The next input position of the alignment feature package corresponds to the "alignment feature package" in S300. Subsequent steps extract the state feature vector and quality proxy vector from the alignment feature package and perform trial sequence arrangement, segmented operation sampling, sensitivity calculation, and trade-off index calculation processing to generate a fingerprint package and advance the gating judgment and candidate centrifugation parameter group update link. Understandably, this step achieves cross-step referencing with S100 through the session configuration package identifier and configuration triplet version field, achieves input-output alignment with S300 through the unified time window index sequence of the alignment feature package, and achieves traceable connection of anomaly handling traces with S400 through the quality mark field sequence and source record field.

[0055] In summary, this step achieves the following technical effects: Under the constraints of the session configuration package, this step completes the synchronous acquisition of multi-source channels and organizes asynchronous data into a unified time window index sequence. The field mapping, time alignment, and denoising processing remain consistent within the same session boundary, and the anomaly handling path is preserved through the quality tag field. The alignment feature package, as the input carrier of the S300, receives the fields required for trial sequence arrangement and sensitivity calculation, and shares the version identifier and source record fields with the subsequent gating decision link.

[0056] S300. Based on the aligned feature package, perform trial sequence arrangement and segmented sampling operations, and perform sensitivity calculation and trade-off index calculation to generate a fingerprint package;

[0057] This step is triggered by the control unit when the transition conditions of the state machine template are met. The transition conditions come from the state machine template registration processing result in the session configuration package obtained in S100 and are associated with the version identifier of the gate threshold group. The triggering time corresponds to the trial state entry, the stable observation window entry after parameter switching, or the fingerprint update window entry before archiving. Specifically, the alignment feature package generated in S200 is written into the step buffer and marked as the alignment input carrier of the current session. The control unit reads the unified time window index sequence, speed feedback sequence, load signal sequence, vibration signal summary sequence, temperature rise signal sequence, feeding position data sequence, quality proxy data sequence, and quality mark field sequence from the alignment feature package. During reading, the consistency of the session configuration package identifier and the source record field is checked, and the consistency check result is registered as the input availability mark of this step. The state feature vector is a structured description vector of the operating state of the centrifugal dehulling equipment, and the quality proxy vector is a structured description vector of the quality proxy data of the separated grains. Both are parsed from the alignment feature package by the feature construction unit in this step and aligned and organized according to the unified time window index sequence.

[0058] During the state feature vector construction process, the feature construction unit first performs field mapping back reading on the aligned feature package. Based on the target field name and channel number of the aligned feature package, it locates various signals and performs secondary cleaning processing. The secondary cleaning processing performs weight reduction or removal operations on the sampling segments marked as failed, out of bounds, maintained, or removed in the quality mark field sequence, and generates a segment gap mark field during removal. Subsequently, the feature construction unit extracts the segment boundaries of the steady-state segment, the transition segment, and the fluctuation segment of the speed feedback sequence. The segment boundaries are jointly determined by the adjacent difference of the unified time window index sequence and the segment length configuration in the state machine template. The load signal sequence extracts the load mean, load fluctuation, load mutation mode, and load hysteresis response features to the feed position data. The load mutation mode is obtained by threshold discrimination of the slope sequence of the load signal sequence and is registered as a mutation event entry. The vibration signal summary sequence extracts the frequency band energy feature and the spectral peak candidate feature. The frequency band energy feature is calculated by the window summary field in the vibration signal summary sequence, and the spectral peak candidate feature is... Features are obtained by performing peak retrieval on the frequency band energy sequence of the window summary field and matching it with the inherent rotational frequency reference table of the rotor component model; temperature rise gradient and temperature rise inflection point features are extracted from the temperature rise signal sequence, and the temperature rise inflection point features are obtained by performing second-order difference on the temperature rise signal sequence and comparing the consistency with the drift mark of the quality mark field sequence; feeding trajectory segment description and feeding position fluctuation features are extracted from the feeding position data sequence, where the feeding trajectory segment description is the discrete expression of the segmented trajectory of the feeding position over time on a unified time window index sequence, and the feeding position fluctuation features are obtained by performing joint frequency domain and time domain statistics on the residual sequence of the feeding position data sequence. Further, the feature construction unit aggregates the above features into a state feature vector according to the unified time window index sequence, and adds a segment identifier field, a segment gap mark field, and a source record reference field to the state feature vector. The segment identifier field is used to mark the running segment number in the trial sequence corresponding to the vector, and the source record reference field is used to backtrack and align the source record field of the feature package during the archiving and registration process.

[0059] During the quality proxy vector construction process, the feature construction unit reads the proportion of shelled particles and the proportion of fragmented particles from the quality proxy data sequence of the aligned feature package and performs cleaning and aggregation in conjunction with frame-level quality markers. The cleaning and aggregation process removes data points corresponding to frame removal markers and performs either preservation or interpolation on a unified time window index sequence. The selection of preservation or interpolation is based on the quality proxy channel gap policy field in the session configuration package. Specifically, if the gap policy field indicates a preservation policy, the quality proxy vector uses the most recently valid proportion of shelled particles and the proportion of fragmented particles when gapped and writes a preservation marker; if the gap policy field indicates an interpolation policy, the quality proxy vector performs linear interpolation on a unified time window index sequence based on adjacent valid points and writes an interpolation marker. Subsequently, the feature construction unit performs windowed statistics on the proportion of shelled particles and the proportion of fragmented particles to generate quality fluctuation features and quality mutation event entries. The quality mutation event entries are obtained by thresholding the slope sequence of the fragmented particle proportions and recording the unified time window index of the event occurrence. Understandably, the proportion of shelled particles and the proportion of broken particles constitute the minimum set of fields of the quality proxy vector, while the quality fluctuation characteristics, quality mutation event entries and frame-level quality markers are preferred extended fields used to support the subsequent calculation of the trade-off index and the feed fluctuation anomaly discrimination link in anomaly classification.

[0060] After completing the state feature vector and quality proxy vector, the control unit enters the trial sequence orchestration processing link. The trial sequence is a combination sequence of operating segments for parameter self-adjustment, and its orchestration is based on the candidate parameter set library and centrifugal parameter group definition from the session configuration package. Specifically, the trial sequence orchestration unit reads the current centrifugal parameter group and extracts a set of candidate centrifugal parameter groups that satisfy the disturbance distance constraint from the candidate parameter set library. The disturbance distance constraint is used as an orchestration constraint in this step. The calculation input of the orchestration constraint is the parameter distance between the candidate centrifugal parameter group and the current centrifugal parameter group. The parameter distance is obtained by measuring the difference between the speed set value, speed ramp, feed trajectory, and gate threshold group. The trial sequence orchestration unit selects parameter pairs or parameter chains for trial within the candidate centrifugal parameter group set and organizes them into a sequence structure of two or more operating segments. The minimum set form of the trial sequence of the present invention is that the feed trajectories of the two operating segments are consistent and the speed ramps are different. The multiple operating segments are a preferred extended form, used to provide finer-grained sensitivity estimation when the rotor component model is switched or the raw material batch fluctuates. Furthermore, the trial sequence arrangement unit generates a run segment description field for each run segment. The run segment description field includes the run segment number, the target centrifugation parameter group reference number, the segment length configuration, the sampling window configuration, and the interlock condition code. The run segment description field is written into the trial sequence structure, and the trial sequence structure is written into the internal intermediate data area of ​​this step for segmented run sampling calls.

[0061] In the segmented operation sampling process, the control unit sends the target centrifugal parameter set segment by segment to the frequency converter drive interface unit and the feeding actuator drive unit according to the trial sequence structure. Within each operation segment, the observation window mechanism of the data acquisition and time service unit is invoked to continuously extract the sampling segments of the corresponding segments from the alignment feature package. Specifically, the segmented operation sampling unit sets a pre-stabilization window, an effective sampling window, and a post-buffer window for each operation segment. The pre-stabilization window is used to shield the transient parameter switching, and the post-buffer window is used to cover the mechanical inertial wake. The above window configuration comes from the sampling window configuration in the operation segment description field. Within the effective sampling window, the segmented operation sampling unit reads the segment identifier field corresponding to the state feature vector and the quality proxy vector and extracts the vector segment of that operation segment to form operation segment sample pairs. The operation segment sample pairs are composed of state feature vector segments and quality proxy vector segments. During the formation, a sample consistency mark field is generated simultaneously. The sample consistency mark field is jointly determined by the segment gap mark field, the quality mark field sequence summary, and the interlock condition code. If the sample consistency flag field indicates an anomaly, the segmented sampling unit marks the sample pair of that segment as low confidence and writes the anomaly cause code to the operation log area, while retaining the reference number of the sample pair for evidence backtracking in subsequent gating decision links. Understandably, the pre-stabilization window, the effective sampling window, and the post-buffer window constitute the minimum set configuration for segmented sampling, while the sample consistency flag field and low confidence labeling are preferred extended configurations used to provide continuous periodic consistency evidence in subsequent convergence gating decisions.

[0062] In the sensitivity calculation process, the sensitivity calculation unit performs differential modeling on the sample pairs of operating segments and outputs the response slope to changes in rotational speed ramp. Specifically, the sensitivity calculation unit extracts sub-feature sets sensitive to rotational speed ramp from the state feature vector fragments corresponding to the two operating segments. These sub-feature sets include rotational speed fluctuation features, load fluctuation features, vibration frequency band energy features, feed position fluctuation features, and temperature rise gradient features. Based on the feature selection configuration fields in the session configuration package, the unit performs minimum set filtering on the above sub-feature sets. Minimum set filtering determines rotational speed fluctuation features, load fluctuation features, and vibration frequency band energy features as mandatory fields, and feed position fluctuation features and temperature rise gradient features as preferred fields. Subsequently, the sensitivity calculation unit reads the difference in rotational speed ramp between the two operating segments and performs intra-segment aggregation on each sub-feature set on a unified time window index sequence. Intra-segment aggregation generates feature statistics for each operating segment. The sensitivity calculation unit performs difference calculations on the characteristic statistics of the two operating segments and divides them by the speed ramp difference to obtain the response slope vector. This response slope vector is then registered as a sensitivity field, which serves as one of the core intermediate products of this step, characterizing the degree of response of the state feature vector to changes in the speed ramp. Further, when the trial sequence is in the form of multiple operating segments, the sensitivity calculation unit performs piecewise regression on the characteristic statistics of the multiple operating segments to obtain multi-point response slopes, and writes a summary of the regression residuals into an extended subfield of the sensitivity field as a low-confidence indicator.

[0063] In the tradeoff index calculation process, the tradeoff index calculation unit performs differential modeling on the mass surrogate vector segments in the sample pairs of operating segments and outputs the response slope of the particle size distribution to changes in rotational speed ramp. Specifically, the tradeoff index calculation unit reads the particle size distribution from the mass surrogate vector segments corresponding to the two operating segments and performs intra-segment aggregation to obtain the mass statistics for each operating segment. Simultaneously, it reads the proportion of shelled particles to generate shelled particle statistics and registers them as an auxiliary mass field. The tradeoff index calculation unit performs differential analysis on the mass statistics of the two operating segments and divides them by the difference in rotational speed ramp to obtain the mass response slope. This mass response slope is then registered as a tradeoff index field. This tradeoff index field is another core intermediate product of this step, used to provide mass constraint information corresponding to the sensitivity field in the subsequent generation of candidate centrifugation parameter groups and comprehensive scoring and ranking. Furthermore, the tradeoff index calculation unit performs consistency verification between the shelled statistics and the quality response slope. If the shelled statistics exhibit an abnormal jump between two running segments and the corresponding frame-level quality tag has an interpolation tag, the tradeoff index calculation unit marks the tradeoff index field as low confidence and writes it into the low confidence reason code. The low confidence reason code is written into the running log area and appended to the quality tag summary field of the subsequent fingerprint packet.

[0064] In the fingerprint package generation process, the fingerprint construction unit encapsulates the sensitivity field, the tradeoff index field, the run segment description field, the sample consistency marker field, and the candidate feature summary related to the anomaly into a fingerprint package. Specifically, the fingerprint package structure includes a session configuration package identifier, an alignment feature package summary reference number, a trial sequence structure reference number, a run segment description field set, a sensitivity field, a tradeoff index field, a state feature vector summary, a quality proxy vector summary, a quality marker summary field, and a source record reference field. The state feature vector summary includes a vibration spectrum peak candidate feature summary, a load mutation morphology summary, and a feed position fluctuation summary, which are pre-embedded as candidate entries for anomaly fingerprint construction in this step. The quality proxy vector summary includes a shelled particle ratio summary and a fragmented particle ratio summary, which are pre-embedded as evidence entries for quality gating judgment in this step. Further, during encapsulation, the fingerprint construction unit backfills the target centrifugation parameter group reference number in the run segment description field set, establishes an association between it and the entry index of the candidate parameter set library, and writes it into the parameter association field. This allows subsequent steps to directly reuse this association field when generating candidate centrifugation parameter groups and sorting the comprehensive score without repeatedly searching the candidate parameter set library. Understandably, the sensitivity field and the trade-off index field constitute the minimum set of fields in the fingerprint packet. The set of running segment description fields and the parameter association fields are necessary supporting fields for the closed-loop self-adjusting link of this invention. The state feature vector summary and the quality proxy vector summary are preferred extended fields used to improve the traceability of evidence in anomaly classification and archiving registration processing.

[0065] In the engineering implementation scenario, after the centrifugal dehulling equipment completes the generation of the alignment feature package, the control unit enters the trial state according to the state machine template and triggers this step. The trial sequence arrangement unit selects two candidate centrifugal parameter groups from the candidate parameter set library that satisfy the disturbance distance constraint with the current centrifugal parameter group, and generates a trial sequence structure for two running segments. The feeding trajectory of the two running segments is consistent, but the speed ramp is different. The feeding actuator drive unit performs segmented feeding position control according to the feeding trajectory, and the frequency conversion drive interface unit performs speed set value update according to the speed ramp. The segmented operation sampling unit extracts the state feature vector fragment and the quality proxy vector fragment of the corresponding segment from the alignment feature package within the effective sampling window of each running segment to form a running segment sample pair. The sensitivity calculation unit calculates the response slope vector of the two running segment sample pairs to obtain the sensitivity field, and the tradeoff index calculation unit calculates the quality response slope of the two running segment sample pairs to obtain the tradeoff index field. The fingerprint construction unit encapsulates the sensitivity field, the tradeoff index field, and the runtime segment description field set to generate a fingerprint package. It also writes references to entries related to the low-confidence reason code and sample consistency marker fields from the runtime log area into the quality marker summary field of the fingerprint package. If, in this embodiment, an abnormality occurs in the image acquisition device's exposure status field, causing a continuous increase in the quality proxy vector interpolation marker, the tradeoff index field is marked as low-confidence, and a reason code reference is retained in the fingerprint package. Subsequent steps read this reason code reference in the quality gating decision to perform weight reduction processing on the gating decision input.

[0066] At the end of this step, the fingerprint packet, as the output of this step, is explicitly recorded as an output field name in the text and written to the step cache. Simultaneously, the session configuration packet identifier and the fingerprint packet summary are registered to form a searchable running record. The next input position of the fingerprint packet corresponds to the "fingerprint packet" in S400. Subsequent steps perform security gating, quality gating, and convergence gating on the fingerprint packet, and under the constraints of the gating results, generate candidate centrifugation parameter groups, perform comprehensive scoring sorting and perturbation distance constraint selection, thereby generating updated centrifugation parameter groups and archived record packets. Understandably, this step achieves cross-step connection through the structured transformation from alignment feature packets to fingerprint packets. The alignment feature packets provide the original alignment evidence carrier, and the fingerprint packets provide a computable summary carrier for gating decisions and parameter updates. The two form a closed-loop traceable link through the session configuration packet identifier and the source record reference field.

[0067] In summary, this step organizes the multi-source aligned data in the alignment feature package into state feature vectors and quality surrogate vectors, and forms reusable sensitivity and tradeoff index fields under the constraint of trial sequences. The fingerprint package encapsulates the runtime description, parameter associations, and anomaly candidate features, while retaining the quality tag summary field and source record reference field. The fingerprint package is called by S400 for gating determination and the candidate centrifugation parameter group update link, and maintains consistent association with the entry index of the candidate parameter set library.

[0068] S400. Based on the fingerprint package, perform security gating, quality gating and convergence gating processing, and under the constraints of the gating results, perform candidate centrifugation parameter group generation, comprehensive score sorting, disturbance distance constraint selection, abnormal fingerprint construction, abnormal classification and handling action sequence generation processing, and generate updated centrifugation parameter group and archived record package.

[0069] This step is automatically triggered by the control unit upon receiving the fingerprint packet output by S300. The fingerprint packet is associated with the session configuration packet identifier and carries an alignment feature packet summary reference number, a trial sequence structure reference number, a set of running segment description fields, a sensitivity field, a tradeoff index field, a quality mark summary field, and a source record reference field. Specifically, when entering this step, the control unit first performs a structural integrity verification on the fingerprint packet. The verification includes the existence of fields, the matching relationship between field definitions and the session configuration packet identifier, and the resolvability of the reference number of the target centrifugation parameter group in the running segment description field set with the index of the candidate parameter set library entry. The verification conclusion is then registered as a fingerprint packet input availability flag. When the fingerprint packet input availability flag is unavailable, the control unit writes the fingerprint packet to the operation log area and generates a candidate entry with an anomaly classification result of feed fluctuation anomaly. At the same time, it generates a combined entry with a disposal action sequence summary of feed reduction action and reset trial action. Subsequently, it enters the archiving and registration process, thereby maintaining an auditable operating trajectory for the closed-loop link under abnormal input scenarios. Understandably, the control unit achieves online updating and version recording of centrifugation parameter groups by gating, scoring, sorting, classifying and archiving the fingerprint packets.

[0070] In the security gating decision process, the gating decision unit extracts security-related evidence fields from the fingerprint packet and generates a comparable gating input vector. The evidence fields include load fluctuation index, vibration frequency band energy index, and temperature rise gradient index. The indexes are parsed from the summary entries pre-embedded in the state feature vector summary in the S300 stage and an index relationship is established with the running segment number in the running segment description field set. Specifically, the load fluctuation index is composed of intra-segment statistics of the load signal sequence. The gating decision unit performs outlier removal processing on the load fluctuation index. The outlier removal processing performs weighted aggregation on the corresponding time window index based on the low confidence cause code in the quality mark summary field. The vibration frequency band energy index is calculated from the window summary field of the vibration signal summary sequence. The gating decision unit loads the inherent rotation frequency reference table according to the rotor assembly model, filters the target frequency band corresponding to the acceleration sensor near the bearing housing area, and generates a frequency band energy comparison value. The frequency band energy comparison value and the vibration frequency band energy index together form a threshold comparison input. The temperature rise gradient index is composed of intra-segment gradient statistics of the temperature rise signal sequence. The gating decision unit performs steady-state segment filtering on the temperature rise gradient index. The steady-state segment boundary is given by the sampling window configuration in the running segment description field set, and the filtered result is registered as the temperature rise gating input value. Subsequently, the gating determination unit reads the gating threshold group and locks the gating threshold group version identifier. It then performs threshold comparisons between the load gating input value, vibration gating input value, and temperature rise gating input value and their corresponding thresholds within the gating threshold group. The threshold comparison employs a combination of in-direction comparison and hysteresis comparison. In-direction comparison is used to determine single-window exceedances, while hysteresis comparison is used to determine continuous window exceedances and avoid gating jitter. The gating determination unit writes the comparison results into the safety gating determination result field, and records the trigger reason code, trigger time window index range, involved running segment number, and gating threshold group version identifier within this field. Understandably, the load fluctuation index, vibration frequency band energy index, and temperature rise gradient index constitute the minimum set of inputs for safety gating determination. Hysteresis comparison and the trigger reason code are preferred extensions used to support the traceable link for subsequent convergence gating determination and archiving registration.

[0071] In the quality gating decision process, the gating decision unit extracts the shelled particle ratio summary and fragmented particle ratio summary from the quality proxy vector summary of the fingerprint packet, and generates a quality gating input vector by combining them with the quality tag summary field. Specifically, both the shelled particle ratio summary and the fragmented particle ratio summary come from the statistical results obtained by target segmentation and counting processing of the separated seed image frames in the quality proxy data. The gating decision unit performs confidence correction processing on the statistical results. The confidence correction processing forms confidence weights based on the proportion of interpolated tags, the proportion of preserved tags, and the proportion of frame culling tags in the quality tag summary field, and performs weighted aggregation on the shelled particle ratio summary and the fragmented particle ratio summary to obtain the shelled gating input value and the fragmented particle gating input value. Subsequently, the gating decision unit reads quality-related thresholds within the gating threshold group and performs threshold comparisons, comparing the shelled gating input value with the shelled threshold and the fragmented gating input value with the fragmented threshold, and writes the comparison results into the quality gating decision result field. When the quality tag summary field indicates low confidence and the low confidence reason code is abnormally related to the image acquisition device exposure status field, the gating decision unit writes a weight reduction decision tag into the quality gating decision result field and registers the source reason code of the weight reduction decision tag. It then still outputs the quality gating decision result field for subsequent comprehensive scoring and ranking. Understandably, the shelled particle ratio summary and the fragmented particle ratio summary constitute the minimum set input for quality gating decision. Confidence correction processing and weight reduction decision tags are preferred extensions used to support the continuous operation of the decision chain even in scenarios with gaps in quality proxy data.

[0072] In the convergence gating decision process, the gating decision unit performs threshold comparisons on the consistency of gating results within a continuous period. This continuous period is defined by a period index sequence associated with a session configuration package identifier and recorded in the operation log area. Specifically, the gating decision unit extracts the security gating decision result field and the quality gating decision result field from the operation log area for the most recent few periods under the same session configuration package identifier, constructs a gating result sequence, and performs consistency statistics on the gating result sequence. These consistency statistics include the correspondence between the consistency interval length, the number of state transitions, the transition interval, and the running segment number. The gating decision unit reads the convergence threshold within the gating threshold group and performs threshold comparisons on the consistency statistics, writing the comparison results into the convergence gating decision result field. When the consistency statistics show that the number of transitions exceeds the convergence threshold, the gating decision unit records the running segment number appearing in the transition set in the convergence gating decision result field and writes a gating jitter flag. This gating jitter flag is used when generating subsequent processing action sequences. Understandably, the convergence gate decision result field serves as a summary constraint term for the three types of gate decisions in this step. Together with the safety gate decision result field and the quality gate decision result field, it constitutes the gate result field set. The gate result field set is a strong constraint input for the subsequent generation of candidate centrifugation parameter groups and the comprehensive scoring and ranking.

[0073] In the process of generating candidate centrifugal parameter groups under gating result constraints, the parameter generation unit takes the set of gating result fields and the parameter association fields in the fingerprint package as input to generate a set of candidate centrifugal parameter groups and performs legality filtering on the entries in the set. Specifically, the parameter generation unit parses the parameter association fields of the fingerprint package to obtain the initial selection set of candidate centrifugal parameter groups associated with the entry index of the candidate parameter set library. The candidate centrifugal parameter groups include four types of parameter fields: speed setpoint, speed ramp, feeding trajectory, and gating threshold group. Among them, the speed setpoint and speed ramp are consistent with the reachable range of the frequency converter drive parameters, the feeding trajectory is the segmented trajectory of the feeding position over time and is consistent with the travel boundary of the feeding actuator parameters, and the gating threshold group is consistent with the version identifier of the gating threshold group locked in this session. Subsequently, the parameter generation unit performs gating filtering on the initial selection set. Gating filtering removes entries with a "reject" result from the initial selection set, while entries with a "reject" result from the quality gating result are marked as low priority and retained for comparison during anomaly classification. When the convergence gating result is "reject," the entire initial selection set is switched to conservative generation mode. Conservative generation mode means that the parameter generation unit only retains entries that are close to the current centrifugal parameter group parameters, narrowing the range of candidate entries. Simultaneously, within the candidate entries, the feeding trajectory remains unchanged, and only the rotation speed ramp is allowed to adjust within its neighborhood. Understandably, the rotation speed setpoint, rotation speed ramp, feeding trajectory, and gating threshold group constitute the minimum set of fields for the candidate centrifugal parameter group. The conservative generation mode is a preferred extension, used to maintain a controllable disturbance boundary in the parameter update chain even in scenarios where gating does not converge.

[0074] In the comprehensive scoring and ranking process, the scoring and ranking unit calculates a comprehensive score for each candidate centrifugation parameter group and outputs the ranking result. The comprehensive score includes quality deviation penalties, risk penalties, and switching cost penalties. Specifically, the quality deviation penalty is calculated by combining the tradeoff index field associated with the candidate centrifugation parameter group with the quality gating judgment result field. The scoring and ranking unit reads the tradeoff index field from the fingerprint packet and maps it to a fragment response slope summary value. The deviation of the fragment response slope summary value from the fragment threshold within the gating threshold group is converted into a penalty value and registered as a quality deviation penalty. The risk penalty is calculated by combining the sensitivity field associated with the candidate centrifugation parameter group with the safety gating judgment result field. The scoring and ranking unit reads the sensitivity field from the fingerprint packet and maps it to a state response. The slope summary value is converted into a penalty value by combining the state response slope summary value with the margin of the load threshold, vibration threshold, and temperature rise threshold within the gate threshold group, and registered as a risk penalty item. The switching cost penalty item is calculated from the parameter distance between the candidate centrifugal parameter group and the current centrifugal parameter group. The scoring and sorting unit reads the current centrifugal parameter group from the current centrifugal parameter group register associated with the session configuration package identifier, calculates the distance between the candidate centrifugal parameter group and the current centrifugal parameter group in four fields: speed setpoint, speed ramp, feeding trajectory, and gate threshold group, and maps the distance to a cost penalty value, registering it as a switching cost penalty item. Subsequently, the scoring and sorting unit performs weighted synthesis on the three types of penalties to obtain a comprehensive score and outputs the comprehensive score sorting result field. At the same time, the comprehensive score sorting result field records the sub-item value, candidate item index, gate threshold group version identifier, and calculation time window index range of the three types of penalties. Understandably, the three types of penalties constitute the minimum set components of the comprehensive score. The sub-item value record and calculation time window index range are preferred extension content, used to support the evidence backtracking of subsequent archiving registration and the updating of the parameter applicability tags of the candidate parameter set library in the partition statistics.

[0075] In the perturbation distance constraint selection process, the constraint selection unit performs perturbation distance constraint selection under the constraint of the comprehensive score ranking result field. The perturbation distance constraint is that the parameter distance does not exceed a preset threshold. Specifically, the constraint selection unit reads the comprehensive score ranking result field and traverses the candidate entries according to the ranking. It calculates the parameter distance between each candidate centrifugation parameter group and the current centrifugation parameter group and compares it with the preset threshold. The first candidate entry that satisfies the condition that the parameter distance does not exceed the preset threshold is selected as the target candidate centrifugation parameter group and recorded in the candidate centrifugation parameter group selection result field. When no candidate entry satisfies the condition that the parameter distance does not exceed the preset threshold after the traversal, the constraint selection unit registers the candidate centrifugation parameter group selection result field as having no valid entries and triggers conservative backoff processing. The conservative backoff processing refers to keeping the current centrifugation parameter group unchanged and generating a disposal action sequence summary as a reset trial action. Further, the constraint selection unit writes a selection reason code into the candidate centrifugation parameter group selection result field. The selection reason code includes three categories: reasons for passing the comprehensive score ranking, reasons for passing the perturbation distance constraint, and reasons for triggering conservative backoff. The code is written to the operation log area for subsequent convergence gating judgment. Understandably, the perturbation distance constraint selection combines the comprehensive score ranking with the parameter distance threshold to form a strong constraint screening step before the generation of updated centrifugal parameter groups. The selection reason code belongs to the preferred extension content and is used to support the auditable link of gating consistency statistics and archived record packages.

[0076] In the anomaly fingerprint construction and anomaly classification processing, the anomaly processing unit constructs anomaly fingerprints and outputs anomaly classification results under the constraints of the gating result field set and aligned evidence summary. Specifically, the anomaly fingerprint consists of vibration spectrum peak features, load mutation morphology features, and feed position fluctuation features. The anomaly processing unit parses the vibration spectrum peak candidate feature summary from the state feature vector summary of the fingerprint package to form vibration spectrum peak features; it parses the load mutation morphology features from the load mutation morphology summary entries and binds them to the occurrence time window index range; and it parses the feed position fluctuation features from the feed position fluctuation summary entries and establishes a mapping with the running segment number. Subsequently, the anomaly processing unit constructs the anomaly fingerprint structure and writes it into the anomaly fingerprint field. The anomaly fingerprint field includes three types of feature values, the time window index range, the running segment number, the gating threshold group version identifier, and the source record reference field. In this step, the anomaly classification rules are loaded by the classification rule engine. The classification rule engine is associated with the session configuration package identifier and has a version number registration mechanism. The version number registration mechanism writes the classification rule version number into the running log area and associates it with the archived record package. Specifically, a blockage anomaly is characterized by a sudden increase in load accompanied by an increase in vibration frequency band energy. The anomaly processing unit reads the sudden increase discrimination flag in the load change morphology characteristics and combines it with the increase discrimination flag of the vibration frequency band energy index for joint discrimination. If the joint discrimination is true, the anomaly classification result is output as a blockage anomaly. A slippage anomaly is characterized by a sudden decrease in load accompanied by an increase in speed fluctuation. The anomaly processing unit reads the sudden decrease discrimination flag in the load change morphology characteristics and maps it from the sensitivity field to obtain the speed fluctuation increase discrimination flag. If the joint discrimination is true, the anomaly classification result is output as a slippage anomaly. An imbalance anomaly is characterized by a continuous increase in the fixed frequency band vibration spectrum peak. The anomaly processing unit determines the anomaly classification based on the speed fluctuation peak. The sub-component model loads an inherent frequency reference table and filters fixed frequency bands within the vibration spectrum peak characteristics. It calculates a discrimination flag indicating the continuous rise of the fixed frequency band's spectral peak with the time window index. If the discrimination flag is true, the output anomaly classification result is an imbalance anomaly. Feed fluctuation anomalies satisfy the condition that the correlation coefficient between feed position fluctuation and quality proxy vector fluctuation exceeds a threshold. The anomaly processing unit constructs a feed fluctuation sequence summary from the feed position fluctuation characteristics and a quality fluctuation sequence summary from the quality proxy vector summary. It calculates the correlation coefficient between the two on a unified time window index and compares it with the correlation coefficient threshold within the gating threshold group. If the comparison exceeds the limit, the output anomaly classification result is a feed fluctuation anomaly. Understandably, all four anomaly classifications are calculated from traceable summary fields within the fingerprint package. The version number registration mechanism of the classification rule engine is a preferred extension, used to provide auditable evidence of rule evolution in subsequent archiving registration and review process responses.

[0077] In the action sequence generation process, the action generation unit generates an action sequence based on the anomaly classification result, the gating result field set, the candidate centrifugal parameter group selection result field, and the gating jitter flag, and writes it into the action sequence summary field. Specifically, under the blockage anomaly classification, the action generation unit generates an action sequence that includes three actions: material reduction, speed reduction, and reset probe. The material reduction action is executed by the feeding actuator drive unit and switches the feeding trajectory to a load reduction segmented trajectory. The speed reduction action is executed by the frequency converter drive interface unit and switches the speed ramp to a conservative ramp. The reset probe action is executed by the control unit and triggers a new round of probe sequence arrangement. The action generation unit sets a reset probe failure count for the reset probe action and associates it with the failure reason code in the operation log area. When the reset probe failure count reaches the threshold, the action generation unit generates a shutdown prompt action and writes it into the action sequence summary field. The shutdown prompt action is a prompt instruction item output by the human-machine interaction unit and a prompt time window index is registered in the operation log area. Under the unbalanced anomaly classification, the action generation unit generates a sequence of actions including a speed reduction and locking action and a maintenance prompt action. The speed reduction and locking action refers to locking the speed setpoint to a preset safety level and freezing the candidate centrifugal parameter group selection result field. The maintenance prompt action is a maintenance instruction entry output by the human-machine interaction unit and a maintenance prompt reason code registered in the operation log area. For slippage anomalies and feed fluctuation anomalies, the action generation unit introduces a gating jitter flag when generating the action sequence. When the gating jitter flag exists, the action generation unit prioritizes generating a reset trial action and reduces the parameter switching frequency. When it does not exist, the action generation unit generates a parameter update action according to the candidate centrifugal parameter group selection result field and switches the feeding trajectory in tandem with the speed ramp. Understandably, the action sequence summary field is referenced by the archived record package as an output evidence field in this step. At the same time, the action generation unit, the feeding actuator drive unit, the frequency converter drive interface unit, and the human-machine interaction unit form a closed-loop execution link of "generation-issuance-recording".

[0078] In the process of generating and updating centrifugal parameter groups, the parameter update unit generates an updated centrifugal parameter group when the gating result field set allows it and the candidate centrifugal parameter group selection result field contains a valid entry. The updated centrifugal parameter group is then written to the centrifugal parameter group register and written back to the operation log area. Specifically, the parameter update unit reads the target candidate centrifugal parameter group corresponding to the candidate centrifugal parameter group selection result field, encapsulates its parameter fields (speed setpoint, speed ramp, feed trajectory, and gating threshold group) into updated centrifugal parameter group fields, and writes an updated centrifugal parameter group version identifier. This updated centrifugal parameter group version identifier is composed of the session configuration package identifier, the gating threshold group version identifier, the classification rule version number, and the time window index range. Subsequently, the parameter update unit sends the updated centrifugal parameter set to the frequency converter drive interface unit and the feeding actuator drive unit and records the sending receipt. The sending receipt includes the sending time window index, the execution unit identifier, the parameter validity flag, and the interlock condition code. When the interlock condition code indicates that the current handling action sequence includes a speed reduction locking action, the parameter update unit freezes the speed setting value of the updated centrifugal parameter set and only allows the feeding trajectory to enter the de-load segment trajectory, thus maintaining consistency with the handling action sequence. Understandably, the updated centrifugal parameter set field is one of the output products of this step and belongs to the source of the centrifugal parameter set referenced by the session configuration package mentioned in S200 in subsequent closed-loop iterations, thereby achieving cross-step cyclical connection.

[0079] In the archiving registration process, the archiving unit generates an archive record package and performs partitioned statistical processing to update the parameter application tags and gating threshold group version identifiers in the candidate parameter set library. Specifically, the archiving unit encapsulates the session configuration package identifier, alignment feature package summary, fingerprint package summary, gating judgment result, updated centrifugation parameter group, anomaly classification result, and disposal action sequence summary into an archive record package, and writes the source record reference field, selection reason code, classification rule version number, gating threshold group version identifier, and issuance receipt summary into the archive record package. After the archived record package is written to the archived storage area, the partition statistics unit performs partition aggregation according to rotor component model, centrifugal dehulling equipment identifier, raw material batch mark, and anomaly classification result, generates a partition statistics summary, and writes the partition statistics summary back to the parameter applicable tag field corresponding to the candidate parameter set library entry index. The parameter applicable tag field includes applicable rotor component model tag, applicable feeding trajectory tag, risk level tag, and gate convergence tag. At the same time, the partition statistics unit writes the version identifier of the locked gate threshold group into the gate threshold group version identifier field of the candidate parameter set library, so that the candidate parameter set library has versioned information that can be directly reused when the candidate parameter set library construction process is triggered in S100. Further, the partition statistics unit performs deduplication registration on the archived record package. The deduplication registration generates a deduplication key based on the session configuration package identifier and the time window index range to avoid statistical deviation caused by repeated writing within the same time window index range, and writes the deduplication registration result to the operation log area. Understandably, the archived record package and the update of the candidate parameter set library form a closed-loop link of "running evidence - statistical aggregation - in-library tag write-back". The archived record package is the second output product of this step, which is used as input for the next round of candidate parameter set library construction processing in S100, thereby realizing the continuous evolution and audit trail of in-library entries.

[0080] In the engineering implementation scenario, after the centrifuge dehulling device completes the trial sequence operation and generates a fingerprint packet, it enters this step. The control unit reads the fingerprint packet and first completes three types of gating judgments. If the safety gating judgment result field shows that the vibration frequency band energy index and the load fluctuation index exceed the limit synchronously within the adjacent time window index and the temperature rise gradient index is in the rising range, then the convergence gating judgment result field is marked as non-convergent. The parameter generation unit switches to conservative generation mode and extracts candidate centrifuge parameter groups with small parameter distances from the candidate parameter set library. When calculating the comprehensive score, the scoring and sorting unit assigns a high weight to the risk penalty item and uses the switching cost penalty item in conjunction with the disturbance distance constraint. The constraint selection unit selects the items that satisfy the parameter distance not exceeding the preset threshold as the candidate centrifuge parameter group selection result field. The anomaly handling unit outputs an anomaly classification result as a blockage anomaly based on the blockage anomaly rules. The action generation unit generates a sequence of actions including material reduction, speed reduction, and reset attempt, and generates a shutdown prompt action when the reset attempt failure count reaches a threshold. The parameter update unit writes the updated centrifugal parameter group into the centrifugal parameter group register and sends it to the execution unit. The archiving unit generates an archive record package and completes partitioned statistical processing to update the parameter applicability tags and gate threshold group version identifiers in the candidate parameter set library. In another embodiment, if the anomaly handling unit outputs an anomaly classification result as an imbalance anomaly, the action generation unit generates a speed reduction and locking action and a maintenance prompt action. The parameter update unit freezes the speed setting value and keeps the feeding trajectory entering the load reduction segment trajectory. The archive record package records the maintenance prompt reason code and increases the risk level tag in the partitioned statistical summary. In another embodiment, if the quality gating judgment result field shows that the particle gating input value exceeds the limit while the safety gating judgment result field is allowed, the scoring and sorting unit assigns higher weight to the quality deviation penalty item and prioritizes the candidate item with the lower trade-off index field. The constraint selection unit is still restricted by the disturbance distance constraint. The parameter update unit writes the updated centrifugation parameter group into the register area and triggers the next round of S200 acquisition, thereby forming an automatic closed-loop operation cycle of "acquisition - fingerprint - gating - update - reacquisition".

[0081] At the end of this step, the control unit writes the updated centrifuge parameter group as an output field name into the centrifuge parameter group register and registers it in association with the session configuration package identifier. The next input position of this updated centrifuge parameter group is set as the source of the centrifuge parameter group referenced by the "session configuration package" in the next round S200. At the same time, the archived record package is written as an output field name into the archived storage area and marked as the input source for the "candidate parameter set library construction" process in the next round S100, thus forming a cyclical link around the candidate parameter set library and the gate threshold group version identifier at the cross-main step level. The control unit synchronously writes the set of gated judgment results, the abnormal classification results, and the summary of the handling action sequence into the operation log area. The operation log area is consistent with the archived record package at the field caliber level and shares the session configuration package identifier, the gate threshold group version identifier, and the classification rule version number, so that this step maintains a searchable, replayable, and auditable process record during automatic operation.

[0082] This step's technical effects can be summarized as follows: Based on the fingerprint packet input, this step completes gating determination, candidate centrifugation parameter group screening, scoring sorting, and perturbation distance constraint selection, and outputs updated centrifugation parameter groups along with the archived record package. Abnormal fingerprint construction, abnormal typing, and the generation of handling action sequences are coupled with the gating result field set within the same operational chain, and the process is recorded by the operational log area and the archived record package. The archived record package, after partitioned statistical processing, writes back the parameter application tags and gating threshold group version identifiers from the candidate parameter set library, enabling subsequent sessions to reuse entries and version information within the library under the same terminology.

Claims

1. A method for self-adjusting centrifugation parameters of a centrifuge for dehulling oats, characterized in that, include: S100: Obtain the centrifugal dehulling equipment identifier, rotor assembly model, and frequency converter drive parameters; perform centrifugal parameter group definition, candidate parameter set library construction, and state machine template registration processing to obtain the session configuration package. S200. Based on the session configuration package, synchronously collect and process rotation speed feedback, feeding position data and quality proxy data to generate an alignment feature package; S300. Based on the alignment feature package, perform trial sequence arrangement and segmented sampling processing, and perform sensitivity calculation and trade-off index calculation processing to generate a fingerprint package; S400. Based on the fingerprint package, perform security gating, quality gating, and convergence gating processing. Under the constraints of the gating results, generate candidate centrifuge parameter groups, sort comprehensive scores, select disturbance distance constraints, construct abnormal fingerprints, classify abnormalities, and generate action sequences for handling actions. Generate updated centrifuge parameter groups and archived record packages. The updated centrifuge parameter groups include speed setting values, speed ramps, feeding trajectories, and gating threshold groups encapsulated from the target candidate centrifuge parameter groups corresponding to the fields selected from the candidate centrifuge parameter groups. Add an updated centrifuge parameter group version identifier to generate archived record packages. The archived record packages include session configuration package identifiers, alignment feature package summaries, fingerprint package summaries, gating result field sets, updated centrifuge parameter groups, abnormal classification results, and action sequence summaries. Perform partitioned statistical processing to update the parameter application tags and gating threshold group version identifiers in the candidate parameter set library.

2. The method according to claim 1, characterized in that, The process of defining centrifugation parameter sets includes: The centrifugal parameter group definition process includes generating a speed reachable domain constraint field based on the available ramp gears in the frequency converter drive parameters and the rated speed level in the rotor assembly model; generating a feeding reachable domain constraint field based on the reachable domain description field in the feeding actuator parameters; performing a consistency comparison process between the three subsets of the gated threshold group and the above-mentioned reachable domain constraint fields to exclude cases where the action range required by the threshold group exceeds the reachable domain; and writing an exception reason code in the session registration area and triggering the confirmation process of the human-machine interaction unit when the consistency comparison fails.

3. The method according to claim 1, characterized in that, The process of constructing the candidate parameter set library and registering the state machine template includes: The candidate parameter set construction process includes generating a basic candidate set based on rotor assembly model, frequency converter drive parameters, and feed actuator parameters. This set consists of conservative candidate items, conventional candidate items, impact-enhanced candidate items, and crushing-suppressed candidate items. For each candidate centrifugal parameter group item, a parameter distance benchmark field is calculated for subsequent comprehensive scoring and ranking, used to calculate the switching cost penalty. Simultaneously, an item index table is constructed and indexed according to applicable condition tags, including raw grain fluctuation level tags, equipment wear level tags, vibration sensitivity level tags, and feed response level tags. A state machine template registration process is then performed, which includes reading three subsets of the gating threshold group. It generates a gating event mapping table to map load fluctuation index threshold out-of-bounds, vibration frequency band energy index threshold out-of-bounds, and temperature rise gradient index threshold out-of-bounds as gating out-of-bounds events; to map shelled particle ratio threshold out-of-bounds and fragmented particle ratio threshold out-of-bounds as subclass reason codes of gating out-of-bounds events; to map continuous period consistency threshold satisfaction as convergence achievement event; and to map continuous period consistency threshold violation accompanied by abnormal fingerprint triggering condition as convergence violation event. It also writes the library version identifier of the candidate parameter set library into the context reference field of the state machine template and writes the health status field of the sensor list into the precondition field of the state machine template to constrain the entry conditions of the trial state and the stable state, and generates a session configuration package.

4. The method according to claim 1, characterized in that, The process of synchronously collecting and processing rotation speed feedback, feed position data, and quality agent data includes: Speed ​​feedback is obtained by the variable frequency drive interface unit from the variable frequency drive's internal speed estimate or encoder feedback value. Load signal is obtained by the variable frequency drive interface unit from the variable frequency drive's output current, output power, or torque estimate. Vibration signal is obtained by the accelerometer located near the bearing housing of the centrifugal dehulling equipment, which collects a triaxial or single-axis acceleration sequence. Temperature rise signal is obtained by the bearing temperature sensor or housing temperature sensor. Feeding position data is obtained by the feeding actuator drive unit from the current position, target position, following error, and interlock status. Quality proxy data includes the separated grain image frames acquired by the image acquisition device, which are then processed by the edge computing unit to perform target segmentation and counting, outputting the proportion of shelled grains and the proportion of broken grains.

5. The method according to claim 1, characterized in that, The process of exploratory sequence arrangement and segmented sampling includes: The trial sequence arrangement process includes extracting a set of candidate centrifugal parameter groups that satisfy the perturbation distance constraint from the candidate parameter set library in the session configuration package. The perturbation distance constraint is calculated by the parameter distance between the candidate centrifugal parameter group and the current centrifugal parameter group. The parameter distance is based on the difference measure of the speed set value, speed ramp, feeding trajectory and gate threshold group, and is organized into a sequence structure of two or more running segments. For each running segment, a running segment description field is generated, including the running segment number, target centrifugal parameter group reference number, segment length configuration, sampling window configuration and interlock condition code. Segmented running sampling processing is performed. The segmented running sampling processing includes sending the target centrifugal parameter group to the frequency conversion drive interface unit and the feeding execution mechanism drive unit segment by segment according to the trial sequence structure. In each running segment, a pre-stabilization window, an effective sampling window and a post-buffer window are set to extract vector fragments from the aligned feature package to form running segment sample pairs, and a sample consistency mark field is generated.

6. The method according to claim 1, characterized in that, The process of performing sensitivity calculation and trade-off index calculation includes: The sensitivity calculation process includes extracting a set of sub-features sensitive to the speed ramp from the state feature vector fragments corresponding to the two operating segments. These sub-features include speed fluctuation features, load fluctuation features, and vibration frequency band energy features. The response slope vector is calculated based on the speed ramp difference and used as the sensitivity field. The trade-off index calculation process is then performed. This process includes reading the fragmentation ratio from the quality proxy vector fragments corresponding to the two operating segments and calculating the quality response slope as the trade-off index field. Simultaneously, consistency checks are performed, and low confidence is marked when the shelled statistics show abnormal jumps. A fingerprint packet is generated, which includes a session configuration packet identifier, an alignment feature packet digest reference number, a trial sequence structure reference number, a set of operating segment description fields, a sensitivity field, a trade-off index field, a state feature vector digest, a quality proxy vector digest, a quality tag digest field, and a parameter association field.

7. The method according to claim 1, characterized in that, The processes for safety gating, quality gating, and convergence gating include: The security gating decision process includes extracting load fluctuation indicators, vibration frequency band energy indicators, and temperature rise gradient indicators from the state feature vector summary of the fingerprint packet, and performing outlier removal processing based on the quality tag summary field. Subsequently, it reads the security gating threshold group from the gating threshold group and performs a combined strategy of same-direction comparison and hysteresis comparison for threshold comparison, generating a security gating decision result field and recording the trigger cause code and time window index range. This is followed by quality gating decision processing, which includes extracting a shelled particle ratio summary and a fragmented particle ratio summary from the quality proxy vector summary of the fingerprint packet, and performing a setting based on the quality tag summary field. The signal correction process generates shell-gated input values ​​and fragmented gated input values ​​through weighted aggregation. Then, it reads the quality gating threshold group from the gating threshold group and performs threshold comparison to generate a quality gating judgment result field. When the image acquisition is abnormal, a weight reduction judgment flag is written. Convergence gating judgment processing is then performed. The convergence gating judgment processing includes extracting the gating result sequence of the most recent period under the same session configuration package identifier from the operation log area, performing consistency statistics on the consistency interval length and the number of state flips, and comparing it with the convergence gating threshold group in the gating threshold group to generate a convergence gating judgment result field. When the number of flips exceeds the limit, a gating jitter flag is written.

8. The method according to claim 1, characterized in that, The process of generating candidate centrifugation parameter sets and ranking them by comprehensive score includes: The candidate centrifugation parameter group generation process includes parsing the initial selection set of candidate centrifugation parameter groups from the parameter association field of the fingerprint package, performing gating filtering to remove entries rejected by the safety gating judgment result field and entries rejected by the quality gating judgment result field with reduced weight, and enabling a conservative generation mode when the convergence gating judgment result field is rejected to retain only entries with small parameter distances and lock the feeding trajectory unchanged, and performing comprehensive scoring and sorting processing. The comprehensive scoring and sorting processing includes calculating a comprehensive score for each candidate centrifugation parameter group set. The comprehensive score includes a quality deviation penalty calculated by the trade-off index field and the quality gating judgment result field, a risk penalty calculated by the sensitivity field and the safety gating judgment result field, and a switching cost penalty calculated by the parameter distance between the candidate centrifugation parameter group and the current centrifugation parameter group. The three types of penalties are weighted and synthesized to output the comprehensive scoring and sorting result field.

9. The method according to claim 1, characterized in that, The process of perturbation distance constraint selection, anomaly fingerprint construction, and anomaly classification includes: The disturbance distance constraint selection process includes traversing candidate entries under the constraint of the comprehensive score ranking result field, calculating the parameter distance and comparing it with a preset threshold, selecting the first candidate entry that meets the distance constraint as the target candidate centrifugal parameter group and recording the candidate centrifugal parameter group selection result field; otherwise, triggering conservative backoff processing to maintain the current centrifugal parameter group, and performing abnormal fingerprint construction and abnormal classification processing. The abnormal fingerprint construction process includes parsing vibration spectrum peak features, load mutation morphology features and feed position fluctuation features from the state feature vector summary of the fingerprint package to construct an abnormal fingerprint field. The abnormal classification processing includes identifying blockage anomalies, slippage anomalies, imbalance anomalies and feed fluctuation anomalies based on the classification rule engine, and outputting the abnormal classification result and classification rule version number.

10. The method according to claim 1, characterized in that, The process of generating and processing the action sequence includes: The process of generating the action sequence includes generating a summary field of the action sequence based on the anomaly classification result, the set of gating result fields, and the candidate centrifuge parameter group selection result fields. This summary field includes actions such as reducing material, slowing down, resetting and probing, slowing down and locking, or maintenance prompts. The process also generates and updates the centrifuge parameter group.

Images