An Adaptive Control Method for Swing Sorting Based on DWS Data

Abstract

This invention relates to the field of logistics sorting control, and proposes an adaptive control method for pendulum sorting based on DWS data. By writing dynamic weight sequences, volume parameters, and conveying speed data using a unified clock source timestamp and associating them with event identifiers, a per-item DWS data record is generated. A per-item state vector is generated by windowing aggregation of the data record, and dynamic risk quantities are calculated. Lateral displacement prediction and prediction uncertainty parameters are obtained by inputting the state vector and pendulum control parameters into a contact dynamics surrogate model constrained by equipment boundaries. The prediction results are solved using chance constraints to obtain the optimal per-item pendulum control parameters, generating a control parameter package to drive lateral flow splitting. Residuals are generated by collecting downstream displacement data and incremental updates or frozen verification branch processing are performed, while simultaneously outputting comparative evaluation results and batch-level correction parameters.

Description

Technical Field

[0001] This invention relates to the field of logistics automation sorting control and industrial algorithm application technology, specifically a swing wheel sorting adaptive control method based on DWS data. Background Technology

[0002] In scenarios such as express delivery, warehousing, and manufacturing logistics, swing wheel sorting devices typically achieve diversion by applying lateral force to items on the conveyor line. Existing swing wheel sorting control often employs fixed parameter or tiered parameter strategies: the system performs diversion based on the conveyor line's set belt speed, fixed swing wheel angle, and action duration; or it performs discrete tiered configuration of the swing wheel's action parameters based on the item's barcode, destination slot, and a small amount of geometric information, and triggers the swing wheel action according to the tiered result.

[0003] In the existing control methods described above, the effects of factors such as dynamic weight changes, volume measurement fluctuations, transport speed perturbations, and changes in object posture on the control link are often not uniformly modeled. Because there is a spatial distance and time difference between the object passing through the DWS acquisition area and the balance wheel's action area, existing systems typically only use DWS data as a basis for billing or static tiering. It is difficult to establish a stable, piece-by-piece correlation between the piece-by-piece DWS data and the control parameters at the moment of balance wheel movement, thus making it difficult to achieve a verifiable closed loop for individual objects at the control level.

[0004] In addition, the following objective characteristics often exist in logistics sites: the shapes of items in the same batch vary greatly, including soft packages, irregularly shaped items, and items with an off-center center of gravity; the conveyor speed fluctuates in a short period, and the speed measured by the encoder deviates from the actual relative motion of the items; volume measurement is affected by obstructions, reflections, and attitude changes, resulting in inter-frame fluctuations; dynamic weighing experiences transient vibrations and sampling noise during the transfer of items. These factors can cause uncertainty in the diversion displacement after the lateral action of the balance wheel. If a control strategy with fixed parameters or empirical thresholds is still used, problems such as diversion deviation, missed diversion, and cross-contamination can easily occur under different batches, different item types, and different operating conditions.

[0005] Meanwhile, existing technologies for online updates to balance wheel control strategies typically rely on manual parameter tuning or offline calibration, lacking a component-based traceable data closed loop with a "prediction-execution-measurement-residual-update" link. Due to the lack of a quantitative expression of control decision uncertainty, existing systems struggle to provide calculable constraints to guide control quantity solutions when facing measurement fluctuations and operating condition drift. They also fail to provide clear verification criteria and freeze-update mechanisms when abnormal flow occurs, thus limiting the stability and maintainability of the control strategy in long-term operation.

[0006] Therefore, this invention proposes an adaptive control method for pendulum sorting based on DWS data. This invention forms DWS data records by associating dynamic weight, volume, and conveying speed data with event identifiers on a per-item basis. It obtains per-item state vectors and generates risk quantities through windowed aggregation. Lateral displacement prediction and uncertainty parameters are output through a contact dynamics proxy model. Under equipment boundary constraints, opportunity constraints are constructed to solve for the optimal per-item pendulum control parameters, forming a control parameter package to drive the flow splitting. Simultaneously, residuals are generated through downstream displacement acquisition, and incremental updates or frozen verification branch processing are performed. Continuous adaptation is achieved based on comparative evaluation and batch-level correction. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an adaptive control method for swing wheel sorting based on DWS data, in order to solve the problems mentioned in the background section.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] S1 acquires the dynamic weight sequence, volume parameters, and conveyor speed data of the items to be sorted through the DWS acquisition module, and generates event identifiers. Associated DWS data records;

[0010] S2, Based on the DWS data records, perform windowed aggregation to generate a per-item state vector. And generate dynamic risk quantity based on the component state vector. ;

[0011] S3, based on the component state vector and the balance wheel control parameters, a contact dynamics proxy model constrained by equipment boundary constraints is adopted. Generate predicted lateral displacement and prediction uncertainty parameters;

[0012] S4. Construct opportunity constraints based on the predicted lateral displacement and the predicted uncertainty parameters, and solve for the optimal control quantity of the component-by-component balance wheel under the condition of satisfying the opportunity constraints and the equipment boundary constraints. Generate component control parameter package ;

[0013] S5, based on the component control parameter package, control the balance wheel to perform lateral flow splitting, and collect the actual lateral displacement to generate residuals. ;

[0014] S6, perform incremental update or freeze update branching on the contact dynamics proxy model based on the residual, and output a review and handling instruction when freezing update;

[0015] S7. Construct a control strategy group and an experimental strategy group, respectively execute the fixed control parameters and the optimal balance wheel control parameters for each component, and generate a comparative evaluation report. ;

[0016] S8, based on batch-level correction. Self-constrained correction parameters are generated, and the sorting accuracy optimization factor is updated based on the risk mapping for solving the opportunity constraint.

[0017] Furthermore, the event identifier The event identifier is generated by concatenating and encoding the channel identifier, trigger time, and local incrementing sequence number, and then... Using index-based association methods, DWS data records and control parameter packages are analyzed. Actual lateral displacement Compared with the comparison and evaluation report object Perform item-by-item association.

[0018] Furthermore, in step S1 of claim 1, the dynamic weight sequence is obtained through a weight channel, the volume parameter is obtained through a volume channel, and the conveying speed data is obtained through a speed channel; wherein the timestamp deviation between the weight channel, the volume channel, and the speed channel is [missing information]. The maximum permissible deviation is set to ;

[0019] when If the verification passes, the event will be written to the item-by-item event ledger. If the verification fails, the item will be marked as an unreliable event in time. At the same time, a review and handling instruction will be output and the self-learning update qualification corresponding to the item will be frozen.

[0020] The DWS acquisition module includes a dynamic weighing module, a volume measurement module, and a velocity acquisition module; wherein the dynamic weighing module is calibrated according to a calibration cycle. Perform weight calibration; perform geometric calibration on the volume measurement module according to standard size blocks; perform pulse coefficient calibration on the speed acquisition module according to the reference rotation speed; and bind the calibration results with the equipment serial number and write them into the calibration parameter table.

[0021] Furthermore, the window length of the windowed aggregation The state vector At least include average weight Standard deviation of weight ,long ,Width ,high ,volume Average speed With velocity and acceleration ;

[0022] The dynamic risk quantity Calculate as follows:

[0023]

[0024] in For normalization function, As an index of volume stability, The error contribution was obtained by performing regression calibration on the historical residual data.

[0025] Furthermore, a risk gating system is set between step S2 and step S4. :

[0026] when When entering the chance constraint solution path, where ;

[0027] when The system outputs a conservative control parameter package to limit the balance wheel angle to a certain value. It outputs a review and handling instruction and records the gating cause vector.

[0028] Furthermore, the device boundary constraints described in step S3 include at least the balance wheel angle boundary. Duration boundary With the trigger lead boundary And write the device boundary constraints into the proxy model. The constraint layer makes the optimal control quantity It satisfies the executability constraints.

[0029] Furthermore, step S4 solves for the optimal control quantity. Previously generated candidate control sets The candidate control set is obtained by... The candidate control set is obtained by taking discrete values ​​and combining them, and the size of the candidate control set satisfies the following conditions: And eliminate candidates that violate the device boundary constraints;

[0030] The optimal control quantity Solve using the following objective function:

[0031]

[0032] This is a factor for optimizing sorting accuracy. These are the weighting coefficients. The control quantity executed for the previous item;

[0033] The opportunity constraint is expressed through the prediction uncertainty parameter. Transform into the following inequality constraints:

[0034]

[0035] in quantiles of the standard normal distribution .

[0036] Furthermore, the residual mentioned in step S6 Defined as follows:

[0037]

[0038] in, This refers to the actual lateral displacement. To predict lateral displacement; and by... Arrival time and event identifier The event log is written using a binding write method, and through the... The associated storage method is used with the component control parameter package. Establish associations by item.

[0039] The incremental update uses a sliding window sample set to adjust the window size. ;when When a sample is identified as learnable, the incremental update is performed.

[0040] when If a sample is identified as an abnormal sample, the incremental update is frozen, and a review and handling instruction is output while the abnormal cause vector is recorded.

[0041] Furthermore, the comparative evaluation report object It should at least include batch number, sample size, omission rate, cross-selling rate, diversion deviation statistics, throughput, and abnormal sample ratio, and be evaluated through the comparison and assessment report objects. The evaluation ledger was written using evidence storage methods.

[0042] Furthermore, the restricted correction described in step S8 satisfy:

[0043]

[0044] Among them, the update cycle Set as Item It is a scaling function, and by using... Boundary projection measures are used to ensure that the device boundary constraints are met.

[0045] The sorting accuracy optimization factor satisfies:

[0046]

[0047] in Based on the weights, The parameters are obtained from the regression calibration of historical residuals and risk and written into the parameter table.

[0048] This invention provides an adaptive control method for swing wheel sorting based on DWS data. It has the following beneficial effects:

[0049] 1. Consistency and Traceability by Item: By writing dynamic weight sequence, volume parameters and conveying speed data with a unified clock source timestamp and associating them with event identifiers by item, DWS data records and control parameter packages that can be indexed by item are generated, thereby realizing the consistency binding of collected data, control decisions and downstream detection results by item and process traceability.

[0050] 2. Executable control under computable constraints: By inputting the component state vector and the control parameters of the balance wheel into the contact dynamics surrogate model, lateral displacement prediction and uncertainty parameters are generated. By constructing opportunity constraints on the prediction results and superimposing equipment boundary constraints, the optimal control parameters of the component are solved, thereby quantifying the uncertainty and transforming it into executable control output under computable constraints.

[0051] 3. Closed-loop adaptive and anomaly-handling: By using differential calculation to generate residuals from the actual and predicted lateral displacements downstream, and performing incremental or frozen update branch processing based on residual constraints, the system outputs verification and handling instructions and comparison evaluation results, thereby achieving closed-loop adaptive updates of the control strategy and ensuring that there are reachable handling paths and verifiable evaluation basis for anomaly samples. Attached Figure Description

[0052] Figure 1 A flowchart of a swing wheel sorting adaptive control method based on DWS data;

[0053] Figure 2 This is a system framework diagram of the present invention. Detailed Implementation

[0054] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0055] The present invention will now be described in detail with reference to the accompanying drawings:

[0056] Example 1 provides an adaptive control method for balance wheel sorting based on DWS data, applied to balance wheel sorting in the scenario of mixed parcels in express delivery.

[0057] 1. Implementation Environment and Equipment Configuration

[0058] 1.1 The conveyor line adopts a linear belt conveyor structure with a conveying speed of 1.20 m / s. The belt speed is collected by an incremental encoder on the drive roller with a resolution of 2048 pulses / revolution. The roller circumference is registered as 0.314 m / revolution according to the calibration table.

[0059] 1.2 The DWS acquisition area is located 3.0m upstream of the pendulum wheel's action area. The dynamic weighing module and the volume measurement module are arranged sequentially in the DWS acquisition area. The volume measurement module uses a structured light 3D camera with a frame rate of 60fps; the sampling rate of the dynamic weighing module is set to 500Hz.

[0060] 1.3 The balance wheel sorting unit adopts servo drive, and the balance wheel control quantity is defined as: balance wheel angle. Duration of action Trigger advance The equipment boundary is registered as follows:

[0061] ; ; The aforementioned boundaries are written into the "Equipment Boundary Table" based on the equipment nameplate specifications and on-site commissioning records.

[0062] 2) Perform the steps of the method of this invention.

[0063] Step S1: Obtain item DWS data and generate an item-by-item event ledger

[0064] S1-1 uses the rising edge of the DWS input photoelectric trigger signal as the trigger time. The system generates event identifiers. :

[0065] .

[0066] in," "" indicates a field concatenation operation, which connects multiple fields in a pre-agreed order and encoding format to form a unique identifier. The channel number (01) represents the DWS acquisition channel number into which the item enters. In this embodiment, the channel number is represented by two digits. For example, "01" represents the first acquisition channel; This indicates the trigger time corresponding to the rising edge of the DWS input photoelectric trigger signal. In this embodiment, A unified clock source output timestamp is used, with the time unit being microseconds, to ensure that different trigger times can still be distinguished during continuous high-speed component processing; the incrementing sequence number This represents the locally incrementing count value under the same channel number. In this embodiment, it is represented by a fixed-length decimal or hexadecimal code, such as a 6-bit code, starting from "000001" and incrementing with each trigger. It is used to match the channel number with... Guarantee the same timestamp or insufficient timestamp granularity. Uniqueness. Through the above concatenation rules, the system can obtain a uniquely resolvable event identifier. and the Write the index fields of the corresponding DWS data records, component control parameter packages, and downstream lateral displacement detection records to achieve component-based consistency binding.

[0067] S1-2 dynamic weighing module outputs weight sequence Structured light camera output size parameters These represent the lengths of the items in the conveying direction. Width in the transverse direction of the conveyor belt and the height perpendicular to the conveyor belt plane ;

[0068] And calculate the volume Encoder output belt speed The system writes the above data to a unified clock source timestamp. This forms a DWS data stream D(t) bound to the event identifier:

[0069]

[0070] The S1-3 unified clock source uses an industrial Ethernet PTP clock, with the edge controller acting as the time master and the DWS acquisition board and camera trigger controller acting as slaves; the timestamp resolution is set to 1µs.

[0071] S1-4 Calibration and Verification:

[0072] The dynamic weighing module performs weight calibration every 72 hours, with weights weighing 5kg and 10kg at two points; the calibration results are written into the "Weighing Calibration Table";

[0073] The structured light camera performs calibration on the calibration board every 168 hours, outputs the reprojection error and writes it into the "visual calibration table";

[0074] The encoder performs a circumference coefficient verification every 168 hours, calculates the pulse equivalent using a 10m reference ruler push method, and writes it into the "encoder coefficient table".

[0075] S1-5 Time Alignment Verification Branch:

[0076] The system calculates the three-channel timestamp deviation. In this embodiment, the maximum permissible deviation is set to .

[0077] The determination method is as follows: The standard target is passed through the acquisition area 500 times consecutively, and the results are calculated. The 99th percentile is used as the upper limit.

[0078] like If the verification is successful, then... Write it into the item-based event log;

[0079] like If the verification fails, the item is marked as an "unreliable time event," a review and handling instruction is output to put it into the bypass review channel, and the self-learning update qualification corresponding to the item is frozen.

[0080] Step S2: Construct the component-specific state vector and dynamic risk quantity

[0081] S2-1 Establish a window for the center In this embodiment, ; Value located at Within the interval, the determination is based on the joint constraint of the balance wheel response time test record and the encoder positioning update cycle.

[0082] S2-2 aggregates within the window to obtain the item-by-item state vector. :

[0083]

[0084] in: Represents the weight sequence within the time window The mean; Represents the weight sequence within the time window Standard deviation; Indicates the conveying speed within the time window. The mean; This represents the acceleration characteristics of the transport velocity sequence within the time window.

[0085] S2.3 Volume stability index Calculation: The sequence of projected areas of the three-dimensional bounding box of the same object within the window. Calculate the coefficient of variation

[0086]

[0087] And order ,in These are the normalization coefficients; Represents a sequence of projected areas; Represents a sequence standard deviation Represents a sequence The mean.

[0088] S2.4 Dynamic Risk Quantity Calculation:

[0089]

[0090] in For the linearly truncated normalized function, make . The determination method is as follows: At least 3000 items are extracted from historical operational data, using residuals... Perform regression calibration for the target quantity and write the regression coefficients into the "Risk Mapping Parameter Table".

[0091] S2-5 Risk Gated Branch:

[0092] This embodiment uses a gate threshold. (falling on) Within the interval), the method for determining it is as follows: under the constraint that the historical omission rate target does not exceed 0.3%, the threshold of the quantile that can make the proportion of abnormal samples stable is calculated in reverse.

[0093] If the dynamic risk quantity and Located within the device's allowable width range If the gate is deemed to be in control, proceed to step S3;

[0094] If the dynamic risk quantity or If the gate is out of bounds, the gate is deemed unqualified, and a conservative control parameter packet is output: The item is then routed to the low-speed verification branch, and the gating cause vector is recorded. .

[0095] Step S3: Construct a contact dynamics surrogate model with device boundary constraints and form a candidate control set.

[0096] S3-1 Constructing the Agent Model The input is The output is :

[0097]

[0098] in, To predict lateral displacement; For predicting uncertainty parameters; The control vector of the balance wheel consists of three control parameters.

[0099] S3-2 Injects device boundary constraints at the model output: Implements a projection operator on the output control quantity to ensure that it satisfies... Furthermore, the upper limits of angular velocity and angular acceleration are cut according to the equipment test curve.

[0100] S3-3 Generate candidate control sets In this embodiment, :

[0101] ; Pick ; Pick Candidates that violate equipment boundaries or angular velocity constraints after combination are eliminated, and the remaining candidates constitute... .

[0102] Step S4: Solve for the optimal control quantity under chance constraints and generate a package of control parameters per unit.

[0103] S4-1 target diversion channel centerline set as In this embodiment, the lateral target displacement of the centerline of the diversion port is set to 250mm, and the measured distance from the geometric center of the diversion port to the center of the main line is written into the "Channel Geometry Table".

[0104] S4-2 Precision Tolerance Determination: The effective width of the diversion port is 320mm, the maximum width of items in the same batch is 280mm, and the safety margin is set at 30mm. Calculations show:

[0105]

[0106] Therefore, this embodiment takes .

[0107] S4-3 Confidence Threshold Determination: Using "missed scores" as negative examples in historical data, calculate the quantiles of the error distribution to ensure the missed score rate meets the target. In this embodiment, we take... .

[0108] S4-4 Constructs opportunity constraints and transforms them into computable inequalities:

[0109]

[0110] Where, Pr(·): calculates or evaluates the probability of the event within the parentheses; is the standard normal quantile function; Y is the lateral displacement random variable.

[0111] S4-5 in the candidate set The above solution is used to determine the optimal control quantity for the balance wheel. :

[0112]

[0113] The parameter values ​​in this embodiment are: And determine according to step S8. .

[0114] S4-6 Generate component control parameter package :

[0115]

[0116] in Indicates the obtained wheel angle Duration of action and trigger lead time The optimal control quantity.

[0117] And write it to the execution queue, where each queue item must contain at least the following fields: Execution time prediction value Verification status, gating status.

[0118] Step S5: Perform balance wheel sorting and collect actual results to form residuals.

[0119] The S5-1 execution side sends commands to the servo driver via EtherCAT. Target angle and Hold duration instructions, and according to To trigger in advance.

[0120] S5-2 has a lateral position detection camera installed 1.5m downstream of the balance wheel's action zone to measure lateral displacement. The measurement resolution is set to 2mm; the system will With arrival time and Bind and write to the event log.

[0121] S5-3 Calculate the residual :

[0122]

[0123] Step S6: Dual Path of Self-Learning Update and Abnormal Freeze

[0124] S6-1 residual constraint is set as follows .

[0125] like Determining learnable samples: Incremental updates using a sliding window. In this embodiment, the window size is taken. And trigger a parameter update every 200 items accumulated; at the same time, update the risk mapping parameter table, so that... Consistent with the actual error variance.

[0126] like For abnormal samples: freeze the update of this item, output a review and handling instruction to put the item into the review branch, and record the abnormal cause vector. Used for offline diagnostics.

[0127] Step S7: Generate Online Comparison and Evaluation Report

[0128] S7-1 constructs two sets of strategies based on shift cycles: the control strategy group uses a fixed control quantity. The experimental strategy group used this method to solve the problem. .

[0129] S7-2 Definition of a complex, irregular set of items: Satisfying Items are categorized into a complex set; this threshold is determined and registered by the upper quantile of the historical residual distribution.

[0130] S7-3 Statistically analyzes the following indicators for the two groups: omission rate, cross-selling rate, and mean diversion deviation | 95th percentile of traffic splitting deviation, throughput; generate report objects Its fields must include at least: statistical period, sample size, proportion of complex sets, values ​​of each indicator, confidence interval, and proportion of abnormal samples, and be bound to the batch number and written into the evaluation ledger.

[0131] Step S8: Batch-level correction amount And sorting accuracy optimization factor Update

[0132] S8-1 defines batch-level correction amounts. Used for Perform limited corrections:

[0133]

[0134] in For the scaling function, make the corrected Do not exceed the equipment boundaries.

[0135] S8-2 The update cycle is set to For each item, the mean drift of the residual is calculated for each period. and according to a fixed step size Make an update, after the update Write it into the "Batch Correction Table".

[0136] S8-3 Definition :

[0137]

[0138] This embodiment takes The method for determining this is as follows: on historical samples, regression calibration is performed with the omission rate and bias as the target, so that the weight of the error term increases synchronously when the risk increases.

[0139] Example 2 provides an adaptive control method for swing wheel sorting based on DWS data, applied to large, irregular packages.

[0140] The difference between this embodiment and Embodiment 1 is that: the downstream lateral measurement uses a photoelectric array, and the chance constraint parameters are configured more strictly to meet the control stability requirements of large, irregularly packaged items; the structure of the remaining steps remains the same.

[0141] 1) Differences in equipment and parameters

[0142] 1.1 The conveying speed is set to 0.90 m / s; the upstream distance of the DWS from the pendulum's action zone is 2.5 m. 1.2 Downstream lateral displacement detection uses a photoelectric array with an array spacing of 5 mm, and lateral displacement... It is calculated from the number of the outermost blocked beam. 1.3 Accuracy tolerance is taken as... The method for determining this is as follows: the effective width of the diversion port is 420mm, the maximum item width is statistically calculated to be 360mm, and the safety margin is taken as 15mm. The result is then calculated.

[0143] 1.4 Confidence threshold selection It is derived by inversely calculating the quantiles of the error distribution of historical large packages, and is used to reduce the probability of missing items.

[0144] 2) Execution process, only listing the key differences from Example 1.

[0145] S1. Time alignment check upper limit The result is determined by the 99th percentile of 500 passes of the standard target; if the verification fails, it will enter the bypass review and freeze the qualification for renewal.

[0146] S2. Risk Governance Threshold Gating does not output conservative control parameter package

[0147]

[0148] S4. Opportunity constraint transformation still follows... Execution, and the candidate set is in Dimension reduction of high-angle terms, making Discrete sets are limited to This is to reduce the risk of boundary collisions induced by the deflection of large components.

[0149] S5. Downstream Calculated and bound by photoelectric array The entry will be made according to the accounting records, and the residual will still be recorded as follows: calculate.

[0150] S6. Residual constraints remain the same. Abnormal samples are frozen, updated, and entered into the review branch.

[0151] S7. Complex sets are defined as... The fixed control value of the strategy group was adjusted to To match the basic configuration for large-scale operating conditions.

[0152] S8. Update cycle set to =150 pieces, to suppress frequent fluctuations in batch-level correction volume over a longer period.

[0153] Comparative Example 1: A swing wheel sorting control method with fixed control parameters

[0154] S1. The dynamic weight sequence, volume parameters and conveying speed data of the items to be sorted are obtained through the DWS acquisition module, and DWS data records associated with event identifiers are generated.

[0155] S2. A per-item status vector is generated by using windowed aggregation on the DWS data records, wherein the window length and aggregation field are configured as fixed parameters; and a risk marker is generated by using fixed rule mapping on the per-item status vector, wherein the risk marker is used to record statistical calibers and is not used for control quantity solving.

[0156] S3. A fixed control parameter set is generated by using fixed control parameter configuration measures for the balance wheel control parameters. The fixed control parameter set consists of at least a fixed balance wheel angle, a fixed action duration, and a fixed trigger advance. The fixed control parameter set is then configured to take effect by writing it into a configuration table.

[0157] S4. Generate a component-based control parameter package by directly issuing the fixed control parameter set. The component-based control parameter package includes the fixed balance wheel angle, the fixed action duration, and the fixed trigger advance; and the component-based control parameter package is associated with the event identifier by adopting a component-based binding measure for subsequent execution control and statistical backtracking.

[0158] S5. Lateral diversion is achieved by using a balance wheel to execute control measures on the component control parameter package, and the actual lateral displacement is collected downstream of the balance wheel's action area; the deviation is generated by using differential calculation measures between the actual lateral displacement and the target lateral displacement, and is used to statistically analyze the diversion deviation index.

[0159] S6. Branch processing is performed by using threshold determination measures for the deviation amount: when the deviation amount does not exceed the deviation threshold, the event is written into the evaluation record by using normal recording measures; when the deviation amount exceeds the deviation threshold, the event is output as a review and handling instruction and written into the exception record; wherein, comparative example 1 does not perform incremental updates on any control model parameters.

[0160] S7. Construct a control statistical group and use the event identifier as the granularity to statistically analyze the omission rate, cross-selling rate, diversion deviation statistics and throughput, generate a control evaluation report object, and bind the control evaluation report object with the batch number and write it into the evaluation ledger.

[0161] S8. By manually configuring or writing back fixed configuration parameters to batch-level statistical results, batch-level correction parameters are generated to adjust the balance wheel angle or duration of action in the fixed control parameter set of the next batch. The control strategy of Comparative 1 adopts the direct issuance and execution of fixed control parameters, without executing the prediction output and uncertainty quantification based on the contact dynamics proxy model, without executing the opportunity constraint solution for the optimal control parameters per piece, and without executing the residual-driven incremental update mechanism.

[0162] To verify the control adaptability and improvement trend of the sorting quality indicators of the method of the present invention under actual sorting conditions, this embodiment sets up a control strategy group and an experimental strategy group, and statistically compares the two groups of samples under consistent test conditions. In the test phase, the system operates on the same conveyor line, with the same swing wheel sorting device and the same batch of items, according to event identifiers... To achieve granularity, records and statistics were compiled for each item, and the samples were divided into a control strategy group and an experimental strategy group.

[0163] The control strategy group follows the method of Comparative Example 1: a fixed set of control parameters is used to control the balance wheel. The fixed control parameter set includes at least a fixed balance wheel angle, a fixed action duration, and a fixed trigger advance. The system only performs target grid mapping and fixed parameter triggering on items, does not generate optimal control parameters per item based on per-item DWS data records, does not construct opportunity constraints based on prediction uncertainty, and does not perform residual-driven model incremental updates.

[0164] Following the method in Example 1, the experimental strategy group performs windowed aggregation of DWS data records for each item to generate an item-by-item state vector and calculates the dynamic risk quantity. It outputs lateral displacement prediction and prediction uncertainty parameters through a contact dynamics proxy model, constructs opportunity constraints under equipment boundary constraints to solve for the optimal pendulum control parameters per item, generates an item-by-item control parameter package for pendulum flow splitting, and forms residuals based on the actual lateral displacement downstream to perform incremental updates or freeze-review branch processing.

[0165] Test results are as follows Figure 2 As shown: The system uses event identifiers For statistical granularity, the following calculations were performed on the two groups of samples: (1) Miss rate: the ratio of the number of events that actually entered the wrong grid or did not enter the target grid to the total number of events in the group; (2) Cross-classification rate: the ratio of the number of events that were misclassified due to cross-grid mixing or interference with adjacent diversions to the total number of events in the group; (3) Mean diversion deviation: the mean of the absolute value of the deviation between the actual lateral displacement and the target lateral displacement in the group; (4) Standard deviation of diversion deviation: the standard deviation of the absolute value of the deviation in the group; (5) Throughput: the number of events that completed diversion per unit time. The system outputs the above statistical results by group and uses them to generate Figure 2 The bar chart shown compares the sorting quality indicators of the control strategy group and the experimental strategy group.

[0166] like Figure 2 As shown, under the same conveyor line, the same wheel sorting device, and the same batch of items, compared with the control strategy group using a fixed set of control parameters, the experimental strategy group showed smaller statistical values ​​in terms of missing sorting rate and cross-selling rate, and smaller statistical values ​​in terms of mean and standard deviation of diversion deviation, while the throughput remained at a comparable level or showed an upward trend. Figure 2 The statistical results show that the present invention generates a control parameter package per piece by piece through DWS data recording and solves the control quantity by combining uncertainty constraints. This enables stable control of the splitting accuracy per piece without introducing additional manual parameter tuning and reduces the impact of batch fluctuations on the splitting quality index.

[0167] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive control method for swing wheel sorting based on DWS data, characterized in that, include: S1 acquires the dynamic weight sequence, volume parameters, and conveyor speed data of the items to be sorted through the DWS acquisition module, and generates event identifiers. Associated DWS data records; S2, Based on the DWS data records, perform windowed aggregation to generate a per-item state vector. And generate dynamic risk quantity based on the component state vector. ; S3, based on the component state vector and the balance wheel control parameters, a contact dynamics proxy model constrained by equipment boundary constraints is adopted. Generate predicted lateral displacement and prediction uncertainty parameters; S4. Construct opportunity constraints based on the predicted lateral displacement and the predicted uncertainty parameters, and solve for the optimal control quantity of the component-by-component balance wheel under the condition of satisfying the opportunity constraints and the equipment boundary constraints. Generate component control parameter package ; S5, based on the component control parameter package, control the balance wheel to perform lateral flow splitting, and collect the actual lateral displacement to generate residuals. ; S6, perform incremental update or freeze update branching on the contact dynamics proxy model based on the residual, and output a review and handling instruction when freezing update; S7. Construct a control strategy group and an experimental strategy group, respectively execute the fixed control parameters and the optimal balance wheel control parameters for each component, and generate a comparative evaluation report. ; S8, based on batch-level correction. Self-constrained correction parameters are generated, and the sorting accuracy optimization factor is updated based on the risk mapping for solving the opportunity constraint.

2. The method according to claim 1, characterized in that, The event identifier The event identifier is generated by concatenating and encoding the channel identifier, trigger time, and local incrementing sequence number, and then... Using index-based association methods, DWS data records and control parameter packages are analyzed. Actual lateral displacement Compared with the comparison and evaluation report object Perform item-by-item association.

3. The method according to claim 1, characterized in that, In step S1 of claim 1, the dynamic weight sequence is obtained through a weight channel, the volume parameter is obtained through a volume channel, and the conveying speed data is obtained through a speed channel; wherein the timestamp deviation between the weight channel, the volume channel, and the speed channel is [missing information]. The maximum permissible deviation is set to ; when If the verification passes, the event will be written to the item-by-item event ledger. If the verification fails, the item will be marked as an unreliable event in time. At the same time, a review and handling instruction will be output and the self-learning update qualification corresponding to the item will be frozen. The DWS acquisition module includes a dynamic weighing module, a volume measurement module, and a velocity acquisition module; wherein the dynamic weighing module is calibrated according to a calibration cycle. Perform weight calibration; Perform geometric calibration on the volume measurement module according to standard size blocks; The speed acquisition module is calibrated using pulse coefficients based on the reference rotational speed. The calibration results are then bound to the device serial number and written into the calibration parameter table.

4. The method according to claim 1, characterized in that, The window length of the windowed aggregation The state vector At least include average weight Standard deviation of weight ,long ,Width ,high ,volume Average speed With velocity and acceleration ; The dynamic risk quantity Calculate as follows: in For normalization function, As an index of volume stability, The error contribution was obtained by performing regression calibration on the historical residual data.

5. The method according to claim 1, characterized in that, Set a risk gating between step S2 and step S4 : when When entering the chance constraint solution path, ; when The system outputs a conservative control parameter package to limit the balance wheel angle to a certain value. It outputs a review and handling instruction and records the gating cause vector.

6. The method according to claim 1, characterized in that, The device boundary constraints described in step S3 include at least the balance wheel angle boundary. Duration boundary With the trigger lead boundary And write the device boundary constraints into the proxy model. The constraint layer makes the optimal control quantity It satisfies the executability constraints.

7. The method according to claim 1, characterized in that, Step S4 solves for the optimal control quantity. Previously generated candidate control sets The candidate control set is obtained by... The candidate control set is obtained by taking discrete values ​​and combining them, and the size of the candidate control set satisfies the following conditions: And eliminate candidates that violate the device boundary constraints; The optimal control quantity Solve using the following objective function: This is a factor for optimizing sorting accuracy. These are the weighting coefficients. The control quantity executed for the previous item; The opportunity constraint is expressed through the prediction uncertainty parameter. Transform into the following inequality constraints: in quantiles of the standard normal distribution .

8. The method according to claim 1, characterized in that, The residual mentioned in step S6 Defined as follows: in, This refers to the actual lateral displacement. To predict lateral displacement; and by... Arrival time and event identifier The event log is written using a binding write method, and through the... The associated storage method is used with the component control parameter package. Establish associations by item. The incremental update uses a sliding window sample set to adjust the window size. ;when When a sample is identified as learnable, the incremental update is performed. when If a sample is identified as an abnormal sample, the incremental update is frozen, and a review and handling instruction is output while the abnormal cause vector is recorded.

9. The method according to claim 1, characterized in that, The comparative evaluation report object It should at least include batch number, sample size, omission rate, cross-selling rate, diversion deviation statistics, throughput, and abnormal sample ratio, and be evaluated through the comparison and assessment report objects. The evaluation ledger was written using evidence storage methods.

10. The method according to claim 1, characterized in that, The restricted correction described in step S8 satisfy: Among them, the update cycle Set as Item It is a scaling function, and by using... Boundary projection measures are used to ensure that the device boundary constraints are met. The sorting accuracy optimization factor satisfies: in Based on the weights, The parameters are obtained from the regression calibration of historical residuals and risk quantities and written into the parameter table.

Images