A method and system for simulating three-dimensional ice loads on the propulsion shafting of an icebreaker
By decomposing and synchronously acquiring triaxial loads in icebreaker propulsion shafting tests, and combining closed-loop cross-response inverse mapping technology, the problem of inconsistent synchronization between the ice impact period and the shaft rotation period was solved, thereby improving the accuracy and reliability of triaxial load simulation tests.
Patent Information
- Application Number
- CN202611619397.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies cannot effectively solve the error reuse problem when the ice impact period and shaft rotation period are not fixed and synchronized in the three-dimensional ice load simulation test of the icebreaker propulsion shaft system. This leads to the deviation of the actual three-dimensional loading relative to the original target time scale, and the cross influence between loading channels and the delay of different force responses cause load reproduction deviation.
By decomposing the target ice load spectrum into a three-dimensional target load in a fixed test coordinate system, using a common time reference to determine the application time, and simultaneously acquiring triaxial forces and shaft rotation angles, historical records are established, qualified records that meet the rotation background are screened, and closed-loop cross-response inverse mapping correction drive is used to ensure that the loading commands of each channel are sent at the same originally scheduled application time.
It effectively reduces erroneous learning caused by historical record mismatch, lowers load reproduction deviation, maintains the relative phase of the components specified in the target spectrum, and improves the accuracy and reliability of the test.
Smart Images

Figure CN123111676A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electro-hydraulic servo measurement and control technology, specifically to a method and system for simulating three-dimensional ice loads on the propulsion shaft system of an icebreaker. Background Technology
[0002] Icebreaker propulsion shafting is subjected to varying loads in the axial and two radial directions when the propeller interacts with ice. Laboratory simulation tests typically mount the propulsion shafting on a test bench. Under specified rotation and support conditions, commands are sent to the loading actuators according to a preset load time history. Force sensors obtain the actual loading force, and the response of the shafting or bearings is collected. For repeated load tests, the deviation between the target and measured forces in the existing loading process can be used to adjust subsequent drives, making the applied load close to a given load spectrum. This provides experimental records for analyzing the force and dynamic response of the shafting under ice loads.
[0003] When the ice impact period and the shaft rotation period are not fixedly synchronized, the shaft rotation angle corresponding to the same moment in the same spectrum and the previous changes in rotation angle may be different. Reusing errors by using a fixed load period relies on the condition that the loading state of the preceding and following periods is approximately the same, making it difficult to distinguish between errors caused by changes in rotation and errors that can be repeatedly compensated. If each channel independently selects historical errors, the combined three-dimensional errors may lack a common loading background. This mismatch, when introduced into mutually coupled loading channels, may cause compensation to deviate from the intended working condition. Furthermore, a unified command transmission clock cannot eliminate the force response delay of different channels, resulting in amplitude or phase deviations in the actual load at the original target time. Subsequent shifting of the measurement curves may also mask the actual timing discrepancies. Therefore, a three-dimensional ice load simulation test method and system for icebreaker propulsion shafting is needed to solve the aforementioned problems of rotational background mismatch in historical error reuse and deviations in the three-dimensional actual loading relative to the original target time scale. Summary of the Invention
[0004] In view of the above-mentioned problems in the prior art, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for simulating three-dimensional ice loads on the propulsion shafting of an icebreaker, comprising: S1. In a fixed test coordinate system, the pre-given and repeatedly played target ice load spectrum is decomposed into three-dimensional target loads, and the target action time is determined by a common time reference. Furthermore, the three-dimensional target load adopts axial, vertical and lateral components, which are obtained by projecting the target ice load spectrum containing the magnitude and direction information of the three-dimensional components or the resultant force according to a given coordinate relationship. The fixed test coordinate does not rotate with the axis, and the projection retains the positive and negative directions, static offset and relative time difference between each component of the original load. The three-dimensional target load is resampled at a common sampling interval. The absolute timestamp of each sampling point is used for actual force comparison. The position of each sampling point within the repeated spectrum is used for historical record filtering. The three-dimensional target load remains unchanged during iteration.
[0005] S2. Perform triaxial loading while the propulsion shaft system is rotating, simultaneously collect the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle. Furthermore, the synchronous acquisition of triaxial force and shaft rotation angle uses the same hardware time base latch loading interface for force measurement values and encoder angles, and corrects the fixed delay of the sensor chain so that the timestamp corresponds to the actual sampling time. The force measurement values are converted to the fixed test coordinates through a fixed direction coefficient and then used to form the triaxial load error. The historical record stores the three-channel drive at the moment of action and the unfolding shaft rotation angle within the corresponding response memory time, and the force measurement value represents the actual force transmitted to the propulsion shaft system through the loading interface.
[0006] S3. Filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed. Select the triaxial load error and corresponding drive in groups. Furthermore, the filtering of the historical records includes: With the target action time to be executed as the endpoint, the corner comparison window is determined according to the response memory length of each channel. The response memory length covers the time required for the closed-loop residual response of the corresponding channel to enter the specified error band. The endpoint angle of the historical record is compared with the predicted endpoint angle of the loading to be performed by the shortest circumference distance, and the expanded angle increment relative to the endpoint is compared point by point within the angle comparison window. Only historical records with consistent rotation direction, endpoint angle difference not exceeding endpoint angle tolerance, and all angle increment differences within the angle comparison window not exceeding increment tolerance are used to select triaxial load errors and corresponding drives in groups.
[0007] Furthermore, the group selection of the triaxial load error and the corresponding drive includes: Taking the target action time to be executed as the endpoint, the comparison window is determined according to the response memory length of each of the three channels, and the candidate records that match the axis rotation angle change in each comparison window are obtained. The common qualified record is determined by the intersection of the candidate records of the three channels. For each record in the common qualified records, the maximum deviation between its axis rotation trajectory and the axis rotation trajectory to be loaded is obtained within the three comparison windows. The non-negative weight is determined according to the reciprocal of the maximum deviation. The non-negative weight is normalized. The lower limit of the maximum deviation is determined by the angle measurement resolution. By simultaneously combining the three load error components and three corresponding drive components in the common qualified record using the same set of normalized weights, a group of selected triaxial load errors and corresponding drives are obtained.
[0008] Furthermore, the change in the shaft rotation angle to be loaded is predicted in a short time. The prediction is based on the collected unfolded shaft rotation angle, current rotation speed and specified rotation speed trajectory. The prediction interval covers the maximum advance transmission time and control processing time. The measured angle and the predicted angle are connected with the target action time as the endpoint. The comparison of the shaft rotation angle change adopts the angle tolerance, which is jointly limited by the encoder uncertainty, prediction error and the upper bound of the rotation-related force change, so that the allowable angle difference and the force change corresponding to the prediction error do not exceed the load error budget allocated to the rotation background difference. The historical records were obtained through a normal loading process that maintained the original ice impact time rhythm.
[0009] S4. Based on the cross-response between channels, the corresponding drive is jointly corrected by the three-dimensional load errors selected in groups to obtain the corrected drive, and the corrected drive is sent in advance to before the target action time according to the force response delay of each channel. Furthermore, the generation of the corrected drive adopts the closed-loop cross-response inverse mapping after removing the pure channel delay under the accepted operating condition. The three load error components at the same moment of action are jointly converted into force equivalent drive increments. The force equivalent drive increments are reduced by learning coefficients and then superimposed on the corresponding drives selected in groups. The compensation frequency band is limited to the accepted frequency band where the closed loop is stable and the closed-loop cross-response inverse mapping is available. The learning coefficients are determined according to the upper bound of the mismatch of the comprehensive model after inverse compensation, so that the residual shrinkage coefficient is less than one. The residual boundary is determined by dividing the upper bound of the single-round uncertainty error by one and the difference with the residual shrinkage coefficient. This residual boundary is used to evaluate whether the allowable absolute error of each force component waveform is greater than the residual boundary.
[0010] Furthermore, when the correction drive is sent ahead of the target action time, the transmission time of each channel is determined by deducting the force response delay of each channel for each target action time. Historical record filtering and drive correction are completed before the earliest transmission time. The command to be sent is formed by using the pre-given subsequent target ice load spectrum value and short-time axis rotation angle prediction. Each channel outputs the corresponding command at its own transmission time using a common time base, and linear interpolation is performed on adjacent commands locally in the driver. The three-way joint correction is completed before the channel time mapping, and the transmission time of each channel corresponds to the same target action time.
[0011] S5. Execute the correction drive, calculate the load reproduction error using the original time-scaled triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
[0012] Furthermore, the evaluation of the load reproduction error includes comparing the measured triaxial force with the peak value, impulse and waveform of the three-dimensional target load under a common original time scale, evaluating the deviation of the measured resultant force direction based on the resultant force direction at the same moment, evaluating the relative phase of the measured components according to the relative phase of the components specified in the target ice load spectrum, and not independently shifting the three-dimensional measurement curves afterward. Once the load reproduction error meets the predetermined requirements, the convergence drive is maintained to perform the specified loading, and the strain of the propulsion shaft system, bearing displacement or vibration response is collected. The actual load and the collected shaft system response are paired according to the original target action time, and the test record is output.
[0013] Secondly, the present invention provides a three-dimensional ice load simulation test system for an icebreaker propulsion shafting system, comprising: The spectrum processing module is used to decompose the pre-given and repeatedly played target ice load spectrum into three-dimensional target loads in a fixed test coordinate system, and to determine the target action time using a common time reference. The synchronous loading acquisition module is used to perform triaxial loading while the propulsion shaft system is rotating, synchronously acquire the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle; The historical filtering module is used to filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed, and the triaxial load error and corresponding drive are selected in groups. The drive correction and scheduling module is used to jointly correct the corresponding drive based on the cross response between channels and the three-dimensional load error selected in groups, to obtain the corrected drive, and to send the corrected drive ahead of the target action time according to the force response delay of each channel. The test evaluation module is used to execute the correction drive, calculate the load reproduction error with the original time-scale triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
[0014] The technical solution provided by this invention has the following beneficial effects: 1. For repeated loading where the ice impact period and shaft rotation period are not fixed and synchronized, this method simultaneously limits the spectral segments, intraspectral moments, and shaft rotation angle changes within the response memory time of historical records. It also takes the intersection of candidate records from three channels, ensuring that the triaxial load errors and corresponding drives entering the correction phase originate from common qualified records. For multiple common qualified records, a shared weight combination or selection of one common qualified record from the entire group is used, ensuring that the errors entering the correction phase maintain a common rotational background and source relationship with the drives. This process can exclude records with similar endpoint angles but inconsistent rotation angle changes within the window, avoiding the direct use of spliced historical components from different channels for joint compensation. Under specified angle tolerance and repeatable rotation conditions, it reduces erroneous learning caused by historical record mismatches.
[0015] 2. To address the cross-effects of loading channels and the response delays of different forces, the three-dimensional drive is first jointly corrected using closed-loop cross-response inverse mapping at the same target application time. Then, the transmission time is determined separately based on the delay of each channel, ensuring that commands at different transmission times correspond to the same originally planned application time. Under the conditions of closed-loop stability, usable inverse mapping, and model mismatch meeting the set boundaries, the correction magnitude is limited by learning coefficients, and subsequent iterations are evaluated using the original time-scale measured load. This helps reduce load reproduction deviations caused by coupling and response time misalignment, while preserving the relative phase of the components specified in the target spectrum, avoiding the use of post-hoc independent shifts of the measured curve to replace the correction of the actual loading timing. Attached Figure Description
[0016] Figure 1 A schematic diagram illustrating the steps of a three-dimensional ice load simulation test method for an icebreaker propulsion shafting system provided by the present invention; Figure 2 This invention provides an overall flowchart of a three-dimensional ice load simulation test method for an icebreaker propulsion shafting system. Figure 3 This is a schematic diagram of the three-way loading measurement and control deployment architecture provided by the present invention; Figure 4 A schematic diagram of the axis rotation angle matching relationship within the response memory window provided by the present invention; Figure 5 This is a schematic diagram illustrating the relationship between the three-channel common qualified records and the shared weight combination provided by the present invention; Figure 6 A schematic diagram of a three-dimensional ice load simulation test system for an icebreaker propulsion shafting provided by the present invention; Figure 7This is the original time-scaled load error iterative evolution diagram provided by the present invention; Figure 8 The corner trajectory screening and load residual distribution diagram provided by this invention; Figure 9 A comparison chart showing the consistency of sources between grouped combinations and channel-by-channel independent combinations provided by this invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0018] This implementation method is used for repeated ice load simulation tests of icebreaker propulsion shafting under rotation and predetermined support conditions. The test input consists of a pre-defined load time history and rotational speed conditions. The three-dimensional loading represents the axial, vertical, and lateral forces at a specified interface, and is not interpreted as an independent reproduction of all bending moments, torques, or six-degree-of-freedom loads. The timing of the ice impacts remains independent and is not adjusted to be locked to the shaft rotation period for the purpose of obtaining comparable records.
[0019] The test program runs on an industrial computer, and historical records are stored in its memory. A real-time controller connects to the servo drive, the force sensor at the loading interface, and the shaft-end encoder, handling timing output and synchronous data acquisition. The rotary drive executes a specified speed trajectory, and the electro-hydraulic servo actuator applies force to the shaft system through the loading interface. During the learning period, the same support and lubrication conditions are used, and the temperature, force, stroke, speed, and oil supply capacity are maintained within the accepted range. If saturation, loss of synchronization, or exceeding operating limits occurs, the current learning update is stopped, the cause and acquired records are saved, and the existing protection circuit performs load limiting or unloading. The following values are for illustrating the processing rules and calculation process and are not considered actual measured performance of the equipment.
[0020] Example 1 Please see Figure 1 and Figure 2 This embodiment of a method for simulating three-dimensional ice loads on an icebreaker propulsion shafting system includes: S1. In a fixed test coordinate system, the pre-given and repeatedly played target ice load spectrum is decomposed into three-dimensional target loads, and the target action time is determined by a common time reference. Specifically, the spectral processing program reads the input file containing time, coordinate definitions, and load components, and establishes a spectral segment identifier and spectral version for each repeating spectral segment. After converting the load to the fixed coordinates of the test bench, a target array is established according to the common sampling interval; each position in the array corresponds to a triaxial force and a target application time. Loading, feedback comparison, and historical retrieval all read this target array; iteration only modifies the driving force, without modifying the target value or target time. Figure 2 The "decompose the three-dimensional target load to determine the target action time" in the text refers to this input processing process.
[0021] Furthermore, the three-dimensional target load adopts axial, vertical and lateral components, which are obtained by projecting the target ice load spectrum containing the magnitude and direction information of the three-dimensional components or the resultant force according to a given coordinate relationship. The fixed test coordinate does not rotate with the axis, and the projection retains the positive and negative directions, static offset and relative time difference between each component of the original load. The three-dimensional target load is resampled at a common sampling interval. The absolute timestamp of each sampling point is used for actual force comparison. The position of each sampling point within the repeated spectrum is used for historical record filtering. The three-dimensional target load remains unchanged during iteration.
[0022] Specifically, the Z-axis is fixed as the axial direction, the X-axis as the vertical direction, and the Y-axis as the lateral direction. Numerical vectors listed below are in the order X, Y, Z. If the input already uses this coordinate system, it is read directly. If the input uses another known coordinate system, a fixed direction cosine matrix between the two coordinate systems is taken, and the force vector at each moment is projected onto the fixed experimental coordinate system. When the input is the magnitude and direction of the resultant force, the magnitude of the resultant force is first multiplied by the unit vector of the direction, and then the same projection is performed. Inputs with only scalar magnitude but lacking direction information cannot be uniquely decomposed; the program marks them as incomplete target input and does not proceed to loading.
[0023] This example uses a 2 Hz repeating spectrum, with each period lasting 0.5 s and a sampling interval of 0.5 ms. 1000 sampling positions are set per period. The absolute timestamp is obtained by adding the absolute time of the start of the repeating period to the position within the spectrum. Sampling sequence numbers within the same spectrum and the same spectrum version are used to match historical records. Inputs with different sampling rates are first processed for anti-aliasing based on the accepted bandwidth, and then interpolated at common sampling points. The three components use the same time axis, and the original relative time differences are not individually zeroed out.
[0024] The target waveform used for continuity calculations employs a squared sine envelope for the first 250 ms of each cycle. This is achieved by dividing the spectral time by 250 ms, multiplying by pi, taking the sine, and then squared. The envelope is zero for the remaining time periods. The X, Y, and Z dynamic components are multiplied by this envelope at 600, -400, and 1800 kN respectively, with an additional 465 kN static bias added to the X direction. Therefore, the total target at 125 ms is 1065, -400, and 1800 kN respectively. This decomposition preserves the negative Y direction and saves both the static bias and dynamic components for subsequent peak, impulse, and phase evaluation.
[0025] S2. Perform triaxial loading while the propulsion shaft system is rotating, simultaneously collect the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle. Specifically, the initial loading utilizes the existing feedforward and force closed-loop mechanisms of the equipment. The controller schedules the drive according to the target time axis and converts the triaxial measured forces at each action moment to fixed experimental coordinates, subtracting them from the target values at the same moment. Records are identified by spectral version, spectral segment identifier, repetition period number, and intraspectral sampling number, saving the signed triaxial load error, the corresponding drive of the three channels actually executed, the action moment, and the rotation angle of the adjacent expansion axis. Only after all three-axis forces, drives, and necessary angle windows are complete and the timestamp verification is passed is the record made searchable; partially complete records continue to wait, while records that time out or are missing are marked as invalid and will not be used as the basis for the next round of learning.
[0026] like Figure 3 As shown, the industrial computer's internal tasks include target spectrum processing, historical record filtering, drive correction, and test evaluation; the real-time controller's internal tasks include timing output, synchronization latching, and timestamp management. The industrial computer transmits the three-dimensional target load, target application time, and correction drive to the real-time controller, which returns the three-axis force and shaft rotation angle with timestamps. The field loading acquisition end includes a three-dimensional electro-hydraulic servo loading channel, a three-axis force measurement channel, a shaft-end encoder, and a propulsion shaft system. Timing commands for each channel are sent from the real-time controller to the loading channel, while the three-axis force sampling signals and shaft rotation angle sampling signals enter the real-time controller along the reverse acquisition path. The three-dimensional loading arrows in the figure indicate the direction of force application, and the arrows pointing from the propulsion shaft system to the sensing channel and encoder indicate the direction of acquiring the measured information.
[0027] Figure 2 The "rotation loading and synchronous acquisition to establish historical records" function generates an initial record during the first load. Subsequent records undergo the same integrity and time verification, and data with only command values or only angle values are not treated as actual load errors that have already been formed.
[0028] Furthermore, the synchronous acquisition of triaxial force and shaft rotation angle uses the same hardware time base latch loading interface for force measurement values and encoder angles, and corrects the fixed delay of the sensor chain so that the timestamp corresponds to the actual sampling time. The force measurement values are converted to the fixed test coordinates through a fixed direction coefficient and then used to form the triaxial load error. The historical record stores the three-channel drive at the moment of action and the unfolding shaft rotation angle within the corresponding response memory time, and the force measurement value represents the actual force transmitted to the propulsion shaft system through the loading interface.
[0029] Specifically, the three force sensing channels employ a common clock-triggered sample-and-hold and analog-to-digital conversion, with encoder counting latched on the same trigger edge. The sensors are mounted on a loading interface capable of characterizing the actual force transmitted to the shaft system. Zero point and gain are first obtained under a known load, and then the electrical signal is converted into a force value. When the sensor's sensing direction does not coincide with the test direction, a fixed direction coefficient obtained from installation calibration is used for conversion. Valve commands or uncalibrated cylinder pressure do not replace the actual force at the loading interface.
[0030] The fixed delay of the sensor chain is calibrated using the same known trigger signal and a reference acquisition chain. The data timestamp, after deducting the fixed delay of each chain, points to the actual sampling time. Fixed chain delay correction is part of sampling alignment, not a post-hoc independent shift to make the waveform meet the standard. After alignment, comparisons are made point-by-point according to the common target timestamp; the time when the message arrives at the industrial computer cannot be used as the actual force application time.
[0031] The encoder continuously accumulates adjacent sampled angles using the shortest circumferential increment to obtain the unfolded shaft rotation angle. The sampling rate should ensure that the rotation angle between normally adjacent samples is less than 180°. If pulse loss or jumps cause the unfolding to be non-unique, the affected record will be marked as invalid. The record stores the three-channel drives facing the same target at the same moment of action. Even if these drives are actually sent at different times, they are grouped together according to the moment of target action. The retention range of force and angle records covers the maximum response memory window and the required filtering history, as described later.
[0032] S3. Filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed. Select the triaxial load error and corresponding drive in groups. Specifically, the historical screening process first searches for completed historical records by spectral version, spectral segment, and intraspectral sampling sequence number, then aligns the endpoint of the angle window for each record with the target action time of the loading process. It compares records occurring at the same relative time position without moving the force measurement curve. Each channel first forms its own candidate record set, then the intersection is taken based on the complete record identifier; only records that simultaneously meet the rotational background conditions of each channel are provided to the drive correction stage with three-dimensional errors and their paired drives.
[0033] Figure 2 After the "Filter historical records by spectral segment, intraspectral time, and axis rotation angle change" step is completed, the "Does a common qualified record exist?" judgment is entered. If a common qualified record exists, group selection and joint correction are performed; if no common qualified record exists, the "Maintain basic feedforward and force closed loop" step is entered. The "NO" branch indicates that the current filtering is complete and the intersection is empty, excluding waiting states where data has not been fully collected or calculations have not been completed; if the waiting exceeds the submission deadline, this correction is not submitted, and the already prepared basic loading is still executed, recording the reason why the deadline was not met.
[0034] For further details, please refer to Figure 4 The filtering of the historical records includes: With the target action time to be executed as the endpoint, the corner comparison window is determined according to the response memory length of each channel. The response memory length covers the time required for the closed-loop residual response of the corresponding channel to enter the specified error band. The endpoint angle of the historical record is compared with the predicted endpoint angle of the loading to be performed by the shortest circumference distance, and the expanded angle increment relative to the endpoint is compared point by point within the angle comparison window. Only historical records with consistent rotation direction, endpoint angle difference not exceeding endpoint angle tolerance, and all angle increment differences within the angle comparison window not exceeding increment tolerance are used to select triaxial load errors and corresponding drives in groups.
[0035] Specifically, the response memory length is taken from the closed-loop response record that has been accepted under the same operating conditions. After deducting the pure channel delay, the residual force response is observed from the end of the disturbance or the end of the command change. The length of time that allows it to enter the specified error band and remain there in the subsequent acceptance observation period is selected, while also taking into account the measurement and operating condition change margins. This length represents the rotation experience that needs to be compared and cannot be directly replaced by the pure transmission delay.
[0036] like Figure 4 As shown, the horizontal axis "relative time" is based on the moment the target takes effect. Negative values indicate the previous time point; The corner comparison windows share this endpoint. For the current point of action to be executed, the trajectory to be executed is denoted as... , No. The historical trajectory is Both adopt the expansion angle and take their respective moments of action as the relative time endpoint; This is a relative time. express One of the channels, This is the memory length for the channel response. For interval The comparison positions are on the common sampling grid within the area. Angles are uniformly expressed in degrees, and the endpoint angle difference and the maximum rotation angle increment difference are calculated as follows:
[0037]
[0038] in, The integer used for whole-circle alignment; Using the shortest distance on the circumference, Compare the "expansion angle increment relative to the endpoint". Under the premise of consistent rotation direction, respectively require... , For the endpoint angle tolerance, This represents the incremental tolerance for the corresponding window. In actual execution, measurement and prediction uncertainties are also factored in according to the rules described later; a condition is considered satisfied when it equals the corrected allowable boundary. If any condition is not met, the candidate channel is excluded.
[0039] Figure 4 In the mechanism example, the predicted endpoint angle of the loading to be executed is 36°, having rotated 12.96° in the previous 40 ms. Historical record A has an endpoint angle of 37°, having rotated 12.80° in the previous 40 ms, with a maximum angle increment difference of approximately 0.40° within its smooth trajectory window. Historical record B has an endpoint angle of 36°, having rotated 9° in the previous 40 ms, with a maximum angle increment difference of 3.96°. Under the illustrated endpoint angle tolerance of 3° and increment tolerance of 1°, A satisfies the geometric comparison of the entire window, while B, although having an endpoint angle difference of 0°, is still excluded in the 30 ms and 40 ms windows. The upper and lower dashed lines in the figure enclose the increment tolerance range relative to the trajectory to be executed; checking only the first and last points of the window cannot replace all sampling positions.
[0040] This example is used to explain Figure 4 , Figure 5 The candidate relationship window is configured as follows: 20 ms axially, 30 ms vertically, and 40 ms laterally. Therefore, candidate B can only enter the axial candidate. The window length is dynamically configured according to the accepted closed loop of each channel and is not fixed by the direction name; the numerical experiments described later use the separately listed X, Y, and Z channel configurations. Figure 4 The numerical comparison first explains the nominal trajectory relationship, and the following prediction and measurement uncertainty checks must be performed before the actual candidate is submitted.
[0041] For further details, please refer to Figure 5 The triaxial load error and the corresponding drive group selection include: Taking the target action time to be executed as the endpoint, the comparison window is determined according to the response memory length of each of the three channels, and the candidate records that match the axis rotation angle change in each comparison window are obtained. The common qualified record is determined by the intersection of the candidate records of the three channels. For each record in the common qualified records, the maximum deviation between its axis rotation trajectory and the axis rotation trajectory to be loaded is obtained within the three comparison windows. The non-negative weight is determined according to the reciprocal of the maximum deviation. The non-negative weight is normalized. The lower limit of the maximum deviation is determined by the angle measurement resolution. By simultaneously combining the three load error components and three corresponding drive components in the common qualified record using the same set of normalized weights, a group of selected triaxial load errors and corresponding drives are obtained.
[0042] Specifically, such as Figure 5 As shown, when the axial candidate record, vertical candidate record, and lateral candidate record all contain A and E, their intersection forms a common qualified record A and a common qualified record E; historical record B is only connected to the axial candidate record and cannot enter the common combination. The axial error, vertical error, and lateral error within the record boundary in the figure, as well as the axial drive, vertical drive, and lateral drive, all come from the same action time of the same historical record, and the missing components are not supplemented from other records.
[0043] In a shared time grid, with Identify the target sampling point to be executed, numbered consecutively across periods according to absolute sampling order; historical retrieval still uses the spectral band and intraspectral position of that point. Let the candidate set of the three channels be denoted as... , and The set of commonly qualified records is as follows:
[0044] The maximum deviation used for weighting Figure 4 The maximum angle increment difference is calculated separately. For the current sampling point, select the one that makes the first... The integer number of the endpoint circumferences between the historical trajectory and the trajectory to be executed that minimizes the difference. And apply this full-week correction to the first The entire comparative interval of this historical trajectory. (Note) For the first The maximum absolute angular deviation of the historical trajectory relative to the trajectory to be executed within the corresponding comparison windows of the three channels. The lower bound for the angle measurement resolution. Normalized weights are used when the common set is not empty. The calculation is as follows:
[0045]
[0046] in, Used for identification Comparison windows corresponding to the three channels , Used to identify common qualified records; the above trajectory, window, and integer number are all obtained with respect to the current target sampling point, summation index. Iterate through the records that are all valid. Preserve the endpoint offset; it cannot be used directly. Figure 4 First subtract the respective endpoints from the results. Use this as the denominator for the weights. Both the angular deviation and the lower limit of the value are in degrees; the weights are dimensionless, non-negative, and summed to a certain value. .
[0047] Figure 5 The lower limit of the value is 0.1°. The maximum deviations of A and E are approximately 1.402° and 2.103°, respectively. Substituting these values into the shared weight relationship yields 0.6 and 0.4. The weight is 1 when there is only one shared record; when the intersection is empty, division is not performed, and the process returns to the basic loading branch.
[0048] remember and For from the first The three-dimensional load error vector and its corresponding driving vector read from the common historical records are all components according to... Arrangement, all units are The drive is represented using a calibrated force equivalent. Here... This indicates the correspondence between the location within the same spectrum and the point to be executed; the historical record itself retains its original absolute timestamp. Grouped triaxial load errors. and the corresponding driver selected in groups for:
[0049]
[0050] The two vector summations use the exact same set of records and the same set of weights, with the six components maintaining the error paired with the driver that produced it. Assuming that the errors of A and E sorted by X, Y, Z at 125 ms are 40, -25, 70 kN and 25, -10, 40 kN respectively, the combined errors are 34, -19, 58 kN. Weights are not assigned to directions with smaller errors.
[0051] Furthermore, the change in the shaft rotation angle to be loaded is predicted in a short time. The prediction is based on the collected unfolded shaft rotation angle, current rotation speed and specified rotation speed trajectory. The prediction interval covers the maximum advance transmission time and control processing time. The measured angle and the predicted angle are connected with the target action time as the endpoint. The comparison of the shaft rotation angle change adopts the angle tolerance, which is jointly limited by the encoder uncertainty, prediction error and the upper bound of the rotation-related force change, so that the allowable angle difference and the force change corresponding to the prediction error do not exceed the load error budget allocated to the rotation background difference. The historical records were obtained through a normal loading process that maintained the original ice impact time rhythm.
[0052] Specifically, the controller reserves calculation and communication time before the earliest transmission time, using the most recent valid encoder sample as the prediction starting point. First, the measured angles are expanded according to a common time grid, then the specified speed trajectory is read. Within the prediction interval, the deviation of the most recently measured speed relative to the specified speed is superimposed using the integral of the specified speed over time, resulting in an angle continuous with the prediction starting point. For constant speed segments, the current speed can be directly multiplied by the prediction duration; when there are known acceleration / deceleration trajectories, the average value of adjacent speeds is accumulated at each sampling interval, without using future measured angles in the current decision.
[0053] exist Figure 4 In this process, the maximum advance transmission time is 20 ms, and the total allowance for processing and communication is 1 ms, which includes the preparation margin for adjacent nodes required for local interpolation. Therefore, the prediction starting point is located at relative time -21 ms. The earlier window portion reads the measured angle, and the predicted angle is read from -21 ms to the target's action time; the continuous trajectories in the figure connect at this position. If a sudden change in rotational speed not included in the specified trajectory occurs within the prediction interval, or if the most recent angle sample is too old, the current prediction is invalid, and no new corrections relying on this prediction are submitted.
[0054] Angle measurement resolution only determines the lower limit of the weighted denominator and is not equivalent to encoder uncertainty. This is the upper bound of the uncertainty of the endpoint angle difference formed by the combined historical and current measurements and predictions. The upper bound of the uncertainty of the corresponding window increment difference is defined, both in degrees, and determined by encoder acceptance and prediction playback error. The actual candidate determination uses:
[0055]
[0056] The incremental uncertainty must cover the error between the angle within the window and the endpoint angle; it cannot simply use the uncertainty of a single angle sample. For example, when the endpoint uncertainty margin is 0.3° and the incremental margin is 0.4°, Figure 4 Comparing A with 1.3° and approximately 0.80° to 3° and 1° respectively, it is still acceptable; B fails when compared with 3.96° plus a margin. These margins cannot be omitted just because the trajectories shown in the diagrams are similar.
[0057] The upper bound of rotation-related force variation can be established by comparing the actual force difference corresponding to different endpoint angles and window rotation trajectories in acceptance records of the same spectral band, the same drive, and the stable temperature range. A conservative envelope covering the acceptance data and measurement margin can be established, giving the maximum force variation that may be caused by a unit endpoint angle difference and a unit increment difference. The candidate tolerance, along with the uncertainty, must not exceed the budget allocated to the rotational background difference when substituted into this envelope. For example, if both upper bounds are 0.5 kN per degree, the 3° endpoint boundary and the 1° increment boundary together correspond to 2 kN, which is lower than the example budget of 3 kN. Therefore, the tolerance cannot be tightened or reuse stopped during acceptance, and this value cannot be treated as a constant common to all equipment.
[0058] Under constant speed of 54 r / min and ice impact conditions of 2 Hz, the axis rotates 162° in each impact cycle, naturally returning to the vicinity of the same rotational phase after 20 impact cycles. The program maintains the original impact rhythm, waiting for this comparable background from normal loading records. When the spectral version, support, or acceptance temperature zone changes, inapplicable old records are discontinued; if no usable records are available after reaching the predetermined learning cycle limit, the program outputs the test state that has not yet met the learning conditions and the records already obtained, without recording them as load compliance.
[0059] S4. Based on the cross-response between channels, the corresponding drive is jointly corrected by the three-dimensional load errors selected in groups to obtain the corrected drive, and the corrected drive is sent in advance to before the target action time according to the force response delay of each channel. Specifically, the drive correction program reads the grouped errors and grouped drives at the same execution time, uses a three-channel joint mapping to convert the signed error into a force equivalent drive increment, and then adds the reduced increment to the corresponding drive. The force equivalent drive is in kN, and the servo driver converts it into an actual command quantity through calibrated gain and zero point. Only after the joint correction is completed are the transmission times allocated according to the delay of each channel; the three error components always come from the same target time, and the components are not shifted to their respective peak positions before being added. Figure 2 The phrase "group selection of three-dimensional load errors and corresponding drive joint correction and advance transmission according to channel delay" indicates this continuous processing branch.
[0060] Furthermore, the generation of the corrected drive adopts the closed-loop cross-response inverse mapping after removing the pure channel delay under the accepted operating condition. The three load error components at the same moment of action are jointly converted into force equivalent drive increments. The force equivalent drive increments are reduced by learning coefficients and then superimposed on the corresponding drives selected in groups. The compensation frequency band is limited to the accepted frequency band where the closed loop is stable and the closed-loop cross-response inverse mapping is available. The learning coefficients are determined according to the upper bound of the mismatch of the comprehensive model after inverse compensation, so that the residual shrinkage coefficient is less than one. The residual boundary is determined by dividing the upper bound of the single-round uncertainty error by one and the difference with the residual shrinkage coefficient. This residual boundary is used to evaluate whether the allowable absolute error of each force component waveform is greater than the residual boundary.
[0061] Specifically, the closed-loop cross-response is taken from the independent channel excitation records in the existing equipment acceptance test. During acceptance, known small-amplitude change commands are applied to each channel under specified speed, temperature range, and support conditions, while the actual forces in the three directions are recorded. Three columns of response data are established based on the command changes of each input channel and the changes of the three output forces. The input must be able to distinguish the effects of each channel, and the complete inverse mapping cannot be inferred from a single test spectrum that is always identical in shape and phase in the three directions.
[0062] For each input channel, the pure delay is separated using known commands and response records. The remaining responses are fitted with finite impulse response coefficients according to a common sampling interval. When dynamic inverse compensation is required, a 3×3 finite impulse response inverse filter is pre-fitted in a selected frequency band: the error of the forward response deviating from the unit mapping after concatenation with the inverse filter is used as the fitting quantity. The coefficients of the real and imaginary parts are calculated using joint least squares, and the worst-case mismatch and noise amplification are checked using acceptance records not involved in the fitting. If a mapping that meets stability, mismatch, and noise constraints cannot be obtained, learning compensation for that frequency band is not enabled.
[0063] The accepted inverse mapping and the unified compensation frequency band limit are combined into a finite impulse response coefficient matrix. ,in , For the maximum number of delayed samples, each matrix is... . For the force equivalent drive increment that is not scaled down according to the learning factor, The learning coefficient, To correct the driver, the joint update is written as:
[0064]
[0065] When both error and drive use force-equivalent units, the discrete coefficient matrix is dimensionless. The filter only reads the current and previously calculated grouped error vectors; across cycles, it inherits the state according to the continuous sampling sequence number, and the initial part without history is set to zero and included in the start-up invalid region. The storage range covers the history length required by the inverse filter; when any required grouped error has not yet formed or is invalid, no learning correction is submitted for that point, the basic loading is performed and the reason is recorded, and the formed valid grouped errors can still be saved for use by subsequent points. Errors outside the compensation frequency band are not amplified, and the remaining out-of-band errors are still included in the original time-scale evaluation; errors caused by candidate switching and weight changes are also included in the uncertainty of acceptance, avoiding the claim that the entire waveform can be stably compensated based solely on the existence of the inverse matrix at a certain frequency point.
[0066] The local static equivalent cross matrix used to illustrate the joint operation is: Its row corresponds Output force, column correspondence Force equivalent drive:
[0067]
[0068] This local calculation is equivalent to taking... , .right The combined error yields an unreduced increment of approximately [missing information]. Learning coefficient is taken as At that time, the actual superimposed increment was approximately This matrix only demonstrates the operation at one point of application; the actual dynamic operating conditions use the verified complete inverse mapping.
[0069] The learning coefficient is selected based on the upper bound of the mismatch in the comprehensive model after inverse compensation. This section uses the maximum absolute error of each force component within the specified evaluation interval as the error norm. ; The dimensionless comprehensive mismatch upper bound, determined under this norm, acceptance frequency band, and local dynamic range, is obtained from the forward and inverse mapping cascade error and independent acceptance records. This represents the upper bound of the single-round uncertainty error, in kN. This is the combined error sequence used for joint correction. This refers to the load error sequence formed in subsequent similar loading processes after the aforementioned corrective drive is executed. And require:
[0070]
[0071]
[0072]
[0073] in, This is the conservative residual shrinkage factor from the current combination error to the corrected load error. For the corresponding residual budget boundary, The allowable absolute waveform error. Upper bound. The residual bound should cover the effects of measurement, rotating background approximation, prediction, residual time difference, out-of-band error, operating condition changes, and candidate switching and weight changes under dynamic response. The allowable tolerance should only be determined when comparable records are continuously obtained, effective updates are made, and the above budget holds; it does not imply strict contraction for each impact cycle adjacent in natural time.
[0074] For example , At that time, ; Single wheel upper limit taken At that time, the residual boundary is It can be set to be greater than this limit. The wavelet is characterized by a decrease in efficiency. If any error term cannot be covered by the budget, the operating conditions should be tightened or the scheme adjusted, and the original convergence conclusion should not be used. This limit only applies to the locally repeatable response that has been accepted and cannot prove that all operating conditions decrease by the same proportion.
[0075] Furthermore, when the correction drive is sent ahead of the target action time, the transmission time of each channel is determined by deducting the force response delay of each channel for each target action time. Historical record filtering and drive correction are completed before the earliest transmission time. The command to be sent is formed by using the pre-given subsequent target ice load spectrum value and short-time axis rotation angle prediction. Each channel outputs the corresponding command at its own transmission time using a common time base, and linear interpolation is performed on adjacent commands locally in the driver. The three-way joint correction is completed before the channel time mapping, and the transmission time of each channel corresponds to the same target action time.
[0076] Specifically, for each future goal, at the moment of action The scheduler adjusts the response delay according to each channel. Arrange for delivery separately. (Note:) For the interval between nodes with a common goal, The total reserved time is for filtering, correction, communication, and advance preparation of adjacent interpolation nodes; this time is verified with worst-case latency and a margin is allowed. This is the start time of this calculation. This refers to the moment when the entire set of instructions is ready and visible to the real-time controller. When using adjacent node linear interpolation, the current node value should be ready before the previous node begins output; therefore:
[0077]
[0078]
[0079] The time in the formula uses a common reference and consistent units. Must be greater than The difference covers the worst-case time consumption and necessary margin for screening, correction, and communication. The actual sampling time of the nearest valid angle sample used for prediction must not be later than the calculation start time; if the sample is earlier, the prediction interval should be extended accordingly. In this example... Delay is For a certain period The points of application and the sending times are respectively The interval between common nodes is... Total reservation is At that time, it is possible Start the calculation, and The entire set was released previously, making Node values can participate in local interpolation of the previous half-node interval; Figure 4 middle The prediction start point corresponds to this. If the computation and communication time cannot meet this condition, the prediction will be started earlier and extended. The transmission time cannot be temporarily postponed, nor will the phase relationship in the target array be changed.
[0080] The three-channel correction values, three transmission times, spectrum version, and point-of-action identifier for the same point of action are first written into the preparation area. After verifying the completeness and ensuring that each transmission time has not expired, the entire group is published to the real-time controller. If a complete submission is not possible, some new values are not allowed to replace the old group. The basic instructions prepared for that point are executed, and the reason for not using the correction is recorded. If a fault occurs after the earliest instruction has been executed, the actual execution information is retained, the loading segment is marked as invalid, and the process ends according to the protection conditions. The remaining channels are not temporarily shifted and retransmitted, and the synchronization is not considered complete.
[0081] The driver performs linear interpolation between scheduled adjacent instructions in a time-proportional manner, with an example local update rate of no less than 20 kHz and a common target node of 2 kHz. The next node and the first node across cycles are pre-queued before interpolation is required, their target spectral values already given, and corrections are based on previously completed historical records. Estimated at a maximum target slope of 23 kN per millisecond, a 50 μs synthesis time difference corresponds to approximately 1.15 kN force error; therefore, the synthesis time difference of the acquisition, communication, and execution chains must be simultaneously verified, and the bus clock accuracy cannot replace the overall system loading timing accuracy.
[0082] S5. Execute the correction drive, calculate the load reproduction error using the original time-scaled triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
[0083] Specifically, the test evaluation procedure first confirms that the current round of instructions has been executed and the synchronous data collection and recording are complete. Then, it searches for the target force and the actual force at the original target action time, calculates the signed error of each component, and saves the results of this round. Figure 2 Both loading branches enter the "Execute loading and synchronously collect and update historical records" phase, followed by "Calculate load reproduction error according to the original time scale". New valid records are added to the historical set of the same spectrum version; if there are lost samples, expired instructions or protection actions in the loading segment, the segment will not participate in the compliance judgment and will not be regarded as a valid learning record.
[0084] "Does the load reproduction error meet the predetermined requirements?" Execute after all records within the specified evaluation interval are complete. If not, use subsequent similar historical records for filtering; if yes, proceed to "Maintain convergence drive, collect shaft system response under specified load, and output test records." If the set number of tests or learning limit is reached before the requirements are met, output the non-compliance status and actual records, and do not substitute a waiting status or a single qualified sampling point for a complete qualified test.
[0085] Furthermore, the evaluation of the load reproduction error includes comparing the measured triaxial force with the peak value, impulse and waveform of the three-dimensional target load under a common original time scale, evaluating the deviation of the measured resultant force direction based on the resultant force direction at the same moment, evaluating the relative phase of the measured components according to the relative phase of the components specified in the target ice load spectrum, and not independently shifting the three-dimensional measurement curves afterward. Once the load reproduction error meets the predetermined requirements, the convergence drive is maintained to perform the specified loading, and the strain of the propulsion shaft system, bearing displacement or vibration response is collected. The actual load and the collected shaft system response are paired according to the original target action time, and the test record is output.
[0086] Specifically, the waveform errors in each direction are subtracted point by point at a common original timescale, and the maximum absolute value within a specified interval is taken. Peak value evaluation identifies dynamic extrema in both positive and negative directions, without converting negative peaks in the Y direction into unsigned positive peaks. Impulse is calculated using trapezoidal integrals at the same start and end times; the dynamic impulse is reduced by a known static bias, while the total force record retains that bias. Component phases are determined by comparing the relative intervals of the corresponding impact peaks or other pre-specified characteristic moments in the target spectrum with the difference between the actual interval and the target interval; when the target itself contains non-zero relative phases, forced in-phase behavior of the actual components is not required.
[0087] The resultant force direction deviation is calculated using the angle between the target force vector and the actual force vector at the same moment. The two non-zero vectors are first normalized, and their inner product is limited to between -1 and 1 before taking the inverse cosine. When the target resultant force or the actual resultant force is close to zero, causing the direction to be below the force measurement resolution, the direction indicator at that point is deemed unevaluable and cannot be set to 0. Whether the direction criterion for this interval is needed is determined before the test begins. The waveform, peak value, impulse, phase, and direction are compared with the tolerances given in the test task; if any required indicator is not met, the formal loading procedure cannot proceed.
[0088] To connect this example into a comprehensive calculation, it is assumed that the local equivalent linear relationship holds within a given envelope. The error waveforms of A and E are respectively multiplied by the aforementioned error vectors within the same envelope, and their paired drives are calculated by subtracting the error from the target. At 125 ms, the equivalent drives of X, Y, and Z forces of A are approximately 1078.721, -537.213, and 1622.128 kN, respectively, and those of E are approximately 1095.504, -555.045, and 1650.450 kN; after sharing and combining, they are approximately 1085.435, -544.346, and 1633.457 kN. After superimposing the aforementioned learning increments and sending them at 117, 113, and 105 ms, under this constructed relationship, the computational power at 125 ms is 1048, -390.5, and 1771 kN, respectively, with residual errors of 17, -9.5, and 29 kN. The peak values and impulse relative errors of the three dynamic components are all less than 3%.
[0089] If the criteria for this calculation experiment are set as follows: maximum absolute waveform error in each direction not exceeding 30 kN, relative error of dynamic peak and impulse not exceeding 3%, relative error of corresponding characteristic moments not exceeding 0.5 ms, and deviation of effective point resultant force direction not exceeding 0.5°, then the maximum directional deviation of the aforementioned ideal local calculation is approximately 0.283°, which can demonstrate a qualified branch. All other points within the spectrum are calculated using the same envelope and the same mapping, and the corresponding drive is only frozen after all required rotating backgrounds and complete evaluation intervals are completed. The above calculations cannot replace bench testing and acceptance.
[0090] The formal test maintains the frozen drive and original impact rhythm, continuously verifying the actual load. Shaft response is obtained using strain measurement channels, bearing displacement sensors, or vibration measurement channels, and paired with the actual triaxial forces at the original target application time. After the specified number of tests, the target, actual load, drive version, response, and pass / fail status are output. If subsequent loads exceed tolerances, the executed records are retained, and the pass / fail count is stopped. Learning loads are also included in the specimen's cumulative load, and the number of learning loads and formal test loads are recorded separately.
[0091] Example 2 This embodiment retains the fixed target spectrum, synchronous force and angle records, rotating background screening, three-dimensional joint correction, and original timescale evaluation, replacing the reciprocal weighting of multiple common records with the selection of a single common record as a whole. This change is suitable for situations where it is desirable to directly use a specific actual load record and its paired drive; it is implemented in parallel with the multi-record weighting of Embodiment 1, and the two combined results are not used simultaneously in the same update.
[0092] The program first searches for records by the same spectral segment and intraspectral time, and then performs checks on the endpoint angular difference, window increment difference, and rotation direction for each channel. After finding the intersection, it calculates the maximum angular deviation of the three windows for each record after integer alignment, and sorts them in ascending order of this deviation; if the deviations are the same, it prioritizes selecting the complete record with the newest acquisition time, and then eliminates ties by record number. Once selected, all three error components and three driving components of that record are read at once, without selecting additional records for a particular direction. This rule preserves the homogeneity of the three-dimensional data, but changes the selection method when there are multiple shared records.
[0093] by Figure 5 The qualification relationship shown is the input. Both A and E are jointly qualified records, while B failed to pass all windows. The maximum deviation of A is approximately 1.402°, less than E's approximately 2.103°. Therefore, A is selected, even if the error of a certain component of E is smaller, the component of E is not spliced. The errors assigned to the joint correction are 40, -25, and 70 kN, corresponding to the complete three-channel values of A being used in the drive. Using the local equivalence relationship and a learning coefficient of 0.5 from Example 1, the drive increments are approximately 21.578, -15.784, and 32.842 kN. The correction results are still sent at 117, 113, and 105 ms respectively, targeting the same 125 ms point of action.
[0094] Under this constructed relationship, the original timescale error of the next similar loading is 20, -12.5, and 35 kN. The 35 kN error in the Z-direction exceeds the 30 kN waveform error, therefore this new record is retained and returned to the screening process, but not included in the formal test. If the next new record with the same rotating background achieves the smallest trajectory deviation, or if the deviation is the same and it is prioritized due to being updated, then after performing the same joint correction again, the errors are 10, -6.25, and 17.5 kN. Under the same envelope assumption, the relative errors of dynamic peak and impulse in each direction are less than 3%, the maximum resultant force direction deviation is approximately 0.175°, and the original component phase is maintained. After the complete evaluation interval and each rotating background to be covered pass the predetermined criteria, the corresponding drive is frozen, the specified loading is executed, and the strain or displacement responses are paired, finally outputting the test record for that route. If a newer record fails the background screening, it is not forcibly replaced due to its newer time; a qualified record continues to be used, or the test enters a branch without candidates.
[0095] When changes in the specified speed operating conditions cause the intersection of candidate channels to be empty, this example does not consider the most recent record as a valid replacement. The controller executes the prepared basic feedforward and force closed loop, continuing to synchronously acquire data at the original impact rhythm; after the actual angle window of the new record is fully formed, it is re-compared with the subsequent trajectory to be executed. Only when a common qualified record appears will the entire selection be restored; if the prediction fails, the temperature range exceeds the limit, or the upper limit of this learning is reached, the time period, cause, and actual loading record of the unmet conditions are saved, and this non-compliance test ends. Thus, a single selection route has a clear driving source and result destination when there are multiple candidates, the error does not meet the standard, or there are no candidates.
[0096] Example 3 Please see Figure 6 This embodiment of a three-dimensional ice load simulation test system for an icebreaker propulsion shafting system includes: The spectrum processing module is used to decompose the pre-given and repeatedly played target ice load spectrum into three-dimensional target loads in a fixed test coordinate system, and to determine the target action time using a common time reference. The synchronous loading acquisition module is used to perform triaxial loading while the propulsion shaft system is rotating, synchronously acquire the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle; The historical filtering module is used to filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed, and the triaxial load error and corresponding drive are selected in groups. The drive correction and scheduling module is used to jointly correct the corresponding drive based on the cross response between channels and the three-dimensional load error selected in groups, to obtain the corrected drive, and to send the corrected drive ahead of the target action time according to the force response delay of each channel. The test evaluation module is used to execute the correction drive, calculate the load reproduction error with the original time-scale triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
[0097] Specifically, the spectrum processing module, corresponding to S1, reads load and coordinate information from the target spectrum file interface and transmits the three-dimensional target load and the target application time to the synchronous loading and acquisition module. The synchronous loading and acquisition module, corresponding to S2, receives the three-axis force from the three-axis force measurement channel and the shaft rotation angle from the shaft end encoder through its input interface, forming a complete three-dimensional load error, corresponding drive and shaft rotation angle historical record, which is then handed over to the historical filtering module, corresponding to S3.
[0098] The historical screening module transmits the selected triaxial load errors and corresponding drives to the drive correction and scheduling module corresponding to S4; the latter then passes the corrected drives and transmission times to the test evaluation module corresponding to S5. After verifying the executable status of the entire set of instructions, the test evaluation module sends a loading execution command to the triaxial electro-hydraulic servo loading channel through the real-time controller. Subsequently, it obtains the actual force and execution status returned by the synchronous loading acquisition module and decides whether to continue iteration, switch to the specified loading, or output the non-compliance record. Figure 6 The vertical lines represent the object transfer between the above functional modules. The data collection and return in the functional closed loop are handled by the same interface. The unidirectional main module chain is not interpreted as having no feedback.
[0099] Figure 3 and Figure 6 The deployment boundaries and functional module relationships are expressed separately. Spectrum processing, historical filtering, and joint correction calculations are executed by an industrial computer processor, with the program and complete historical records stored in memory; the timing queue, sampling latch, and execution status are maintained by a real-time controller. Each set of data transmitted by both carries the spectrum version and target action point identifier. The experimental evaluation module only allows the execution of instructions with the same version and complete submission. The basic loading branch also forms a new record through the same execution and acquisition interface, avoiding the use of different time bases between the correction branch and the basic branch.
[0100] The following combination Figures 7 to 9 Numerical experiments illustrating the above processing relationships are presented. Three figures are used to observe cross-cycle evolution, the correspondence between trajectory qualification and updated residuals, and the differences in the consistency of three-dimensional historical record sources under different combinations. The data are generated using a computational model and are not actual bench measurements, nor are the full-condition accuracy inferred from the local results in the figures.
[0101] like Figure 7 As shown, the horizontal axis represents the time within the spectrum, and the vertical axis represents the ice impact cycle number. The color code "Triaxial Load Error Modulus" represents the square root of the sum of the squares of the three load error components at the same original target time. The calculation includes 100 impact cycles, 1000 sampling points per cycle, a rotational speed of 54 r / min, a spectral frequency of 2 Hz, and a learning coefficient of 0.5. This numerical experiment uses response memory lengths of 20, 30, and 40 ms and pure delays of 8, 12, and 20 ms in the X, Y, and Z order. Therefore, a 40 ms window is used for the Z-axis. Figure 5 The candidate relationship examples belong to different channel configurations.
[0102] Figure 7The loading model first calculates the first-order residual response in each input channel, then multiplies and sums the corresponding coefficients in the cross-mapping of each channel response to form the anisotropic output force. The time constant is taken as the natural logarithm of the corresponding memory length divided by 100, so that the first-order free response decays to 1% of the initial value after that length; the nominal inverse compensation time constants are the actual values divided by 1.10, 1.08, and 1.12, respectively, and a small cross-response mismatch is retained. The correction adopts a common frequency band limitation of passing within 40 Hz, cosine transition from 40 to 80 Hz, and no compensation above 80 Hz; the 2 kHz instruction node is linearly interpolated on the 20 kHz local grid. The model continuously retains the filtering state and adds rotationally correlated loads, small and slow changes, prediction errors, and residual time differences of 30, -20, and 40 μs, without resetting the response state in each cycle.
[0103] There are no common qualified records in periods 1 to 20, so basic loading is used. Period 21 is the first time records from period 1 can be reused; the dashed line in the figure indicates the first available position of a common qualified record. The cross mark at 125 ms indicates the learning update position, taken from the actual executed update record; other positions within the spectrum are processed according to the same rules. The color generally lightens with the accumulation of comparable history, but there are periodic differences and non-zero residuals. This figure illustrates the relationship between qualification waiting and subsequent iterations under normal impact rhythms; it does not imply a strict halving in each round, nor does it prove that the entire measurement error budget has been met. The calculations in the figure construct the known model trajectory offline; actual online applications still require meeting the aforementioned prediction interval and transmission cutoff conditions.
[0104] like Figure 8 As shown, the horizontal axis represents the endpoint angle difference, and the vertical axis represents the maximum angle increment difference. A 3° endpoint angle tolerance and a 1° increment tolerance form two boundaries. The 12-level endpoint deviation and the 19-level speed deviation constitute 228 smooth, unidirectional rotational trajectories. The three channels still use 20, 30, and 40 ms windows for X, Y, and Z respectively. Dots represent records that are generally acceptable, triangles represent records that are acceptable only in terms of endpoint angle, and squares represent records that are unacceptable in terms of endpoint angle, with quantities of 30, 103, and 95 respectively. The position of the points is derived from the comparison of complete angle trajectories. The point shape is classified according to the qualification conditions that the endpoint angle difference does not exceed 3° and the increment difference does not exceed 1°, and the category is not selected by reverse selection based on the size of the calculated residual.
[0105] Figure 8The color-coded "Original Time-Scaled Load Residual" indicates the difference between the error of a historical record and the target force minus the calculated force vector magnitude obtained at the original action time of 125 ms after the entire driving group is used for the same reference trajectory and jointly updated. Historical records are first formed through three learning iterations under their respective rotational backgrounds, and then substituted offline into the reference working condition; filtered records are only used for comparison calculations and are not sent to the actual equipment. This model places the equivalent loads related to angle and rotational speed within the closed-loop memory response. For every 1° increase per second in rotational speed difference, the equivalent loads in X, Y, and Z increase by 0.055, -0.040, and 0.095 kN, respectively. These are constructed conditions and cannot be used to verify the equipment's 3 kN rotational background error budget.
[0106] Figure 8 Records that simultaneously satisfy both types of angle constraints occupy only the lower left area; some records with small endpoint angle differences are excluded due to different variations within the window. The calculated residuals are approximately 1.90 to 16.63 kN, with overlap between the residual ranges of qualified and rejected records. Therefore, this figure supports the ability of complete trajectory constraints to distinguish records with similar endpoints but different rotational experiences, but does not support the notion that the residuals of each rejected record must be larger. The geometric qualifications in the figure are calculated using the complete given trajectory; online prediction should also incorporate measurement and prediction uncertainty margins.
[0107] like Figure 9 As shown, six reference endpoint angles (0°, 60°, 120°, 180°, 240°, and 300°) are used, with four historical trajectories per group, totaling 24 historical records. Each group has 4, 3, and 2 candidates in the X, Y, and Z windows, respectively, with an intersection of 2 records in each group. Grouped combinations use shared weights to form three-dimensional load errors and corresponding drivers within a common range of qualified records. Independent channel-by-channel combinations use candidate records from each channel to form corresponding components. Both combination methods maintain the pairing of load errors and corresponding drivers within a single historical record; the difference lies in whether the three channels use the same historical record set and the same set of weights.
[0108] Figure 9 The horizontal axis is the reference working condition group, and the vertical axis is the number of distinct historical records used in the X, Y, and Z directions. Hollow circles represent group combinations, solid triangles represent independent channel-by-channel combinations, and connecting lines are only used to associate two combination methods under the same reference working condition, not to represent the time trajectory or the direction of result change. The record number ranges such as "01-02" and "01-04" in the figure represent the historical records involved in the combination in the corresponding direction.
[0109] For grouped combinations, the X, Y, and Z directions all use a common set of qualified records and the same set of weights. Therefore, the historical record sets corresponding to the three directions are the same, and the number of historical record sets that are different under each reference working condition is 1. For example, in group 01, the X, Y, and Z directions all use records 01-02, thus corresponding to one set of historical records.
[0110] For each channel-by-channel independent combination, each of the three directions uses its own candidate record set. In this numerical experiment, the historical record sets used in the three directions are different from each other. Therefore, the number of different historical record sets under each reference condition is 3. For example, in group 01, the X direction uses records 01-04, the Y direction uses records 01-03, and the Z direction uses records 01-02, thus forming three different historical record sets.
[0111] thus, Figure 9 This illustrates that grouped combinations can avoid triaxial load errors and cross-historical splicing caused by corresponding drives being formed from different historical record sets, ensuring that the three directions maintain a common historical record source relationship. This figure is used to illustrate the difference in historical record source relationship between grouped combinations and channel-by-channel independent combinations, and does not imply that the reproducibility index of any single load under all operating conditions is superior to that of channel-by-channel independent combinations.
[0112] Figure 9 Continue Figure 8 The dynamic model is constructed, and the same reference working condition and candidate record basis are used in each group for both combination methods. When comparing, the candidate qualifications and combination results are frozen, and the comparison of the source relationship of the figure is not interpreted as full-spectrum online execution or full-cycle compliance. Three numerical experiments are used to observe when comparable records appear, which trajectories are allowed to be reused, and the differences in the source relationship of the three-dimensional historical records under different combination methods. The actual experiment still verifies the original time-scaled load and shaft system response item by item according to S5.
[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for simulating three-dimensional ice loads on the propulsion shafting of an icebreaker, characterized in that, include: S1. In a fixed test coordinate system, the pre-given and repeatedly played target ice load spectrum is decomposed into three-dimensional target loads, and the target action time is determined by a common time reference. S2. Perform triaxial loading while the propulsion shaft system is rotating, simultaneously collect the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle. S3. Filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed. Select the triaxial load error and corresponding drive in groups. S4. Based on the cross-response between channels, the corresponding drive is jointly corrected by the three-dimensional load errors selected in groups to obtain the corrected drive, and the corrected drive is sent in advance to before the target action time according to the force response delay of each channel. S5. Execute the correction drive, calculate the load reproduction error using the original time-scaled triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
2. The method according to claim 1, characterized in that, The three-dimensional target load uses axial, vertical and lateral components, and is obtained by projecting the target ice load spectrum containing the magnitude and direction information of the three-dimensional components or the resultant force according to a given coordinate relationship. The fixed test coordinate does not rotate with the axis, and the projection retains the positive and negative directions, static offset and relative time difference between each component of the original load. The three-dimensional target load is resampled at a common sampling interval. The absolute timestamp of each sampling point is used for actual force comparison. The position of each sampling point within the repeated spectrum is used for historical record filtering. The three-dimensional target load remains unchanged during iteration.
3. The method according to claim 1, characterized in that, The synchronous acquisition of triaxial force and shaft rotation angle uses the same hardware time base latch loading interface for force measurement values and encoder angles, and corrects the fixed delay of the sensor chain so that the timestamp corresponds to the actual sampling time. The force measurement values are converted to the fixed test coordinates through a fixed direction coefficient and then used to form the triaxial load error. The historical record stores the three-channel drive at the moment of action and the unfolding shaft rotation angle within the corresponding response memory time, and the force measurement value represents the actual force transmitted to the propulsion shaft system through the loading interface.
4. The method according to claim 1, characterized in that, The filtering of the historical records includes: With the target action time to be executed as the endpoint, the corner comparison window is determined according to the response memory length of each channel. The response memory length covers the time required for the closed-loop residual response of the corresponding channel to enter the specified error band. The endpoint angle of the historical record is compared with the predicted endpoint angle of the loading to be performed by the shortest circumference distance, and the expanded angle increment relative to the endpoint is compared point by point within the angle comparison window. Only historical records with consistent rotation direction, endpoint angle difference not exceeding endpoint angle tolerance, and all angle increment differences within the angle comparison window not exceeding increment tolerance are used to select triaxial load errors and corresponding drives in groups.
5. The method according to claim 1, characterized in that, The triaxial load error and the corresponding drive group selection include: Taking the target action time to be executed as the endpoint, the comparison window is determined according to the response memory length of each of the three channels, and the candidate records that match the axis rotation angle change in each comparison window are obtained. The common qualified record is determined by the intersection of the candidate records of the three channels. For each record in the common qualified records, the maximum deviation between its axis rotation trajectory and the axis rotation trajectory to be loaded is obtained within the three comparison windows. The non-negative weight is determined according to the reciprocal of the maximum deviation. The non-negative weight is normalized. The lower limit of the maximum deviation is determined by the angle measurement resolution. By simultaneously combining the three load error components and three corresponding drive components in the common qualified record using the same set of normalized weights, a group of selected triaxial load errors and corresponding drives are obtained.
6. The method according to claim 1, characterized in that, The change in the shaft rotation angle to be loaded is predicted in a short time. The prediction is based on the collected unfolded shaft rotation angle, current rotation speed and specified rotation speed trajectory. The prediction interval covers the maximum advance transmission time and control processing time. The measured angle and the predicted angle are connected with the target action time as the endpoint. The comparison of the shaft rotation angle change adopts the angle tolerance, which is jointly limited by the encoder uncertainty, prediction error and the upper bound of the rotation-related force change, so that the allowable angle difference and the force change corresponding to the prediction error do not exceed the load error budget allocated to the rotation background difference. The historical records were obtained through a normal loading process that maintained the original ice impact time rhythm.
7. The method according to claim 1, characterized in that, The generation of the correction drive adopts the closed-loop cross-response inverse mapping after removing the pure channel delay under the accepted operating condition. The three load error components at the same moment of action are jointly converted into the force equivalent drive increment. The force equivalent drive increment is reduced by the learning coefficient and then superimposed on the corresponding drive selected in a group. The compensation frequency band is limited to the accepted frequency band where the closed loop is stable and the closed-loop cross-response inverse mapping is available. The learning coefficient is determined according to the upper bound of the mismatch of the comprehensive model after inverse compensation, so that the residual shrinkage coefficient is less than one. The residual boundary is determined by dividing the upper bound of the single-round uncertainty error by one and the difference with the residual shrinkage coefficient. This residual boundary is used to evaluate whether the allowable absolute error of each force component waveform is greater than the residual boundary.
8. The method according to claim 1, characterized in that, When the correction drive is sent ahead of the target action time, the transmission time of each channel is determined by deducting the force response delay of each channel for each target action time. Historical record filtering and drive correction are completed before the earliest transmission time. The command to be sent is formed by using the pre-given subsequent target ice load spectrum value and short-time axis rotation angle prediction. Each channel outputs the corresponding command at its own transmission time using a common time base, and linear interpolation is performed on adjacent commands locally in the driver. The three-way joint correction is completed before the channel time mapping. The transmission time of each channel corresponds to the same target action time.
9. The method according to claim 1, characterized in that, The evaluation of the load reproduction error includes comparing the measured triaxial force with the peak value, impulse and waveform of the three-dimensional target load under a common original time scale, evaluating the deviation of the measured resultant force direction based on the resultant force direction at the same moment, evaluating the relative phase of the measured components according to the relative phase of the components specified in the target ice load spectrum, and not independently shifting the three-dimensional measurement curves afterward. Once the load reproduction error meets the predetermined requirements, the convergence drive is maintained to perform the specified loading, and the strain of the propulsion shaft system, bearing displacement or vibration response is collected. The actual load and the collected shaft system response are paired according to the original target action time, and the test record is output.
10. A three-dimensional ice load simulation test system for an icebreaker propulsion shafting system, characterized in that, include: The spectrum processing module is used to decompose the pre-given and repeatedly played target ice load spectrum into three-dimensional target loads in a fixed test coordinate system, and to determine the target action time using a common time reference. The synchronous loading acquisition module is used to perform triaxial loading while the propulsion shaft system is rotating, synchronously acquire the triaxial force and shaft rotation angle, subtract the triaxial force from the triaxial target load at the same time the target is applied to obtain the triaxial load error, establish a historical record, and save the triaxial load error, the corresponding drive and shaft rotation angle; The historical filtering module is used to filter the historical records so that they have the same spectral segment and intraspectral time as the load to be executed, and the shaft rotation angle change during the response memory time matches the shaft rotation angle change of the load to be executed, and the triaxial load error and corresponding drive are selected in groups. The drive correction and scheduling module is used to jointly correct the corresponding drive based on the cross response between channels and the three-dimensional load error selected in groups, to obtain the corrected drive, and to send the corrected drive ahead of the target action time according to the force response delay of each channel. The test evaluation module is used to execute the correction drive, calculate the load reproduction error with the original time-scale triaxial force, and complete the simulation test by iterating through subsequent similar historical records based on the load reproduction error.
