Servo system control optimization method and system applied to intelligent creep machine

Abstract

The application provides a servo system control optimization method and system applied to an intelligent creep machine, relates to the technical field of material performance test equipment control, and first acquires a servo driving time sequence signal set of the intelligent creep machine in a continuous loading time period, which contains time sequence signals such as servo motor winding temperature, current, rotating speed, screw shaft axial force and displacement; then the servo driving time sequence signal set is subjected to multi-resolution signal decomposition processing to generate a time-frequency localized feature set reflecting signal mutation characteristics; then a pre-constructed creep system dynamic response model is called to perform creep state correlation analysis to generate a creep process state transition representation vector; finally, the servo control parameter adjustment constraint boundary is determined according to the creep process state transition representation vector to generate a control parameter optimization instruction. The application can dynamically adjust the servo control parameter according to the material creep stage to improve test accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of control technology for material performance testing equipment, and more specifically, to a servo system control optimization method and system applied to an intelligent creep machine. Background Technology

[0002] In materials performance testing, intelligent creep machines are used to study the creep characteristics of materials under sustained loads. Traditional intelligent creep machines employ a relatively simple servo system control method, typically controlling the servo motor based solely on preset fixed parameters. However, materials undergo different stages during creep, including transient creep, steady-state creep, and accelerated creep, each with significantly different mechanical properties and deformation characteristics.

[0003] Existing servo system control methods struggle to adjust control parameters in real time according to the dynamic changes during the material creep stage. For example, in the transient creep stage, the material deforms rapidly, and if the servo control parameters cannot adapt to these changes in a timely manner, it may lead to inaccurate test data. In the steady-state creep stage, the material deformation is relatively stable, but unreasonable control parameters may introduce unnecessary errors. In the accelerated creep stage, the material is close to failure, requiring even higher precision in the control parameters, which traditional methods struggle to meet. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a servo system control optimization method applied to an intelligent creep machine, the method comprising: The set of servo drive timing signals of the intelligent creep machine during the continuous loading period is obtained. The set of servo drive timing signals includes servo motor winding temperature timing signal, servo motor current timing signal, servo motor speed timing signal, lead screw axial force timing signal, and lead screw displacement timing signal. The servo drive timing signal set is subjected to multi-resolution signal decomposition processing to obtain the approximate component sequence and detail component sequence of each signal component on the multi-order decomposition layer, and a time-frequency localization feature set reflecting the signal change characteristics is generated based on the approximate component sequence and the detail component sequence. The pre-constructed dynamic response model of the creep system is invoked to perform creep state correlation analysis on the time-frequency localized feature set, generating a creep process state transition characterization vector. The creep process state transition characterization vector is used to characterize the evolution direction and degree of the material in the intelligent creep machine from the transient creep stage to the steady-state creep stage and then to the accelerated creep stage under continuous load. Based on the creep process state transition characterization vector, the adjustment constraint boundary of the servo control parameters under the current creep stage is determined, and based on the adjustment constraint boundary, the control parameter optimization instruction for dynamically calibrating the servo control parameters is generated.

[0005] Furthermore, embodiments of the present invention also provide a servo system control optimization system applied to an intelligent creep machine, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described servo system control optimization method for intelligent creep machines by executing the machine-executable instructions.

[0006] Based on the above, by acquiring a set of servo drive timing signals containing various time-series signals such as servo motor winding temperature, current, speed, lead screw axial force, and displacement during the continuous loading period of the intelligent creep machine, and performing multi-resolution signal decomposition processing on the servo drive timing signal set to generate a time-frequency localized feature set reflecting the signal abrupt change characteristics, it is possible to deeply explore the subtle changes hidden in the signals and effectively identify key feature points in the material creep process. A pre-constructed creep system dynamic response model is used for creep state correlation analysis to generate a creep process state transition characterization vector, which can accurately characterize the evolution direction and degree of the material at each creep stage under continuous load. Based on this creep process state transition characterization vector, the adjustment constraint boundary of the servo control parameters under the current creep stage is determined, and control parameter optimization instructions are generated. This allows for real-time and precise adjustment of servo control parameters according to the dynamic changes in the material creep stage, significantly improving the accuracy, reliability, and adaptability of intelligent creep machine testing. Attached Figure Description

[0007] Figure 1 This is a schematic diagram of the execution flow of the servo system control optimization method for intelligent creep machines provided in an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of exemplary hardware and software components of a servo system control optimization system for an intelligent creep machine, provided in an embodiment of the present invention. Detailed Implementation

[0009] Figure 1 This is a flowchart illustrating a servo system control optimization method for an intelligent creep machine according to an embodiment of the present invention, which will be described in detail below.

[0010] The servo system control optimization method for intelligent creep machines provided in this application can be applied to material creep performance testing scenarios. In this scenario, the intelligent creep machine applies a constant or stepped load to the specimen, and uses a servo motor to drive a lead screw to perform force and displacement control, while simultaneously acquiring the operating status signals of the servo drive system. Under continuous load, the specimen undergoes transient creep, steady-state creep, and accelerated creep stages. The control parameters of the servo system are dynamically adjusted according to the creep stage evolution to ensure testing accuracy and equipment operational safety.

[0011] Step S110: Obtain the set of servo drive timing signals of the intelligent creep machine during the continuous loading period. The set of servo drive timing signals includes servo motor winding temperature timing signal, servo motor current timing signal, servo motor speed timing signal, lead screw axial force timing signal, and lead screw displacement timing signal.

[0012] The servo motor winding temperature timing signal is acquired by a platinum resistance temperature sensor embedded in the end of the servo motor stator winding, converted into a digital signal by an analog-to-digital converter, with a sampling period of [missing information]. The servo motor current timing signal is acquired from the servo driver current detection circuit by a Hall current sensor, with the sampling period synchronized with the temperature signal. The servo motor speed timing signal is obtained by quadrature pulses output from an incremental photoelectric encoder, counted using a fourfold frequency multiplication, and converted to a period. The lead screw axial force timing signal is acquired by a strain gauge force sensor installed at the end of the lead screw, and converted into a digital quantity after conditioning by a Wheatstone bridge and an instrumentation amplifier. The lead screw displacement timing signal is acquired by a grating ruler reading head. All five signals are acquired synchronously at a unified sampling clock, each forming a one-dimensional array. The same index position in the array corresponds to the same sampling time, collectively constituting the servo drive timing signal set. Each signal has been converted into an engineering value with physical units using sensor calibration coefficients, and is dimensionless before entering multi-resolution decomposition using its own preset normalization parameters.

[0013] Step S120: Perform multi-resolution signal decomposition processing on the servo drive timing signal set to obtain the approximate component sequence and detail component sequence of each signal component on the multi-level decomposition layer, and generate a time-frequency localization feature set reflecting the signal change characteristics based on the approximate component sequence and detail component sequence.

[0014] Step S121: Perform multi-resolution signal decomposition processing on the servo motor winding temperature timing signal, and peel the servo motor winding temperature timing signal layer by layer into a temperature slowly varying component sequence and a temperature transient component sequence. The temperature slowly varying component sequence retains the low-frequency contour features of the servo motor winding temperature timing signal on each decomposition layer, and the temperature transient component sequence retains the high-frequency oscillation features of the servo motor winding temperature timing signal on each decomposition layer.

[0015] Multi-resolution signal decomposition processing employs the discrete wavelet transform algorithm, with the Daubechies wavelet chosen as the wavelet basis function due to its compact support and high-order vanishing moment characteristics. The total decomposition level is set to J. This is applied to the servo motor winding temperature timing signal. Layer-by-layer decomposition is employed, with the filter bank using orthogonal mirror filters. The low-pass filter coefficient sequence is H, and the high-pass filter coefficient sequence is G. In the j-th layer, the approximate coefficient sequence of the (j-1)-th layer is convolved with H and G respectively, followed by a 2x downsampling to obtain the approximate coefficient sequence of the j-th layer. and the detail coefficient sequence of the j-th layer .right Single-branch reconstruction yields temperature-gradient component sequences. ,right Single-branch reconstruction yields the temperature transient component sequence. . reserve Low-frequency contour features at layer j reserve High-frequency oscillation characteristics at the j-th layer. j is traversed from 1 to J to complete the J-th layer decomposition.

[0016] Step S122: Perform multi-resolution signal decomposition processing on the servo motor current timing signal, and peel the servo motor current timing signal layer by layer into a current slowly varying component sequence and a current transient component sequence. The current slowly varying component sequence retains the low-frequency contour features of the servo motor current timing signal on each decomposition layer, and the current transient component sequence retains the high-frequency oscillation features of the servo motor current timing signal on each decomposition layer.

[0017] Using the same discrete wavelet transform algorithm and parameters as in step S121, the servo motor current timing signal is processed. Perform J-level decomposition to obtain the j-th level slowly varying current component sequence. and the transient component sequence of the j-th layer current .

[0018] Step S123: Perform multi-resolution signal decomposition processing on the servo motor speed timing signal, and peel the servo motor speed timing signal layer by layer into a speed gradually changing component sequence and a speed transient component sequence. The speed gradually changing component sequence retains the low-frequency contour features of the servo motor speed timing signal on each decomposition layer, and the speed transient component sequence retains the high-frequency oscillation features of the servo motor speed timing signal on each decomposition layer.

[0019] For servo motor speed timing signal Perform the same processing to obtain the j-th layer of slowly varying speed components sequence. and the j-th layer speed transient component sequence .

[0020] Step S124: Perform multi-resolution signal decomposition processing on the screw axial force timing signal, and peel the screw axial force timing signal into force gradually varying component sequence and force transient component sequence layer by layer. The force gradually varying component sequence retains the low-frequency profile features of the screw axial force timing signal on each decomposition layer, and the force transient component sequence retains the high-frequency oscillation features of the screw axial force timing signal on each decomposition layer.

[0021] The same processing is performed on the axial force timing signal Xf of the lead screw to obtain the j-th layer of force gradually varying component sequence. and the j-th layer force transient component sequence .

[0022] Step S125: Perform multi-resolution signal decomposition processing on the lead screw displacement time sequence signal, and peel the lead screw displacement time sequence signal layer by layer into a displacement slowly varying component sequence and a displacement transient component sequence. The displacement slowly varying component sequence retains the low-frequency contour features of the lead screw displacement time sequence signal on each decomposition layer, and the displacement transient component sequence retains the high-frequency oscillation features of the lead screw displacement time sequence signal on each decomposition layer.

[0023] For the timing signal of the lead screw displacement Perform the same processing to obtain the j-th layer displacement gradually varying component sequence. and the transient component sequence of the j-th layer displacement .

[0024] Step S126: Perform abrupt change point localization processing on each layer of temperature transient component sequence, current transient component sequence, rotational speed transient component sequence, force transient component sequence, and displacement transient component sequence, extract the time position and intensity of energy abrupt changes in each time series signal of each layer, and generate a subset of abrupt change position identifiers and a subset of abrupt change intensity identifiers for each signal in each decomposition layer; align the subsets of abrupt change intensity identifiers of all signals in the same decomposition layer according to the time index to obtain the combined abrupt change intensity sequence of each decomposition layer.

[0025] For the temperature transient component sequence of the j-th layer Perform mutation point localization processing. Set the sliding window length to Lw and the sliding step size to 1. For the k-th time index, calculate the value within the sliding window. The local energy value is obtained by summing the squares of the sampled values, and the calculation formula is as follows: .right The sequence calculation yields first-order and second-order difference sequences. It detects zero-crossing points where adjacent elements in the first-order difference sequence have opposite signs. If the second-order difference value corresponding to a zero-crossing point is negative, this position is identified as a breakpoint, and the sequence is marked accordingly. The value is used as the mutation strength. The set of detected mutation point locations is denoted as... The set of mutation intensity is denoted as Use the same method on Each signal undergoes mutation point localization processing to obtain a subset of mutation position identifiers and a subset of mutation intensity identifiers for each signal at layer j. The mutation intensity identifier subsets of the five signals at layer j are aligned along the same time index. If a signal has no mutation point at a certain time index, the mutation intensity at that position is set to 0. After alignment, a five-dimensional vector is obtained at each time index, and these vectors are arranged along the time indices to form the mutation intensity sequence of the combined signals at layer j. .

[0026] Step S127: Perform mutation point localization processing on each layer of mutation intensity sequence, extract the time position and mutation intensity of energy mutation in the time series signal of each layer, and generate the mutation position identifier set and mutation intensity identifier set corresponding to each layer.

[0027] mutation intensity sequence of the j-th layer combination The maximum component value of the five-dimensional vector at each time index is extracted as the comprehensive mutation intensity value for that index. The mutation point location processing is performed on the comprehensive mutation intensity value sequence using the same sliding window energy detection and peak search method as in step S126, and the detected mutation point locations are stored in the mutation location identifier set corresponding to the j-th layer. Store the comprehensive mutation intensity value at the corresponding position into the mutation intensity identifier set corresponding to the j-th layer. Perform the above processing on j from 1 to J.

[0028] Step S128: Perform inter-layer trend coupling processing on the temperature slowly varying component sequence, current slowly varying component sequence, rotational speed slowly varying component sequence, force slowly varying component sequence, and displacement slowly varying component sequence. The inter-layer trend coupling processing is used to extract the degree of trend consistency and trend deviation between each slowly varying component sequence on the same decomposition layer, and generate the trend coupling degree distribution on each decomposition layer.

[0029] For the j-th layer, obtain and Set the local fitting window length to Lfit. At the k-th time index, take the index interval for each slowly varying component sequence. Local linear fitting is performed on the sampled values ​​within the range to obtain the local trend slope sequence of each slowly varying component sequence at that index, denoted as . For each of the five slope values ​​mentioned above, calculate the Pearson correlation coefficient pairwise. Use the arithmetic mean of all Pearson correlation coefficients as the trend coupling degree value at the k-th time index. By iterating through all k, the trend coupling degree distribution of the j-th layer is obtained. .

[0030] Step S129: Based on the combined mutation intensity sequence, mutation location identifier subset, and trend coupling degree distribution, construct a time-frequency localization feature map with the decomposition layer as the vertical axis and the time index as the horizontal axis. Concatenate the time-frequency localization feature maps corresponding to all decomposition layers along the vertical axis to generate a time-frequency localization feature set that reflects the distribution law of signal mutation features across the entire frequency range. Each coordinate node in the time-frequency localization feature map records the mutation intensity value and trend coupling degree value at the corresponding time on the decomposition layer.

[0031] For the j-th layer, and Align on a unified time index. At the k-th time index, take... The overall mutation intensity value and The trend coupling degree values ​​are grouped into tuples, which serve as the coordinate node feature values ​​of the j-th row and k-th column. The j-th row is obtained by traversing all k. The above operation is repeated for all j from 1 to J, and the J rows are stacked to form a time-frequency localized feature set F, which is a two-dimensional feature structure with J rows and multiple columns.

[0032] Step S130: Call the pre-built dynamic response model of the creep system, perform creep state correlation analysis on the time-frequency localized feature set, and generate creep process state transition characterization vector. The creep process state transition characterization vector is used to characterize the evolution direction and degree of the material in the intelligent creep machine from the transient creep stage to the steady-state creep stage and then to the accelerated creep stage under continuous load.

[0033] Step S131: Input the time-frequency localized feature set into the feature coding layer of the dynamic response model of the creep system. The feature coding layer performs time-dimensional sequence coding on the mutation intensity value and trend coupling degree value of each decomposition layer in the time-frequency localized feature set to generate the time-series feature coding vector corresponding to each decomposition layer.

[0034] The feature encoding layer consists of J parallel independent recurrent neural network units. For the j-th decomposition layer, the sequence of all pairs in the j-th row is taken from F, and the two values ​​of each pair are used as a two-dimensional input vector. Where t is the time step number. The hidden state update formula for the j-th independent recurrent neural network unit is: ,in Let be the hidden state vector at time step t. Let be the two-dimensional input vector at time step t, W be the input weight matrix, and u be the cyclic weight vector. Let represent element-wise multiplication, b be the bias vector, and σ be the hyperbolic tangent activation function. After processing all time steps of the j-th layer, the hidden state vector of the last time step is taken as the temporal feature encoding vector of the j-th decomposition layer.

[0035] Step S132: Input the temporal feature encoding vectors corresponding to each decomposition layer into the inter-layer information transfer layer of the dynamic response model of the creep system. The inter-layer information transfer layer transfers the temporal feature encoding vectors of the lower decomposition layers to the higher decomposition layers layer by layer and fuses them with the temporal feature encoding vectors of the higher decomposition layers to generate inter-layer progressive fusion feature vectors.

[0036] Let the temporal feature encoding vector of the j-th decomposition layer be... Starting from layer 1, let the transfer vector... Where f1 is a fully connected mapping. For layers 2 to J, the gate vector... , pass vector ,in This represents vector concatenation, where σ is the logistic activation function, and W_z and b_z are gating parameters. The final g_J is the inter-layer progressively fused feature vector.

[0037] Step S133: Input the interlayer progressive fusion feature vector into the creep stage discrimination layer of the creep system dynamic response model. The creep stage discrimination layer performs stage boundary identification processing on the interlayer progressive fusion feature vector and outputs the preliminary stage identification sequence corresponding to the current time window. The preliminary stage identification sequence is used to distinguish between transient creep stage, steady-state creep stage and accelerated creep stage.

[0038] The creep stage discrimination layer consists of a fully connected layer and a flexible maximum transfer function layer. The fully connected layer maps the inter-layer progressively fused feature vector g_J to a three-dimensional stage scoring vector s = W_s·g_J + b_s. The flexible maximum transfer function layer normalizes s to obtain the stage probability vector pstage = softmax(s). The stage index corresponding to the maximum value in pstage is taken as the stage determination result for the current time window. Discrimination is performed one by one for multiple consecutive time windows covered by F to obtain a preliminary stage identifier sequence.

[0039] Step S134: Input the preliminary stage identifier sequence into the stage evolution tracking layer of the creep system dynamic response model. The stage evolution tracking layer tracks the stage migration path of the preliminary stage identifier sequence in a continuous time window and generates a stage migration trajectory descriptor. The stage migration trajectory descriptor records the direction and number of changes of the creep stage between adjacent time windows.

[0040] The stage evolution tracking layer is implemented using a finite state machine, with the state set {0, 1, 2} corresponding to three creep stages. The initial stage identifier sequence is input sequentially according to the time window. Each input value is compared with the current state: if the input value equals the current state, no transition event is recorded; if the input value does not equal the current state, a transition event is recorded. The transition event attributes include the starting state, the target state, and the transition direction. The transition direction is determined by subtracting the sign of the starting state from the target state. The current state is then updated to the input value. After traversing all time windows, the positive transition event sequence, the total number of positive transitions, and the total number of negative transitions are counted and combined to form the stage transition trajectory descriptor.

[0041] Step S135: Input the interlayer progressive fusion feature vector and the stage migration trajectory descriptor into the state transition modeling layer of the creep system dynamic response model. The state transition modeling layer performs dynamic weighting processing on the interlayer progressive fusion feature vector according to the stage change direction recorded in the stage migration trajectory descriptor, so that the temporal features in the interlayer progressive fusion feature vector that are consistent with the stage change direction obtain enhanced response and generate state transition driving features.

[0042] The progressively fused feature vector g_J is denoted as Fvec. Fvec is then mapped to a query vector through a fully connected layer. The target state index of the latest positive migration event is read from the stage migration trajectory descriptor, and from three sets of preset key vector templates. Select the corresponding template. For each key vector in the template... Calculate attention score Attention weights are obtained by normalization using the flexible maximum transfer function. The state transition-driven feature Fdrv is obtained by concatenating the attention weights corresponding to each local feature interval in Fvec after weighting each local feature vector.

[0043] Step S136: Input the state transition driving features into the evolution parameter estimation layer of the dynamic response model of the creep system. The evolution parameter estimation layer performs evolution direction projection processing and evolution degree quantification processing on the state transition driving features to generate evolution direction vector and evolution degree parameter. The evolution direction vector is used to indicate the transition tendency of the creep state from the current stage to the next stage, and the evolution degree parameter is used to indicate the progress depth in the creep stage at the current moment.

[0044] The evolution parameter estimation layer contains two parallel fully connected output branches. The first branch is the evolution direction branch, where the output 3D vector is normalized using a flexible maximum transfer function to obtain the evolution direction vector. The second branch is the evolution degree branch. The output scalar is mapped to the interval [0, 1] through a logical activation function to obtain the evolution degree parameter. .

[0045] Step S137: Combine and encode the stage identifier, evolution direction vector and evolution degree parameter of the current moment in the preliminary stage identifier sequence to generate a creep process state transition representation vector containing stage attribute dimension, direction attribute dimension and degree attribute dimension.

[0046] The current stage identifier is converted into a one-hot encoded vector Soh, which is a three-dimensional vector. Soh, Vdir, and r are concatenated along the feature dimension to obtain the creep process state transition representation vector Vcreep=[Soh; Vdir; r], which is a 7-dimensional vector.

[0047] Step S140: Based on the state transition characterization vector of the creep process, determine the adjustment constraint boundary of the servo control parameters under the current creep stage, and generate control parameter optimization instructions for dynamic calibration of the servo control parameters based on the adjustment constraint boundary.

[0048] Step S141: Extract the stage attribute dimension component from the creep process state transition representation vector, and determine the current creep stage as one of the following stage types: transient creep stage, steady-state creep stage, or accelerated creep stage, based on the stage attribute dimension component.

[0049] The first three dimensions of the vector are extracted from Vcreep as the stage attribute dimension components. The maximum value is taken as the index 0, 1, and 2, which correspond to the transient creep stage, steady-state creep stage, and accelerated creep stage, respectively.

[0050] Step S142: Extract the directional attribute dimension component from the creep process state transition representation vector, and determine the transition trend direction corresponding to the current creep stage based on the directional attribute dimension component. The transition trend direction is used to indicate the downstream stage type of creep state evolution.

[0051] Extract the 4th to 6th dimensions of the vector from Vcreep as the directional attribute dimension component Vdir, and take the stage type corresponding to the maximum value. If the stage type is not equal to the current stage type, the transition direction points to the stage type; otherwise, the transition direction remains unchanged.

[0052] Step S143: Extract the degree attribute dimension component from the creep process state transition representation vector, and determine the progress depth parameter in the current creep stage based on the degree attribute dimension component. The progress depth parameter is used to characterize the position ratio of the current time in the current creep stage.

[0053] Extract the 7th scalar from Vcreep as the progress depth parameter p, p∈[0,1].

[0054] Step S144: Obtain the pre-stored stage parameter constraint mapping relationship. The stage parameter constraint mapping relationship records the first parameter constraint range corresponding to the transient creep stage, the second parameter constraint range corresponding to the steady-state creep stage, and the third parameter constraint range corresponding to the accelerated creep stage. The first parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The second parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The third parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The value ranges of the first, second, and third parameter constraint ranges are different from each other. According to the stage type and progress depth parameter of the current creep stage, perform constraint boundary shrinkage processing within the corresponding parameter constraint range. The constraint boundary shrinkage processing adjusts the corresponding upper limit threshold of motor current and the upper limit threshold of motor speed from the initial boundary value inward according to the progress depth parameter and the preset shrinkage rule, generating current constraint boundary, speed constraint boundary, and force constraint boundary.

[0055] Let the stage parameter constraint mapping relationship be: the upper limit of current corresponding to the transient creep stage. and maximum speed The upper limit of current corresponding to the steady-state creep stage and maximum speed The upper limit of current corresponding to the accelerated creep stage and maximum speed ,and Determine the initial current upper limit based on the current stage type. and initial speed limit The contraction rule is a nonlinear contraction function with current-constrained boundaries. Rotational speed constraint boundary The force constraint boundary FB is determined by multiplying the upper limit of the force sensor's range by a safety factor η. .

[0056] Step S145: Obtain the real-time current value of the servo motor current timing signal, the real-time speed value of the servo motor speed timing signal, and the real-time force value of the lead screw axial force timing signal. Generate the current adjustment direction based on the difference between the current constraint boundary and the real-time current value, generate the speed adjustment direction based on the difference between the speed constraint boundary and the real-time speed value, and generate the force adjustment direction based on the difference between the force constraint boundary and the real-time force value.

[0057] Read the real-time current value I_R, real-time speed value V_R, and real-time force value F_R at the current sampling moment. Calculate the current deviation EI = I_R - IB. If EI > 0, the current adjustment direction DI is to decrease the current; otherwise, it is to maintain or increase the current. Similarly, calculate the speed deviation EV = V_R - VB to determine the speed adjustment direction DV, and calculate the force deviation EF = F_R - FB to determine the force adjustment direction DF.

[0058] Step S146: The current adjustment direction, speed adjustment direction, and force adjustment direction are used as the current setpoint adjustment amount of the servo current loop, the speed setpoint adjustment amount of the servo speed loop, and the force setpoint adjustment amount of the servo force loop, respectively, to generate dynamic calibration values ​​of servo control parameters for the servo current loop, servo speed loop, and servo force loop.

[0059] ΔIref = K_P·EI, where K_P is the proportional control gain function, and the output, after being limited, is used as the current setpoint adjustment. ΔVref = K_P·EV, after being limited, is used as the speed setpoint adjustment. ΔFref = K_P·EF, after being limited, is used as the force setpoint adjustment. ΔIref, ΔVref, and ΔFref together constitute the dynamic calibration values ​​of the servo control parameters.

[0060] Step S147: Generate control parameter optimization instructions based on the dynamic calibration values ​​of the servo control parameters. The control parameter optimization instructions are used to drive the servo current loop to perform current adjustment operations, drive the servo speed loop to perform speed adjustment operations, and drive the servo force loop to perform force adjustment operations.

[0061] ΔIref, ΔVref, and ΔFref are encapsulated according to a preset protocol format. The message header contains an instruction type identifier and a length field, and the payload section stores the three adjustment values ​​sequentially. The encapsulated instruction data packet is written to the instruction receive buffer of the servo driver via the fieldbus. After parsing, the servo driver adds ΔIref to the current loop current setpoint, ΔVref to the current speed loop current setpoint, and ΔFref to the current force loop current setpoint, respectively.

[0062] Step S210: Obtain the set of historical servo drive signals accumulated by the intelligent creep machine during the historical loading cycle. The set of historical servo drive signals includes historical servo motor winding temperature timing signals, historical servo motor current timing signals, historical servo motor speed timing signals, historical lead screw axial force timing signals, and historical lead screw displacement timing signals, as well as manually marked records of creep stages corresponding to each historical timing signal.

[0063] The historical test database retrieves records of all previous creep tests of the same model as the current intelligent creep machine and the same material grade as the current specimen. Each historical test record contains a complete set of historical servo drive signals, as well as manually annotated records written by the test engineer after the test, based on the deformation curve characteristics and fracture microscopic analysis results of the specimen. The manually annotated records use the sampling time as an index to clearly mark the creep stage at each sampling time. A stage label value of 0 corresponds to the transient creep stage, a value of 1 corresponds to the steady-state creep stage, and a value of 2 corresponds to the accelerated creep stage.

[0064] Step S220: Perform multi-resolution signal decomposition processing on the historical servo motor winding temperature timing signal, historical servo motor current timing signal, historical servo motor speed timing signal, historical lead screw axial force timing signal, and historical lead screw displacement timing signal respectively to obtain the historical approximate component sequence and historical detail component sequence of each historical timing signal on the multi-level decomposition layer.

[0065] Using the same discrete wavelet transform parameters and filter banks as in steps S121 to S125, J-level decomposition is performed on each historical time series signal. Single-branch reconstruction is performed on the j-th level to obtain the historical approximate component sequence and the historical detail component sequence. The decomposition results maintain the same frequency resolution characteristics as in the online processing stage.

[0066] Step S230: Based on the historical approximate component sequence and the historical detail component sequence, perform abrupt feature extraction and trend coupling degree calculation on each historical time series signal to generate a historical time-frequency localized feature sample set. Each sample in the historical time-frequency localized feature sample set is labeled with a corresponding creep stage label.

[0067] Using the same mutation point localization processing, combined mutation intensity sequence generation, inter-layer trend coupling processing, and time-frequency localized feature map construction methods as steps S126 to S129, the multi-resolution decomposition results of each historical time series signal are processed to generate a historical time-frequency localized feature set. For the feature data at each time window position in the historical time-frequency localized feature set, the creep stage label is extracted from the manually labeled record corresponding to that time window as the supervision label for that sample. Each sample consists of input features and a label. The input features are the feature value matrix of the coordinate nodes of the corresponding time window in the time-frequency localized feature map, and the label is the labeled creep stage index value.

[0068] Step S240: Construct an initial creep system dynamic response model. The initial creep system dynamic response model includes a feature encoding layer, an inter-layer information transmission layer, a creep stage discrimination layer, a stage evolution tracking layer, a state transition modeling layer, and an evolution parameter estimation layer connected in sequence. The network weight parameters of each layer are set using a random initialization method.

[0069] The input weight matrices of the independent recurrent neural network units in the feature encoding layer are initialized using a Xavier uniform distribution, the recurrent weight vectors are initialized as constant vectors with all elements having fixed values, and the bias terms are initialized to zero. The weight matrices of the fully connected mapping in the inter-layer information transfer layer and the fully connected weight matrices of the gated fusion units are initialized using a Xavier uniform distribution. The weight matrices of the fully connected layers in the creep stage discrimination layer are initialized using a Xavier uniform distribution. The weight matrices of the fully connected layers in the query vector mapping of the state transition modeling layer are initialized using a Xavier uniform distribution, and the three sets of key vector templates are randomly initialized using a truncated normal distribution with a mean of zero and a set standard deviation. The weight matrices of the fully connected layers in the two output branches of the evolution parameter estimation layer are initialized using a Xavier uniform distribution. The stage evolution tracking layer is a parameterless finite state machine and does not involve weight parameters.

[0070] Step S250: Input the set of historical time-frequency localized feature samples into the initial creep system dynamic response model, generate sample time-series feature encoding vectors through the feature encoding layer, generate sample inter-layer progressive fusion feature vectors through the inter-layer information transfer layer, output sample preliminary stage identification sequence through the creep stage discrimination layer, output sample stage migration trajectory descriptor through the stage evolution tracking layer, output sample state transition driving features through the state transition modeling layer, and output sample evolution direction vector and sample evolution degree parameter through the evolution parameter estimation layer.

[0071] Each sample in the historical time-frequency localized feature sample set is propagated forward sequentially through each layer of the initial creep system dynamic response model. The processing logic of each layer is exactly the same as the processing logic of the online inference stage described in steps S131 to S136, the only difference being that the processing object is historical sample data. The intermediate variables output by each layer are respectively denoted as the sample time-series feature encoding vector, the sample inter-layer progressive fusion feature vector, the sample initial stage identifier sequence, the sample stage migration trajectory descriptor, the sample state transition driving feature, the sample evolution direction vector, and the sample evolution degree parameter.

[0072] Step S260: Perform cross-entropy loss calculation on the preliminary stage identifier sequence of the sample and the creep stage label to obtain the stage discrimination loss value; perform direction deviation loss calculation on the sample evolution direction vector and the standard evolution direction vector determined according to the creep stage label to obtain the evolution direction loss value; and perform degree deviation loss calculation on the sample evolution degree parameter and the internal progress reference value of the stage determined according to the creep stage label to obtain the evolution degree loss value.

[0073] Stage discriminant loss value The calculation formula is: ,in Iterate through the three creep stage categories, where y_c is the one-hot encoded vector of the creep stage label. The dimensional component, p_c, represents the stage probability vector of the time window corresponding to the initial stage identifier sequence of the sample. Dimensional components. Evolutionary direction loss value. The calculation formula is: Where Vdir is the sample evolution direction vector, and Vstd is the one-hot encoded vector for the next stage determined based on the creep stage label. Evolution degree loss value. The calculation formula is: , where r is the sample evolution degree parameter, rref is the internal progress reference value of the stage, and rref is calculated based on the proportion of the start and end time of the current creep stage marked in the manual annotation record of the time window.

[0074] Step S270: The stage discrimination loss value, evolution direction loss value, and evolution degree loss value are weighted and fused to generate a joint training total loss value. Based on the joint training total loss value, the network weight parameters of each layer in the initial creep system dynamic response model are updated by back gradient propagation. The weighted fusion process and back gradient propagation update process are iteratively executed until the joint training total loss value meets the preset convergence condition. The network weight parameters that meet the preset convergence condition are solidified to obtain the creep system dynamic response model and store it in the model storage unit.

[0075] Total loss value of joint training The calculation formula is: ,in For the preset weighting coefficients, satisfy An adaptive moment estimation optimizer is used to update the weight parameters of each network layer through gradient backpropagation. The parameters of the adaptive moment estimation optimizer are set to a preset learning rate, and the first-order moment decay coefficient and the second-order moment decay coefficient are taken as default values. During training, the joint training total loss value of the current epoch is calculated after each complete training epoch. Training is terminated when the decrease of the joint training total loss value in Q consecutive training epochs is less than a preset convergence threshold, or when the training epochs reach a preset maximum number of training epochs. The weight matrices, weight vectors, and bias term parameters of each layer at the time of training termination are serialized and saved to the model storage unit as a trained creep system dynamic response model for use in the online inference stage.

[0076] Step S310: Based on the progress depth parameter corresponding to the degree attribute dimension component in the creep process state transition characterization vector, generate the calibration rate adjustment coefficient in the dynamic calibration process of servo control parameters. The calibration rate adjustment coefficient is positively correlated with the progress depth parameter. When the progress depth parameter increases, the calibration rate adjustment coefficient increases synchronously.

[0077] Let the depth of advance parameter be p, and the reference rate coefficient be... The formula for calculating the calibration rate adjustment coefficient Kspd is as follows: Since p∈[0,1], Kspd increases linearly with p. When p approaches 1, Kspd reaches its maximum value. The increase in the calibration rate adjustment coefficient makes the rate at which the servo control parameters are adjusted from the current value to the target value faster. The deeper the creep stage progresses, the faster the deformation state of the specimen changes, and the servo system needs to complete the dynamic calibration of parameters with a higher response rate.

[0078] Step S320: Obtain the initial value of the current loop proportional gain and the initial value of the current loop integral gain of the servo current loop. Multiply the initial value of the current loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic current loop proportional gain. Multiply the initial value of the current loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic current loop integral gain. Write the dynamic current loop proportional gain and the dynamic current loop integral gain into the control register of the servo current loop.

[0079] Let the initial value of the current loop proportional gain be K_pi0, and the initial value of the current loop integral gain be K_ii0. The dynamic current loop proportional gain K_pi = K_pi0 × Kspd, and the dynamic current loop integral gain K_ii = K_ii0 × Kspd. K_pi and K_ii are written to the servo current loop control register via the fieldbus. The proportional gain and integral gain increase synchronously with Kspd, and the adjustment stiffness and error elimination rate of the current loop are correspondingly improved.

[0080] Step S330: Obtain the initial value of the speed loop proportional gain and the initial value of the speed loop integral gain of the servo speed loop; multiply the initial value of the speed loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic speed loop proportional gain; multiply the initial value of the speed loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic speed loop integral gain; and write the dynamic speed loop proportional gain and the dynamic speed loop integral gain into the control register of the servo speed loop.

[0081] Let the initial value of the speed loop proportional gain be K_pv0, and the initial value of the speed loop integral gain be K_iv0. The dynamic speed loop proportional gain K_pv = K_pv0 × Kspd, and the dynamic speed loop integral gain K_iv = K_iv0 × Kspd. Write K_pv and K_iv into the servo speed loop control register.

[0082] Step S340: Obtain the initial value of the force loop proportional gain and the initial value of the force loop integral gain of the servo force loop; multiply the initial value of the force loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic force loop proportional gain; multiply the initial value of the force loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic force loop integral gain; and write the dynamic force loop proportional gain and the dynamic force loop integral gain into the control register of the servo force loop.

[0083] Let the initial value of the force loop proportional gain be K_pf0, and the initial value of the force loop integral gain be K_if0. The dynamic force loop proportional gain K_pf = K_pf0 × Kspd, and the dynamic force loop integral gain K_if = K_if0 × Kspd. Write K_pf and K_if into the servo force loop control register.

[0084] Step S350: Generate feedforward compensation gain adjustment coefficient based on the advance depth parameter. The feedforward compensation gain adjustment coefficient is negatively correlated with the advance depth parameter. When the advance depth parameter increases, the feedforward compensation gain adjustment coefficient decreases synchronously.

[0085] The feedforward compensation gain adjustment coefficient Kff = 1 - p. As p increases, Kff decreases, and the gain of the feedforward compensation channel decreases as the creep stage progresses. In the later stages of creep, the specimen deformation behavior enters an accelerated irreversible phase, and excessive feedforward compensation may introduce overshoot. Therefore, it is necessary to gradually reduce the feedforward compensation intensity.

[0086] Step S360: Obtain a preset initial gain value for speed feedforward compensation; multiply the initial gain value for speed feedforward compensation by the feedforward compensation gain adjustment coefficient to obtain the dynamic speed feedforward compensation gain; and write the dynamic speed feedforward compensation gain into the feedforward compensation control channel of the servo speed loop. Obtain a preset initial gain value for force feedforward compensation; multiply the initial gain value for force feedforward compensation by the feedforward compensation gain adjustment coefficient to obtain the dynamic force feedforward compensation gain; and write the dynamic force feedforward compensation gain into the feedforward compensation control channel of the servo force loop.

[0087] Let the initial gain value of the speed feedforward compensation be Vff0, and the dynamic speed feedforward compensation gain be Vff = Vff0 × Kff. Let the initial gain value of the force feedforward compensation be Fff0, and the dynamic force feedforward compensation gain be Fff = Fff0 × Kff. Write Vff to the gain register of the servo speed loop feedforward compensation control channel, and write Fff to the gain register of the servo force loop feedforward compensation control channel.

[0088] Step S370: Generate the filter cutoff frequency adjustment coefficient based on the advance depth parameter. The filter cutoff frequency adjustment coefficient is negatively correlated with the advance depth parameter. When the advance depth parameter increases, the filter cutoff frequency adjustment coefficient decreases synchronously.

[0089] The filter cutoff frequency adjustment coefficient Kflt = 1 - p. As p increases, Kflt decreases, shifting the filter's cutoff frequency towards lower frequencies and increasing the filtering strength. In the later stages of the creep phase, high-frequency disturbances in the servo drive signal need to be suppressed more effectively to prevent the controller from malfunctioning due to noise.

[0090] Step S380: Obtain the preset initial cutoff frequency of the current loop filter in the servo current loop, the preset initial cutoff frequency of the speed loop filter in the servo speed loop, and the preset initial cutoff frequency of the force loop filter in the servo force loop. Multiply the initial cutoff frequency of the current loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic current loop filter cutoff frequency. Multiply the initial cutoff frequency of the speed loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic speed loop filter cutoff frequency. Multiply the initial cutoff frequency of the force loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic force loop filter cutoff frequency. Write the dynamic current loop filter cutoff frequency into the filter parameter register of the servo current loop, the dynamic speed loop filter cutoff frequency into the filter parameter register of the servo speed loop, and the dynamic force loop filter cutoff frequency into the filter parameter register of the servo force loop.

[0091] Let the initial cutoff frequency of the current loop filter be f_ci0, and the dynamic cutoff frequency of the current loop filter be f_ci = f_ci0 × Kflt. Let the initial cutoff frequency of the speed loop filter be f_cv0, and the dynamic cutoff frequency of the speed loop filter be f_cv = f_cv0 × Kflt. Let the initial cutoff frequency of the force loop filter be f_cf0, and the dynamic cutoff frequency of the force loop filter be f_cf = f_cf0 × Kflt. Write f_ci into the filter parameter register of the servo current loop, f_cv into the filter parameter register of the servo speed loop, and f_cf into the filter parameter register of the servo force loop.

[0092] Step S410: Obtain the lead screw displacement timing signal, perform displacement change rate extraction processing on the lead screw displacement timing signal, calculate the ratio of displacement change to time interval at adjacent sampling times, and generate the lead screw movement speed timing signal.

[0093] Let the value of the lead screw displacement timing signal at the kth sampling time be... The sampling period is Lead screw movement rate The timing signal Vs of the lead screw movement speed is obtained by iterating through k from 1 to N-1.

[0094] Step S420: Obtain the timing signal of the axial force of the lead screw and the timing signal of the lead screw movement speed. Combine the timing signal of the axial force of the lead screw and the timing signal of the lead screw movement speed on a unified time axis to generate a force-velocity phase plane trajectory diagram. The force-velocity phase plane trajectory diagram uses the axial force of the lead screw as the first dimension coordinate and the lead screw movement speed as the second dimension coordinate.

[0095] Let the value of the lead screw axial force timing signal at the k-th sampling time be Xfk. Pair the values ​​corresponding to each sampling time with (Xfk, ... As a phase point on a two-dimensional phase plane, adjacent phase points are connected by straight line segments according to the sampling time sequence to form a force-velocity phase plane trajectory diagram. The horizontal axis of the force-velocity phase plane trajectory diagram represents the axial force of the lead screw, in Newtons, and the vertical axis represents the lead screw movement speed, in millimeters per second.

[0096] Step S430: Extract graphic morphological features from the force-velocity phase plane trajectory diagram, identify the closed loop area features, trajectory spiral direction features, and trajectory boundary offset distance features in the force-velocity phase plane trajectory diagram, and generate a phase plane trajectory morphological feature set.

[0097] The closed loop area feature S_loop is extracted by calculating the area of ​​the phase point sequence forming the closed loop in the force-velocity phase plane trajectory diagram using the polygon area formula.

[0098] The process involves summing and iterating through all phase point pairs on the closed loop. The trajectory spiral direction feature D_spiral is extracted by calculating the directed area of ​​the triangle formed by three consecutive phase points in the force-velocity phase plane trajectory diagram. A positive sign indicates a counter-clockwise spiral direction, and a negative sign indicates a clockwise spiral direction. The variation of the positive and negative signs of the directed area across the entire trajectory is statistically analyzed as the spiral direction feature. The trajectory boundary offset distance feature D_bound is extracted by calculating the convex hull polygon of the force-velocity phase plane trajectory diagram, calculating the perpendicular distances from all phase points to each edge of the convex hull, and taking the arithmetic mean and maximum value of these distances as the boundary offset distance feature. S_loop, D_spiral, and D_bound are combined into the phase plane trajectory morphology feature set F_phase.

[0099] Step S440: Extract the stage attribute dimension component from the creep process state transition representation vector, determine the current creep stage type based on the stage attribute dimension component, extract the degree attribute dimension component from the creep process state transition representation vector, and determine the current progress depth parameter based on the degree attribute dimension component.

[0100] The stage attribute dimension component takes the stage index corresponding to the maximum value. Index value 0 corresponds to the transient creep stage, index value 1 corresponds to the steady-state creep stage, and index value 2 corresponds to the accelerated creep stage. The degree attribute dimension component is directly used as the current progress depth parameter p.

[0101] Step S450: Obtain the phase plane trajectory morphology feature record corresponding to the accelerated creep stage in the pre-stored historical creep failure cases. The phase plane trajectory morphology feature record of the historical creep failure cases includes the closed loop area feature sequence before failure, the trajectory spiral direction feature sequence before failure, and the trajectory boundary offset distance feature sequence before failure.

[0102] Using the material grade and test temperature of the current specimen as joint search criteria, matching records are retrieved from the historical creep failure case database. The retrieved records contain a sequence of phase plane trajectory morphological features at each sampling time before failure, arranged in chronological order, with the last record corresponding to the feature value at the last measurable time before failure.

[0103] Step S460: The closed loop area feature, trajectory spiral direction feature, and trajectory boundary offset distance feature in the phase plane trajectory morphology feature set are compared with the closed loop area feature sequence, trajectory spiral direction feature sequence, and trajectory boundary offset distance feature sequence before failure to obtain the similarity matching result. The similarity matching result is used to characterize the degree of similarity between the current phase plane trajectory morphology and the historical phase plane trajectory morphology before failure.

[0104] For the closed loop area feature sequence, a dynamic time warping algorithm is used to calculate the warped path distance. The smaller the warped path distance, the higher the similarity. The closed loop area similarity is denoted as S_sim. For the trajectory spiral direction feature, a direct comparison method is used. If the current spiral direction is consistent with the spiral direction before failure, the similarity is set to 1; otherwise, it is set to 0. The spiral direction similarity is denoted as D_sim. For the trajectory boundary offset distance feature, the difference between the current offset distance and the average offset distance of the last M records before failure is calculated. The smaller the difference, the higher the similarity. The boundary offset similarity is denoted as B_sim. The total similarity matching result R_sim = (S_sim + D_sim + B_sim) / 3.

[0105] Step S470: When the current creep stage type is steady-state creep stage and the similarity matching result exceeds the preset similarity threshold, an accelerated creep warning trigger instruction is generated. The accelerated creep warning trigger instruction includes the comparison difference between the current closed loop area feature and the last record in the closed loop area feature sequence before failure, the comparison difference between the current trajectory spiral direction feature and the last record in the trajectory spiral direction feature sequence before failure, and the current progress depth parameter.

[0106] Let the preset similarity threshold be R_th. If the current stage type is steady-state creep stage and R_sim > R_th, then an accelerated creep warning trigger command is generated. The comparison difference ΔS = S_loop - S_loop_his_last, and the comparison difference ΔD = D_spiral - D_spiral_his_last. The data structure of the accelerated creep warning trigger command is a triple (ΔS, ΔD, p).

[0107] Step S480: Perform pre-shrinkage processing on the adjusted constraint boundary according to the accelerated creep warning trigger command. The pre-shrinkage processing reduces the values ​​of the upper limit threshold of motor current and the upper limit threshold of motor speed in advance on the basis of the original constraint boundary shrinkage adjustment.

[0108] Let the current constraint boundary generated in step S144 be IB, and the rotational speed constraint boundary be VB. The pre-shrinkage coefficient β_pre = 1 - μ × (ΔS + ΔD), where μ is the pre-shrinkage scaling factor, and β_pre < 1. The pre-shrinked current constraint boundary IB_pre = IB × β_pre, and the pre-shrinked rotational speed constraint boundary VB_pre = VB × β_pre. IB_pre and VB_pre replace the original IB and VB for deviation calculation in subsequent step S145. The pre-shrinkage process allows for the application of stricter servo parameter constraints in advance when a phase plane trajectory pattern similar to historical failure cases is detected, preventing rapid failure during the accelerated creep stage of the specimen.

[0109] Step S510: Obtain non-contact deformation measurement signals within the gauge length section of the specimen during the creep test performed by the intelligent creep machine. The non-contact deformation measurement signals include the axial strain timing signal and the radial strain timing signal of the gauge length section. The axial strain timing signal of the gauge length section is acquired by a non-contact deformation measurement unit installed in the axial direction of the gauge length section of the specimen, and the radial strain timing signal of the gauge length section is acquired by a non-contact deformation measurement unit installed in the radial direction of the gauge length section of the specimen.

[0110] The non-contact deformation measurement unit consists of dual cameras in a digital image correlation measurement system. The dual cameras simultaneously acquire speckle images of the gauge length surface of the specimen, and the axial strain and radial strain of the gauge length are calculated using an image matching algorithm. The axial strain time series signal of the gauge length is denoted as ε_a, and the radial strain time series signal is denoted as ε_r, both of which are dimensionless engineering strain values.

[0111] Step S520: Perform strain rate extraction processing on the axial strain time series signal of the gauge length segment, calculate the ratio of the axial strain change to the time interval at adjacent sampling times, and generate the axial strain rate time series signal. Perform strain rate extraction processing on the radial strain time series signal of the gauge length segment, calculate the ratio of the radial strain change to the time interval at adjacent sampling times, and generate the radial strain rate time series signal.

[0112] Axial strain rate Radial strain rate Both V_a and V_r take absolute values.

[0113] Step S530: Based on the axial strain rate time-series signal and the radial strain rate time-series signal, calculate the ratio of radial strain rate to axial strain rate at each sampling moment to generate a strain rate ratio time-series signal.

[0114] The strain rate ratio R_b_k = V_r_k / V_a_k, and if V_a_k = 0, then R_b_k = 0.

[0115] Step S540: Perform trend extraction processing on the strain rate ratio time series signal, extract the monotonically changing trend segments and trend inflection point positions in the strain rate ratio time series signal, and generate a strain rate ratio change trend descriptor. The strain rate ratio change trend descriptor records the start and end times and the increasing slope parameter of the monotonically increasing segment, the start and end times and the decreasing slope parameter of the monotonically decreasing segment, and the time position of the trend inflection point.

[0116] Piecewise linear fitting is performed on the strain rate ratio time-series signal R_b. Starting from the first sampling point, a sliding window is used to scan the R_b sequence segment by segment. The data within the window are subjected to least-squares linear fitting to obtain the fitting slope. When the fitting slope is greater than zero, the segment is marked as a monotonically increasing segment, and the start time index t_s_inc, end time index t_e_inc, and increasing slope parameter k_inc = (R_b(t_e_inc) - R_b(t_s_inc)) / (t_e_inc - t_s_inc) are recorded. When the fitting slope is less than zero, the segment is marked as a monotonically decreasing segment, and the start time index t_s_dec, end time index t_e_dec, and decreasing slope parameter k_dec = (R_b(t_e_dec) - R_b(t_s_dec)) / (t_e_dec - t_s_dec) are recorded. When the sign of the slope changes between adjacent segments, the time index at the point of change is marked as the trend turning point t_turn. The parameters of each increasing segment, each decreasing segment, and the positions of all trend inflection points are combined into a strain rate ratio change trend descriptor D_trend.

[0117] Step S550: Extract the stage attribute dimension component from the creep process state transition representation vector, determine the current creep stage type based on the stage attribute dimension component, extract the degree attribute dimension component from the creep process state transition representation vector, and determine the current progress depth parameter based on the degree attribute dimension component.

[0118] The first three dimensions of the vector are extracted from Vcreep as the stage attribute dimension components. The index value corresponding to the maximum value is taken. Index value 0 corresponds to the transient creep stage, index value 1 corresponds to the steady-state creep stage, and index value 2 corresponds to the accelerated creep stage. The seventh scalar dimension is extracted from Vcreep as the progress depth parameter p, p∈[0,1].

[0119] Step S560: Obtain the phase plane trajectory morphology feature record corresponding to the accelerated creep stage in the pre-stored historical creep failure cases. The phase plane trajectory morphology feature record of the historical creep failure cases includes the closed loop area feature sequence before failure, the trajectory spiral direction feature sequence before failure, and the trajectory boundary offset distance feature sequence before failure.

[0120] Using the material grade and test temperature of the current specimen as joint search criteria, matching records are retrieved from the historical creep failure case database. The retrieved records contain a sequence of phase plane trajectory morphological features at each sampling time before failure, arranged chronologically, with the last record corresponding to the feature value at the last measurable time before failure. The closed loop area feature sequence before failure is denoted as {S_loop_his_q}, the trajectory spiral direction feature sequence before failure is denoted as {D_spiral_his_q}, and the trajectory boundary offset distance feature sequence before failure is denoted as {D_bound_his_q}, where q is the sequence index.

[0121] Step S570: The closed loop area feature, trajectory spiral direction feature, and trajectory boundary offset distance feature in the phase plane trajectory morphology feature set are compared with the closed loop area feature sequence, trajectory spiral direction feature sequence, and trajectory boundary offset distance feature sequence before failure. The similarity matching result is used to characterize the degree of similarity between the current phase plane trajectory morphology and the historical phase plane trajectory morphology before failure.

[0122] For the closed loop area feature, the current closed loop area S_loop is dynamically time-warped and matched with the closed loop area feature sequence {S_loop_his_q} before failure. The warped path distance DTW_loop is calculated. The smaller the warped path distance, the higher the similarity. The closed loop area similarity S_sim = 1 / (1+DTW_loop). For the trajectory spiral direction feature, if the current spiral direction D_spiral is consistent with the spiral direction of the last record in the trajectory spiral direction feature sequence before failure, then D_sim = 1; otherwise, D_sim = 0. For the trajectory boundary offset distance feature, the absolute value of the difference between the current boundary offset distance D_bound and the mean D_bound_his_avg of the last M records in the trajectory boundary offset distance feature sequence before failure is calculated. The boundary offset similarity B_sim = 1 / (1+D_bound-D_bound_his_avg). The total similarity matching result R_sim=(S_sim+D_sim+B_sim) / 3.

[0123] Step S580: When the current creep stage type is steady-state creep stage and the similarity matching result exceeds the preset similarity threshold, an accelerated creep warning trigger instruction is generated. The accelerated creep warning trigger instruction includes the comparison difference between the current closed loop area feature and the last record in the closed loop area feature sequence before failure, the comparison difference between the current trajectory spiral direction feature and the last record in the trajectory spiral direction feature sequence before failure, and the current progress depth parameter.

[0124] Let the preset similarity threshold be R_th. If the current stage type is a steady-state creep stage and R_sim is greater than R_th, then an accelerated creep warning trigger command is generated. The comparison difference ΔS is equal to S_loop minus the last record value in {S_loop_his_q}. The comparison difference ΔD is equal to D_spiral minus the numerical representation of the last record value in {D_spiral_his_q}. The data structure of the accelerated creep warning trigger command is a triple (ΔS, ΔD, p).

[0125] Step S590: Perform pre-shrinkage processing on the adjusted constraint boundary according to the accelerated creep warning trigger command. The pre-shrinkage processing reduces the values ​​of the upper limit threshold of motor current and the upper limit threshold of motor speed in advance on the basis of the original constraint boundary shrinkage adjustment.

[0126] Let the current constraint boundary generated in step S144 be IB, and the speed constraint boundary be VB. The formula for calculating the pre-shrinkage coefficient β_pre is β_pre = 1 - μ × (ΔS + ΔD), where μ is the pre-shrinkage scaling factor, and β_pre is less than 1. The pre-shrinked current constraint boundary IB_pre = IB × β_pre. The pre-shrinked speed constraint boundary VB_pre = VB × β_pre. Replace the original IB and VB with IB_pre and VB_pre for the deviation calculation in the subsequent step S145.

[0127] Step S610: Obtain non-contact deformation measurement signals within the gauge length section of the specimen during the creep test performed by the intelligent creep machine. The non-contact deformation measurement signals include the axial strain timing signal and the radial strain timing signal of the gauge length section. The axial strain timing signal of the gauge length section is acquired by a non-contact deformation measurement unit installed in the axial direction of the gauge length section of the specimen, and the radial strain timing signal of the gauge length section is acquired by a non-contact deformation measurement unit installed in the radial direction of the gauge length section of the specimen.

[0128] The non-contact deformation measurement unit consists of dual cameras in a digital image correlation measurement system. The dual cameras simultaneously acquire speckle images of the gauge length surface of the specimen, and the axial strain and radial strain of the gauge length are calculated using an image matching algorithm. The axial strain time series signal of the gauge length is denoted as ε_a, and the radial strain time series signal is denoted as ε_r, both of which are dimensionless engineering strain values.

[0129] Step S620: Perform strain rate extraction processing on the axial strain time series signal of the gauge length segment, calculate the ratio of the axial strain change to the time interval at adjacent sampling times, and generate the axial strain rate time series signal. Perform strain rate extraction processing on the radial strain time series signal of the gauge length segment, calculate the ratio of the radial strain change to the time interval at adjacent sampling times, and generate the radial strain rate time series signal.

[0130] Axial strain rate Radial strain rate Both V_a and V_r are absolute values, expressed in seconds.

[0131] Step S630: Based on the axial strain rate time-series signal and the radial strain rate time-series signal, calculate the ratio of radial strain rate to axial strain rate at each sampling moment to generate a strain rate ratio time-series signal.

[0132] The strain rate ratio R_b_k = V_r_k / V_a_k, and if V_a_k = 0, then R_b_k = 0. R_b is a dimensionless ratio sequence.

[0133] Step S640: Perform trend extraction processing on the strain rate ratio time series signal, extract the monotonically changing trend segments and trend inflection point positions in the strain rate ratio time series signal, and generate a strain rate ratio change trend descriptor. The strain rate ratio change trend descriptor records the start and end times and the increasing slope parameter of the monotonically increasing segment, the start and end times and the decreasing slope parameter of the monotonically decreasing segment, and the time position of the trend inflection point.

[0134] A sliding window is used to scan the R_b sequence segment by segment. Least-squares linear fitting is performed on the data within the window to obtain the fitted slope. When the fitted slope is greater than zero, the segment is marked as a monotonically increasing segment, and the start time index t_s_inc, end time index t_e_inc, and increasing slope parameter k_inc = (R_b(t_e_inc) - R_b(t_s_inc)) / (t_e_inc - t_s_inc) are recorded. When the fitted slope is less than zero, the segment is marked as a monotonically decreasing segment, and the start time index t_s_dec, end time index t_e_dec, and decreasing slope parameter k_dec = (R_b(t_e_dec) - R_b(t_s_dec)) / (t_e_dec - t_s_dec) are recorded. When the sign of the slope changes between adjacent segments, the time index at the point of change is marked as the trend turning point t_turn. The parameters of each increasing segment, each decreasing segment, and the positions of all trend inflection points are combined into a strain rate ratio change trend descriptor D_trend.

[0135] Step S650: Extract the stage attribute dimension component from the creep process state transition characterization vector, determine the current creep stage type based on the stage attribute dimension component, and when the current creep stage type is a steady-state creep stage, read the time position of the most recent trend inflection point from the strain rate ratio change trend descriptor.

[0136] The first three dimensions of the vector are extracted from Vcreep as the stage attribute dimension components. The index value corresponding to the maximum value is taken. Index value 0 corresponds to the transient creep stage, index value 1 corresponds to the steady-state creep stage, and index value 2 corresponds to the accelerated creep stage. If it is determined to be a steady-state creep stage, the position Tb_turn of the trend inflection point with the largest time index is read from D_trend.

[0137] Step S660: Calculate the time difference between the most recent trend inflection point and the starting time of the steady-state creep stage to obtain the duration of the steady-state creep stage. Compare the duration of the steady-state creep stage with the historical duration of the steady-state creep stage of similar materials under the same test conditions in the pre-stored data to generate an estimated remaining duration of the steady-state creep stage.

[0138] Let the starting time of the steady-state creep stage be... The duration of the steady-state creep phase, ΔT_ss, is equal to Tb_turn minus... The duration of the steady-state creep stage history of pre-stored similar materials under the same experimental conditions is L_his. The estimated remaining duration L_rem is equal to L_his minus ΔT_ss.

[0139] Step S670: Generate a time margin parameter based on the estimated remaining duration of the steady-state creep stage. The time margin parameter indicates the remaining available time before entering the accelerated creep stage. Write the time margin parameter into the control parameter optimization instruction, adding time margin prompt information to the control parameter optimization instruction. When the current creep stage type is the accelerated creep stage, perform acceleration trend analysis processing on the axial strain rate time series signal, extract the accelerated growth slope parameter of the axial strain rate, generate an estimated termination time of the accelerated creep stage based on the accelerated growth slope parameter and the degree attribute dimension component, and write the estimated termination time of the accelerated creep stage into the control parameter optimization instruction, adding termination time prompt information to the control parameter optimization instruction.

[0140] The time margin parameter is set to L_rem, and L_rem is written into the time margin field of the control parameter optimization instruction load segment.

[0141] If the current stage is the accelerated creep stage, perform linear fitting on the V_a sequence to extract the accelerated growth slope parameter k_acc. Let the current axial strain value be ε_a_curr, and the preset termination strain threshold for the accelerated creep stage be ε_a_fail. Estimated termination time for the accelerated creep stage. Write T_fail into the termination time field of the control parameter optimization instruction payload segment.

[0142] Step S710: Obtain the servo motor current timing signal and the servo motor voltage timing signal, perform power calculation processing on the servo motor current timing signal and the servo motor voltage timing signal, multiply the servo motor current value and the servo motor voltage value at the same sampling time to generate the real-time power timing signal of the servo motor.

[0143] Let the value of the servo motor current timing signal at the k-th sampling moment be I_k, and the value of the servo motor voltage timing signal at the k-th sampling moment be U_k. P_elec_k = I_k × U_k, where P_elec is the real-time power timing signal of the servo motor, in watts.

[0144] Step S720: Perform power fluctuation analysis on the real-time power timing signal of the servo motor, extract the power fluctuation amplitude characteristics and power fluctuation frequency characteristics of the real-time power timing signal of the servo motor within a preset time window, and generate a power fluctuation feature descriptor.

[0145] Let the preset time window length be W_p. The power fluctuation amplitude feature A_fluct is the standard deviation of the P_elec sequence within the window. The power fluctuation frequency feature F_fluct is the frequency value corresponding to the maximum value in the amplitude spectrum of the P_elec sequence within the window after Fast Fourier Transform. Combine A_fluct and F_fluct into a power fluctuation feature descriptor D_fluct.

[0146] Step S730: Obtain the timing signal of the axial force of the lead screw and the timing signal of the lead screw displacement. Perform work calculation processing on the timing signal of the axial force of the lead screw and the timing signal of the lead screw displacement. Multiply the average value of the axial force of the lead screw at adjacent sampling times with the change in the lead screw displacement and perform time integration to generate the timing signal of the cumulative work done by the lead screw.

[0147] Let the axial force of the leadscrew at the k-th sampling time be F_k, and the displacement of the leadscrew at the k-th sampling time be S_k. The work increment at the k-th sampling time is ΔW_k = (F_k + F_{k+1}) / 2 × (S_{k+1} - S_k). The cumulative work done by the leadscrew is... , i ranges from 1 to k-1.

[0148] Step S740: Perform work rate extraction processing on the cumulative work of the lead screw timing signal, calculate the ratio of the change in cumulative work at adjacent sampling times to the time interval, and generate the lead screw work rate timing signal.

[0149] P_mech is the timing signal for the working rate of the leadscrew, in watts.

[0150] Step S750: Compare the real-time power timing signal of the servo motor with the working speed timing signal of the lead screw on the same time axis, calculate the difference between the real-time power value of the servo motor and the working speed value of the lead screw at each sampling moment, and generate the transmission efficiency loss power timing signal.

[0151] P_loss_k = P_elec_k - P_mech_k, where P_loss is the timing signal of transmission efficiency loss power, in watts.

[0152] Step S760: Perform cumulative analysis processing on the transmission efficiency loss power time sequence signal, and perform time integration on the transmission efficiency loss power value within the preset time window to generate the transmission efficiency loss cumulative energy time sequence signal.

[0153] Let the preset time window length be W_loss. k is summed over the window length, and E_loss is the cumulative energy of transmission efficiency loss in joules.

[0154] Step S770: Extract the stage attribute dimension component and the degree attribute dimension component from the creep process state transition characterization vector, and input the stage attribute dimension component, the degree attribute dimension component, the power fluctuation feature descriptor, and the transmission efficiency loss cumulative energy time series signal into the pre-built servo drive health assessment model to generate the servo drive health assessment value.

[0155] The servo drive health assessment model is a three-layer fully connected neural network. The input layer has 7 dimensions: 3 for stage attribute components, 1 for degree attribute components, 2 for power fluctuation feature descriptors, and 1 for transmission efficiency loss cumulative energy scalar. The first hidden layer has 16 nodes, and the activation function is a linear rectified function. The second hidden layer has 8 nodes, and the activation function is also a linear rectified function. The output layer has 1 node, and the activation function is a logistic function, outputting the servo drive health assessment value H_health ∈ [0, 1].

[0156] Step S780: Generate a transmission system maintenance prompt instruction based on the servo drive health assessment value. When the servo drive health assessment value is lower than the preset health threshold, attach the transmission system maintenance prompt instruction to the control parameter optimization instruction.

[0157] If H_health is less than the preset health threshold H_th, a transmission system maintenance prompt instruction is generated. The instruction payload includes the H_health value, A_fluct, and E_loss. This transmission system maintenance prompt instruction is then appended as an extended field to the control parameter optimization instruction.

[0158] For example, the method may further include: step S810, acquiring the servo motor winding temperature time sequence signal, performing temperature change trend extraction processing on the servo motor winding temperature time sequence signal, extracting the temperature rise slope sequence and temperature fluctuation amplitude sequence of the servo motor winding temperature time sequence signal within a continuous time window, and generating a set of temperature dynamic change feature descriptions.

[0159] Let the length of the continuous time window be W_temp. A linear fit is performed on the temperature sequence within the window to obtain the temperature rise slope k_T. The temperature fluctuation amplitude A_T is the standard deviation of the temperature sequence within the window. k_T and A_T are combined to form the temperature dynamic change feature description set D_temp.

[0160] Step S820: Obtain the time sequence signal of the axial force of the lead screw, perform force drift analysis on the time sequence signal of the axial force of the lead screw, extract the force attenuation slope sequence and the force fluctuation standard deviation sequence of the axial force time sequence signal during the constant displacement holding stage, and generate a set of force relaxation feature descriptions.

[0161] During the constant displacement holding phase, the force value sequence is linearly fitted to obtain the force value attenuation slope k_F. The standard deviation of force value fluctuation σ_F is the standard deviation of the force value sequence. k_F and σ_F are combined to form the force value relaxation feature description set D_force.

[0162] Step S830: Extract the degree attribute dimension component from the creep process state transition characterization vector, determine the current progress depth parameter based on the degree attribute dimension component, and perform coupled analysis on the current temperature rise slope in the temperature rise slope sequence and the current progress depth parameter to establish a mapping relationship between the temperature rise slope and the creep progress depth.

[0163] The 7th scalar dimension is extracted from Vcreep as the progress depth parameter p. k_T is coupled with p to establish a mapping function k_T=f_T(p). The function parameters are determined by linear regression using the k_T values ​​corresponding to different p values ​​in historical data.

[0164] Step S840: The current force attenuation slope in the force attenuation slope sequence is coupled with the current creep depth parameter for analysis to establish a mapping relationship between the force attenuation slope and the creep creep depth.

[0165] Couple k_F with p to establish a mapping function k_F=f_F(p), and determine the function parameters by performing linear regression on the k_F values ​​corresponding to different p values ​​in historical data.

[0166] Step S850: Based on the mapping relationship between the temperature rise slope and the creep progression depth, determine whether the change in the current temperature rise slope relative to the temperature rise slope corresponding to the previous creep progression depth exceeds the preset slope change judgment threshold, and generate a temperature slope change judgment result; based on the mapping relationship between the force attenuation slope and the creep progression depth, determine whether the change in the current force attenuation slope relative to the stress attenuation slope corresponding to the previous creep progression depth exceeds the preset slope change judgment threshold, and generate a force slope change judgment result.

[0167] Calculate the temperature slope change magnitude Δk_T = k_T(p_current) - k_T(p_prev) / k_T(p_prev). If Δk_T is greater than the preset slope abrupt change threshold Δ_th, the temperature slope abrupt change is determined to have occurred. Calculate the force slope change magnitude Δk_F = k_F(p_current) - k_F(p_prev). / k_F(p_prev). If Δk_F is greater than Δ_th, then the result of the force slope abrupt change is that a sudden change has occurred.

[0168] Step S860: When the temperature slope change determination result indicates that a slope change has occurred and the force value slope change determination result indicates that a slope change has occurred, a creep mechanism conversion warning sign is generated. The creep mechanism conversion warning sign is used to indicate that the micro-damage mechanism inside the specimen material in the intelligent creep machine changes from grain boundary slip dominance to pore nucleation and growth dominance.

[0169] If Δk_T is greater than Δ_th and Δk_F is greater than Δ_th, then a creep mechanism transition warning flag Flag_trans=1 is generated. This creep mechanism transition warning flag indicates that the servo system is currently undergoing a shift in the dominant creep deformation mechanism, and the control strategy needs to be adjusted to cope with the accelerated accumulation of microscopic damage in the material.

[0170] Step S870: The creep mechanism conversion early warning identifier is attached to the creep process state transition representation vector to generate an enhanced creep process state transition representation vector containing mechanism conversion early warning information. The enhanced creep process state transition representation vector adds a mechanism early warning attribute dimension component on the basis of the original stage attribute dimension component, direction attribute dimension component and degree attribute dimension component.

[0171] Add Flag_trans as the 8th dimension to Vcreep to obtain the enhanced creep process state transition representation vector Vcreep_enh=[Vcreep;Flag_trans], which has an 8-dimensional dimension.

[0172] Step S880: Based on the mechanism warning attribute dimension components, perform mechanism conversion additional constraint adjustment on the adjustment constraint boundary of the servo control parameters under the current creep stage. The mechanism conversion additional constraint adjustment further corrects the original upper limit threshold of motor current and upper limit threshold of motor speed downward. The correction magnitude is positively correlated with the change magnitude of temperature rise slope and the change magnitude of force attenuation slope.

[0173] Let the current constraint boundary generated in step S144 be IB, and the speed constraint boundary be VB. The correction amplitude factor γ = 1 - λ × (Δk_T + Δk_F) / 2, where λ is the correction scaling factor, and γ is less than 1. The corrected current constraint boundary IB_trans = IB × γ. The corrected speed constraint boundary VB_trans = VB × γ. IB_trans and VB_trans replace the original IB and VB for subsequent dynamic calibration of servo control parameters.

[0174] Step S890: Based on the adjusted constraint boundary after the additional constraint adjustment of the execution mechanism transformation, generate the control parameter optimization instruction of the additional mechanism transformation response strategy.

[0175] Using IB_trans and VB_trans as new constraint boundaries, the current, velocity, and force setpoint adjustments are recalculated according to steps S145 to S147, generating control parameter optimization instructions for the additional mechanism transition response strategy. A mechanism transition response identifier field is added to the instruction load, recording the Flag_trans value and the correction amplitude factor γ value.

[0176] Step S910: Obtain the lead screw displacement timing signal, perform displacement holding characteristic analysis on the lead screw displacement timing signal, extract the deviation sequence and deviation pullback compensation sequence of the lead screw displacement deviating from the target displacement value during the holding stage, and generate a displacement holding accuracy feature description set.

[0177] Let the target displacement be S_target. The displacement deviation ΔS_k = S_k - S_target. The deviation pullback compensation is the displacement compensation output by the controller. The ΔS_k sequence and the compensation sequence are combined into a displacement holding accuracy feature description set D_pos.

[0178] Step S920: Obtain the force value maintenance characteristics of the lead screw axial force timing signal during the holding phase, extract the force value tracking error sequence and the force value compensation response delay time sequence of the lead screw axial force timing signal during the holding phase, and generate a set of force value maintenance accuracy feature descriptions.

[0179] Let the target force be F_target. The force tracking error ΔF_k = F_k - F_target. The force compensation response delay time is the time interval from the occurrence of the force deviation to the effective application of the compensation. The ΔF_k sequence and the delay time sequence are combined into a force maintenance accuracy feature description set D_force_maint.

[0180] Step S930: Extract the displacement deviation sequence from the displacement holding accuracy feature description set, perform deviation direction consistency analysis on the displacement deviation sequence, and statistically analyze the duration sequence and deviation direction switching frequency sequence of displacement deviations maintaining the same deviation direction in continuous sampling time to generate displacement deviation direction adhesion feature description.

[0181] Record the sign sequence of ΔS_k. Calculate the duration T_stick of consecutive identical signs and the number of sign switches N_switch. Describe the displacement deviation direction adhesion characteristics D_stick, which includes the average duration and switching frequency N_switch / T_total, where T_total is the total duration of the hold phase.

[0182] Step S940: Extract the force tracking error sequence from the force value maintenance accuracy feature description set, perform error accumulation trend analysis on the force tracking error sequence, calculate the error accumulation area sequence and error zero crossing number sequence of the force tracking error sequence within a continuous time window, and generate a force error accumulation feature description.

[0183] The cumulative area of ​​error A_err=∑ΔF_k The number of zero-crossing errors, N_zero, is the number of sign changes between adjacent sampling points in the ΔF_k sequence. The force error accumulation feature F_err_accum includes A_err and N_zero.

[0184] In step S950, the adhesion feature description of displacement deviation direction and the cumulative feature description of force error are input into the pre-constructed micro-motion crawling behavior discrimination model. The micro-motion crawling behavior discrimination model outputs the probability value of micro-motion crawling behavior based on the joint distribution characteristics between the switching frequency of displacement deviation direction and the number of zero crossings of force tracking error.

[0185] The micro-movement crawling behavior discrimination model is a support vector machine classifier with a radial basis function kernel. The input feature vector is [T_stick_mean, N_switch / T_total, A_err, N_zero]. The output is the probability value of micro-movement crawling behavior P_stick∈[0,1].

[0186] Step S960: Extract the stage attribute dimension component from the creep process state transition characterization vector. When the stage attribute dimension component indicates that the current creep stage is a steady-state creep stage, extract the degree attribute dimension component from the creep process state transition characterization vector. Compare the probability value of the micro-creep behavior with the preset creep probability judgment threshold. When the probability value of the micro-creep behavior exceeds the preset creep probability judgment threshold, generate the micro-creep working condition identification result. Based on the micro-creep working condition identification result and the degree attribute dimension component, generate an anti-creep compensation control strategy. The anti-creep compensation control strategy includes a displacement deviation feedforward compensation sequence and a force value flutter suppression filter parameter. The displacement deviation feedforward compensation sequence is determined based on the deviation pullback compensation sequence and the deviation direction duration in the displacement deviation direction adhesion feature description. The force value flutter suppression filter parameter is determined based on the error fluctuation frequency characteristics of the force value tracking error sequence.

[0187] If P_stick is greater than the preset crawling probability judgment threshold P_th, then a micro-motion crawling condition identification result Flag_stick=1 is generated. The displacement deviation feedforward compensation sequence is equal to the deviation pullback compensation sequence multiplied by the compensation gain coefficient, and the compensation gain coefficient is proportional to the duration T_stick_mean of the deviation direction. The force flutter suppression filter parameters are the center frequency and bandwidth of the notch filter, and the center frequency is determined by the peak frequency in the amplitude spectrum after the force tracking error sequence is subjected to a fast Fourier transform.

[0188] Step S970: The displacement deviation feedforward compensation sequence is superimposed on the position setpoint signal of the servo position loop to generate a corrected position setpoint signal containing anti-creep feedforward compensation. The force value chatter suppression filter parameters are written into the notch filter parameter register of the servo force loop to update the filtering characteristics of the servo force loop.

[0189] The corrected position setpoint signal is equal to the original position setpoint signal plus the displacement deviation feedforward compensation sequence. The notch filter parameters are written into the filter parameter register of the servo force loop.

[0190] Step S980: Based on the execution effect of the anti-crawling compensation control strategy, continuously monitor the probability value of micro-motion crawling behavior. When the probability value of micro-motion crawling behavior decreases to below the preset crawling release threshold after the execution of the anti-crawling compensation control strategy, generate an anti-crawling compensation effective confirmation flag.

[0191] After the anti-crawling compensation control strategy is executed, P_stick_new is recalculated. If P_stick_new is less than the preset crawling release threshold P_rel, an anti-crawling compensation activation confirmation flag Flag_rel=1 is generated.

[0192] Step S990: Combine the micro-creep condition identification result, anti-creep compensation control strategy, and anti-creep compensation effective confirmation flag into a micro-creep response record, and attach the micro-creep response record to the control parameter optimization instruction.

[0193] The micro-motion crawling response record data structure includes Flag_stick, displacement deviation feedforward compensation sequence, force value flutter suppression filter parameters, and Flag_rel. This record is appended as an extended field to the control parameter optimization instruction.

[0194] In one exemplary embodiment, a servo system control optimization system for an intelligent creep machine is provided. This servo system control optimization system for the intelligent creep machine can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, the servo system control optimization system for an intelligent creep machine includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a servo system control optimization method for an intelligent creep machine. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the housing of the servo system control optimization system used in intelligent creep machines, or an external keyboard, touchpad, or mouse, etc.

[0195] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A servo system control optimization method applied to an intelligent creep machine, characterized in that, The method includes: The set of servo drive timing signals of the intelligent creep machine during the continuous loading period is obtained. The set of servo drive timing signals includes servo motor winding temperature timing signal, servo motor current timing signal, servo motor speed timing signal, lead screw axial force timing signal, and lead screw displacement timing signal. The servo drive timing signal set is subjected to multi-resolution signal decomposition processing to obtain the approximate component sequence and detail component sequence of each signal component on the multi-order decomposition layer, and a time-frequency localization feature set reflecting the signal change characteristics is generated based on the approximate component sequence and the detail component sequence. The pre-constructed dynamic response model of the creep system is invoked to perform creep state correlation analysis on the time-frequency localized feature set, generating a creep process state transition characterization vector. The creep process state transition characterization vector is used to characterize the evolution direction and degree of the material in the intelligent creep machine from the transient creep stage to the steady-state creep stage and then to the accelerated creep stage under continuous load. Based on the creep process state transition characterization vector, the adjustment constraint boundary of the servo control parameters under the current creep stage is determined, and based on the adjustment constraint boundary, the control parameter optimization instruction for dynamically calibrating the servo control parameters is generated.

2. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The process of performing multi-resolution signal decomposition on the servo drive timing signal set to obtain approximate component sequences and detail component sequences of each signal component at multiple decomposition layers, and generating a time-frequency localized feature set reflecting the signal abrupt change characteristics based on the approximate component sequences and the detail component sequences, includes: The servo motor winding temperature timing signal is subjected to multi-resolution signal decomposition processing, which peels the servo motor winding temperature timing signal into temperature slowly varying component sequences and temperature transient component sequences layer by layer. The temperature slowly varying component sequences retain the low-frequency profile features of the servo motor winding temperature timing signal on each decomposition layer, and the temperature transient component sequences retain the high-frequency oscillation features of the servo motor winding temperature timing signal on each decomposition layer. The servo motor current timing signal is subjected to multi-resolution signal decomposition processing, which peels the servo motor current timing signal into a current slowly varying component sequence and a current transient component sequence layer by layer. The current slowly varying component sequence retains the low-frequency contour features of the servo motor current timing signal at each decomposition layer, and the current transient component sequence retains the high-frequency oscillation features of the servo motor current timing signal at each decomposition layer. The servo motor speed timing signal is subjected to multi-resolution signal decomposition processing, which peels the servo motor speed timing signal into a speed gradually changing component sequence and a speed transient component sequence layer by layer. The speed gradually changing component sequence retains the low-frequency contour features of the servo motor speed timing signal at each decomposition layer, and the speed transient component sequence retains the high-frequency oscillation features of the servo motor speed timing signal at each decomposition layer. The axial force timing signal of the lead screw is subjected to multi-resolution signal decomposition processing, and the axial force timing signal of the lead screw is peeled into a force gradually changing component sequence and a force transient component sequence layer by layer. The force gradually changing component sequence retains the low-frequency profile features of the axial force timing signal of the lead screw at each decomposition layer, and the force transient component sequence retains the high-frequency oscillation features of the axial force timing signal of the lead screw at each decomposition layer. The lead screw displacement timing signal is subjected to multi-resolution signal decomposition processing, which peels the lead screw displacement timing signal into a displacement slowly varying component sequence and a displacement transient component sequence layer by layer. The displacement slowly varying component sequence retains the low-frequency contour features of the lead screw displacement timing signal at each decomposition layer, and the displacement transient component sequence retains the high-frequency oscillation features of the lead screw displacement timing signal at each decomposition layer. For each layer of transient component sequences of temperature, current, rotational speed, force, and displacement, abrupt change point localization processing is performed to extract the time position and intensity of energy abrupt changes in each time series signal of each layer, generating a subset of abrupt change position identifiers and a subset of abrupt change intensity identifiers for each signal in each decomposition layer; the subsets of abrupt change intensity identifiers for all signals in the same decomposition layer are aligned according to time index to obtain a combined abrupt change intensity sequence in each decomposition layer; The mutation point localization process is performed on the mutation intensity sequence of each layer, and the time position and mutation intensity of energy mutation in the time series signal of each layer are extracted to generate the mutation position identifier set and mutation intensity identifier set corresponding to each layer. Interlayer trend coupling processing is performed on the temperature slowly varying component sequence, the current slowly varying component sequence, the rotational speed slowly varying component sequence, the force slowly varying component sequence, and the displacement slowly varying component sequence. The interlayer trend coupling processing is used to extract the degree of trend consistency and trend deviation between each slowly varying component sequence on the same decomposition layer, and generate the trend coupling degree distribution on each decomposition layer. Based on the combined mutation intensity sequence, the mutation location identifier subset, and the trend coupling degree distribution, a time-frequency localized feature map is constructed with the decomposition layer as the vertical axis and the time index as the horizontal axis. The time-frequency localized feature maps corresponding to all decomposition layers are spliced ​​along the vertical axis to generate a time-frequency localized feature set that reflects the distribution law of signal mutation features across the entire frequency range. Each coordinate node in the time-frequency localized feature map records the mutation intensity value and trend coupling degree value at the corresponding time on the decomposition layer.

3. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The process involves calling a pre-built dynamic response model of the creep system, performing creep state correlation analysis on the time-frequency localized feature set, and generating a creep process state transition representation vector, including: The time-frequency localization feature set is input into the feature encoding layer of the dynamic response model of the creep system. The feature encoding layer performs time-dimensional sequence encoding processing on the mutation intensity value and trend coupling degree value of each decomposition layer in the time-frequency localization feature set to generate the time-series feature encoding vector corresponding to each decomposition layer. The temporal feature encoding vectors corresponding to each decomposition layer are input into the inter-layer information transfer layer of the dynamic response model of the creep system. The inter-layer information transfer layer transfers the temporal feature encoding vectors of the lower decomposition layers to the higher decomposition layers layer by layer and fuses them with the temporal feature encoding vectors of the higher decomposition layers to generate inter-layer progressive fusion feature vectors. The interlayer progressive fusion feature vector is input into the creep stage discrimination layer of the creep system dynamic response model. The creep stage discrimination layer performs stage boundary identification processing on the interlayer progressive fusion feature vector and outputs the preliminary stage identification sequence corresponding to the current time window. The preliminary stage identification sequence is used to distinguish between transient creep stage, steady-state creep stage and accelerated creep stage. The preliminary stage identifier sequence is input into the stage evolution tracking layer of the creep system dynamic response model. The stage evolution tracking layer tracks the stage migration path of the preliminary stage identifier sequence in a continuous time window and generates a stage migration trajectory descriptor. The stage migration trajectory descriptor records the direction and number of changes of the creep stage between adjacent time windows. The interlayer progressive fusion feature vector and the stage migration trajectory descriptor are input into the state transition modeling layer of the creep system dynamic response model. The state transition modeling layer dynamically weights the interlayer progressive fusion feature vector according to the stage change direction recorded in the stage migration trajectory descriptor, so that the temporal features in the interlayer progressive fusion feature vector that are consistent with the stage change direction obtain enhanced response and generate state transition driving features. The state transition driving features are input into the evolution parameter estimation layer of the dynamic response model of the creep system. The evolution parameter estimation layer performs evolution direction projection processing and evolution degree quantification processing on the state transition driving features to generate an evolution direction vector and an evolution degree parameter. The evolution direction vector is used to indicate the transition tendency of the creep state from the current stage to the next stage, and the evolution degree parameter is used to indicate the progress depth within the creep stage at the current moment. The current stage identifier, the evolution direction vector, and the evolution degree parameter in the preliminary stage identifier sequence are combined and encoded to generate a creep process state transition representation vector containing stage attribute dimension, direction attribute dimension, and degree attribute dimension.

4. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The step of determining the adjustment constraint boundary of the servo control parameters in the current creep stage based on the creep process state transition characterization vector, and generating control parameter optimization instructions for dynamically calibrating the servo control parameters based on the adjustment constraint boundary, includes: Extract the stage attribute dimension component from the creep process state transition characterization vector, and determine the current creep stage as one of the following stage types: transient creep stage, steady-state creep stage, or accelerated creep stage, based on the stage attribute dimension component. The directional attribute dimension component is extracted from the creep process state transition characterization vector, and the transition tendency direction corresponding to the current creep stage is determined based on the directional attribute dimension component. The transition tendency direction is used to indicate the downstream stage type of creep state evolution. The degree attribute dimension component is extracted from the creep process state transition representation vector, and the progress depth parameter within the current creep stage is determined based on the degree attribute dimension component. The progress depth parameter is used to characterize the position ratio within the current creep stage at the current moment. Obtain the pre-stored stage parameter constraint mapping relationship. The stage parameter constraint mapping relationship records the first parameter constraint range corresponding to the transient creep stage, the second parameter constraint range corresponding to the steady-state creep stage, and the third parameter constraint range corresponding to the accelerated creep stage. The first parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The second parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The third parameter constraint range includes the upper limit threshold of motor current and the upper limit threshold of motor speed. The value ranges of the first parameter constraint range, the second parameter constraint range, and the third parameter constraint range are all different. Based on the current creep stage type and the progress depth parameter, constraint boundary shrinkage processing is performed within the corresponding parameter constraint range. The constraint boundary shrinkage processing adjusts the corresponding motor current upper limit threshold and motor speed upper limit threshold from the initial boundary value inward according to the progress depth parameter and a preset shrinkage rule, generating current constraint boundary, speed constraint boundary and force constraint boundary. The system acquires the real-time current value of the servo motor current timing signal, the real-time speed value of the servo motor speed timing signal, and the real-time force value of the lead screw axial force timing signal. It generates a current adjustment direction based on the difference between the current constraint boundary and the real-time current value, generates a speed adjustment direction based on the difference between the speed constraint boundary and the real-time speed value, and generates a force adjustment direction based on the difference between the force constraint boundary and the real-time force value. The current adjustment direction, the speed adjustment direction, and the force adjustment direction are respectively used as the current setpoint adjustment amount of the servo current loop, the speed setpoint adjustment amount of the servo speed loop, and the force setpoint adjustment amount of the servo force loop, generating dynamic calibration values ​​of servo control parameters for the servo current loop, the servo speed loop, and the servo force loop. The control parameter optimization instruction is generated based on the dynamic calibration value of the servo control parameters. The control parameter optimization instruction is used to drive the servo current loop to perform current adjustment operation, drive the servo speed loop to perform speed adjustment operation, and drive the servo force loop to perform force adjustment operation.

5. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The method further includes: The set of historical servo drive signals accumulated by the intelligent creep machine during the historical loading cycle is obtained. The set of historical servo drive signals includes historical servo motor winding temperature timing signal, historical servo motor current timing signal, historical servo motor speed timing signal, historical lead screw axial force timing signal and historical lead screw displacement timing signal, as well as the creep stage manual annotation record corresponding to each historical timing signal. Multi-resolution signal decomposition processing is performed on the historical servo motor winding temperature timing signal, the historical servo motor current timing signal, the historical servo motor speed timing signal, the historical lead screw axial force timing signal, and the historical lead screw displacement timing signal, respectively, to obtain the historical approximate component sequence and historical detail component sequence of each historical timing signal on the multi-level decomposition layer; Based on the historical approximate component sequence and the historical detail component sequence, abrupt feature extraction and trend coupling degree calculation are performed on each historical time series signal to generate a historical time-frequency localized feature sample set. Each sample in the historical time-frequency localized feature sample set is labeled with a corresponding creep stage label. An initial creep system dynamic response model is constructed, which includes a feature encoding layer, an inter-layer information transmission layer, a creep stage discrimination layer, a stage evolution tracking layer, a state transition modeling layer, and an evolution parameter estimation layer connected in sequence. The network weight parameters of each layer are set using a random initialization method. The historical time-frequency localized feature sample set is input into the initial creep system dynamic response model. The feature encoding layer generates sample time-series feature encoding vectors, the inter-layer information transfer layer generates sample inter-layer progressive fusion feature vectors, the creep stage discrimination layer outputs sample preliminary stage identifier sequences, the stage evolution tracking layer outputs sample stage migration trajectory descriptors, the state transition modeling layer outputs sample state transition driving features, and the evolution parameter estimation layer outputs sample evolution direction vectors and sample evolution degree parameters. The sample preliminary stage identifier sequence and the creep stage label are subjected to cross-entropy loss calculation to obtain the stage discrimination loss value. The sample evolution direction vector and the standard evolution direction vector determined according to the creep stage label are subjected to direction deviation loss calculation to obtain the evolution direction loss value. The sample evolution degree parameter and the stage internal progress reference value determined according to the creep stage label are subjected to degree deviation loss calculation to obtain the evolution degree loss value. The stage discrimination loss value, the evolution direction loss value, and the evolution degree loss value are weighted and fused to generate a joint training total loss value. Based on the joint training total loss value, the network weight parameters of each layer in the initial creep system dynamic response model are updated by back gradient propagation. The weighted fusion process and the back gradient propagation update process are iteratively executed until the joint training total loss value meets the preset convergence condition. The network weight parameters that meet the preset convergence condition are solidified to obtain the creep system dynamic response model and stored in the model storage unit.

6. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The method further includes: Based on the progress depth parameter corresponding to the degree attribute dimension component in the creep process state transition characterization vector, a calibration rate adjustment coefficient is generated in the dynamic calibration process of servo control parameters. The calibration rate adjustment coefficient is positively correlated with the progress depth parameter. When the progress depth parameter increases, the calibration rate adjustment coefficient increases synchronously. Obtain the initial values ​​of the current loop proportional gain and the current loop integral gain of the servo current loop. Multiply the initial value of the current loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic current loop proportional gain. Multiply the initial value of the current loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic current loop integral gain. Write the dynamic current loop proportional gain and the dynamic current loop integral gain into the control register of the servo current loop. Obtain the initial value of the speed loop proportional gain and the initial value of the speed loop integral gain of the servo speed loop. Multiply the initial value of the speed loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic speed loop proportional gain. Multiply the initial value of the speed loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic speed loop integral gain. Write the dynamic speed loop proportional gain and the dynamic speed loop integral gain into the control register of the servo speed loop. Obtain the initial value of the force loop proportional gain and the initial value of the force loop integral gain of the servo force loop. Multiply the initial value of the force loop proportional gain by the calibration rate adjustment coefficient to obtain the dynamic force loop proportional gain. Multiply the initial value of the force loop integral gain by the calibration rate adjustment coefficient to obtain the dynamic force loop integral gain. Write the dynamic force loop proportional gain and the dynamic force loop integral gain into the control register of the servo force loop. A feedforward compensation gain adjustment coefficient is generated based on the progress depth parameter. The feedforward compensation gain adjustment coefficient is negatively correlated with the progress depth parameter. When the progress depth parameter increases, the feedforward compensation gain adjustment coefficient decreases synchronously. Obtain a preset initial gain value for speed feedforward compensation, multiply the initial gain value for speed feedforward compensation by the feedforward compensation gain adjustment coefficient to obtain the dynamic speed feedforward compensation gain, and write the dynamic speed feedforward compensation gain into the feedforward compensation control channel of the servo speed loop. Obtain a preset initial gain value for force feedforward compensation, multiply the initial gain value for force feedforward compensation by the feedforward compensation gain adjustment coefficient to obtain the dynamic force feedforward compensation gain, and write the dynamic force feedforward compensation gain into the feedforward compensation control channel of the servo force loop. A filter cutoff frequency adjustment coefficient is generated based on the progress depth parameter. The filter cutoff frequency adjustment coefficient is negatively correlated with the progress depth parameter. When the progress depth parameter increases, the filter cutoff frequency adjustment coefficient decreases synchronously. The following steps are taken: obtain the preset initial cutoff frequency of the current loop filter in the servo current loop, the preset initial cutoff frequency of the speed loop filter in the servo speed loop, and the preset initial cutoff frequency of the force loop filter in the servo force loop; multiply the initial cutoff frequency of the current loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic current loop filter cutoff frequency; multiply the initial cutoff frequency of the speed loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic speed loop filter cutoff frequency; and multiply the initial cutoff frequency of the force loop filter by the filter cutoff frequency adjustment coefficient to obtain the dynamic force loop filter cutoff frequency. Write the cutoff frequency of the dynamic current loop filter into the filter parameter register of the servo current loop, write the cutoff frequency of the dynamic speed loop filter into the filter parameter register of the servo speed loop, and write the cutoff frequency of the dynamic force loop filter into the filter parameter register of the servo force loop.

7. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The method further includes: The lead screw displacement timing signal is acquired, and the displacement change rate is extracted from the lead screw displacement timing signal. The ratio of the displacement change to the time interval at adjacent sampling times is calculated to generate the lead screw movement speed timing signal. The axial force timing signal and the moving speed timing signal of the lead screw are obtained. The axial force timing signal and the moving speed timing signal of the lead screw are combined on a unified time axis to generate a force-velocity phase plane trajectory diagram. The force-velocity phase plane trajectory diagram uses the axial force of the lead screw as the first dimension coordinate and the moving speed of the lead screw as the second dimension coordinate. The force-velocity phase plane trajectory diagram is processed by extracting graphic morphological features, identifying the closed loop area features, trajectory spiral direction features, and trajectory boundary offset distance features in the force-velocity phase plane trajectory diagram, and generating a phase plane trajectory morphological feature set; Extract stage attribute dimension components from the creep process state transition characterization vector, determine the current creep stage type based on the stage attribute dimension components, extract degree attribute dimension components from the creep process state transition characterization vector, and determine the current progress depth parameter based on the degree attribute dimension components; Obtain the phase plane trajectory morphology feature record corresponding to the accelerated creep stage in the pre-stored historical creep failure cases. The phase plane trajectory morphology feature record of the historical creep failure cases includes the closed loop area feature sequence before failure, the trajectory spiral direction feature sequence before failure, and the trajectory boundary offset distance feature sequence before failure. The closed loop area feature, trajectory spiral direction feature, and trajectory boundary offset distance feature in the phase plane trajectory morphology feature set are compared with the closed loop area feature sequence, the trajectory spiral direction feature sequence, and the trajectory boundary offset distance feature sequence before failure to obtain a similarity matching result. The similarity matching result is used to characterize the degree of similarity between the current phase plane trajectory morphology and the historical phase plane trajectory morphology before failure. When the current creep stage type is a steady-state creep stage and the similarity matching result exceeds a preset similarity threshold, an accelerated creep warning trigger instruction is generated. The accelerated creep warning trigger instruction includes the comparison difference between the current closed loop area feature and the last record in the sequence of closed loop area features before failure, the comparison difference between the current trajectory spiral direction feature and the last record in the sequence of trajectory spiral direction features before failure, and the current progress depth parameter. According to the accelerated creep warning trigger command, the adjustment constraint boundary is pre-shrinked. The pre-shrinking process reduces the values ​​of the upper limit threshold of motor current and the upper limit threshold of motor speed in advance on the basis of the original constraint boundary shrinkage adjustment.

8. The servo system control optimization method for intelligent creep machines according to claim 4, characterized in that, The method further includes: The non-contact deformation measurement signal within the gauge length section of the specimen is acquired during the creep test performed by the intelligent creep machine. The non-contact deformation measurement signal includes the axial strain timing signal and the radial strain timing signal of the gauge length section. The axial strain timing signal is acquired by a non-contact deformation measurement unit installed in the axial direction of the gauge length section of the specimen, and the radial strain timing signal is acquired by a non-contact deformation measurement unit installed in the radial direction of the gauge length section of the specimen. The axial strain time series signal of the gauge length segment is subjected to strain rate extraction processing, and the ratio of the axial strain change at adjacent sampling times to the time interval is calculated to generate an axial strain rate time series signal. The radial strain time series signal of the gauge length segment is subjected to strain rate extraction processing, and the ratio of the radial strain change at adjacent sampling times to the time interval is calculated to generate a radial strain rate time series signal. Based on the axial strain rate time-series signal and the radial strain rate time-series signal, the ratio of radial strain rate to axial strain rate at each sampling moment is calculated to generate a strain rate ratio time-series signal. The strain rate ratio time series signal is subjected to trend extraction processing to extract the monotonically changing trend segments and trend inflection point positions in the strain rate ratio time series signal, and a strain rate ratio change trend descriptor is generated. The strain rate ratio change trend descriptor records the start and end times and the increasing slope parameter of the monotonically increasing segment, the start and end times and the decreasing slope parameter of the monotonically decreasing segment, and the time position of the trend inflection point. Extract the stage attribute dimension component from the creep process state transition characterization vector, determine the current creep stage type based on the stage attribute dimension component, and when the current creep stage type is a steady-state creep stage, read the time position of the most recent trend inflection point from the strain rate ratio change trend descriptor. The time difference between the most recent trend inflection point and the start time of the steady-state creep stage is calculated to obtain the duration of the steady-state creep stage. The duration of the steady-state creep stage is then compared with the historical duration of the steady-state creep stage of similar materials under the same test conditions in a pre-stored database to generate an estimated remaining duration of the steady-state creep stage. A time margin parameter is generated based on the estimated remaining duration of the steady-state creep stage. The time margin parameter is used to indicate the remaining available time before entering the accelerated creep stage. The time margin parameter is written into the control parameter optimization instruction, so that the control parameter optimization instruction includes a time margin prompt. When the current creep stage type is accelerated creep stage, the axial strain rate time series signal is subjected to acceleration trend analysis processing, the acceleration growth slope parameter of the axial strain rate is extracted, and the acceleration creep stage termination time estimate is generated based on the acceleration growth slope parameter and the degree attribute dimension component. The acceleration creep stage termination time estimate is written into the control parameter optimization instruction, so that the termination time prompt information is added to the control parameter optimization instruction.

9. The servo system control optimization method for intelligent creep machines according to claim 1, characterized in that, The method further includes: The servo motor current timing signal and servo motor voltage timing signal are acquired, and the servo motor current timing signal and servo motor voltage timing signal are processed for power calculation. The servo motor current value and servo motor voltage value at the same sampling time are multiplied to generate the real-time power timing signal of the servo motor. The real-time power timing signal of the servo motor is subjected to power fluctuation analysis and processing. The power fluctuation amplitude characteristics and power fluctuation frequency characteristics of the real-time power timing signal of the servo motor within a preset time window are extracted to generate a power fluctuation feature descriptor. The axial force timing signal and the displacement timing signal of the lead screw are acquired. The work done is calculated and processed by the axial force timing signal and the displacement timing signal of the lead screw. The average value of the axial force of the lead screw at adjacent sampling times is multiplied by the change in the displacement of the lead screw and integrated over time to generate the cumulative work done by the lead screw timing signal. The working rate extraction processing is performed on the cumulative work time sequence signal of the lead screw, and the ratio of the change in cumulative work at adjacent sampling times to the time interval is calculated to generate the working rate time sequence signal of the lead screw. The real-time power timing signal of the servo motor and the working speed timing signal of the lead screw are compared on the same time axis. The difference between the real-time power value of the servo motor and the working speed value of the lead screw at each sampling moment is calculated to generate the transmission efficiency loss power timing signal. The transmission efficiency loss power time-series signal is subjected to cumulative analysis and processing. The transmission efficiency loss power value within a preset time window is integrated over time to generate a transmission efficiency loss cumulative energy time-series signal. Extract the stage attribute dimension component and the degree attribute dimension component from the creep process state transition characterization vector, and input the stage attribute dimension component, the degree attribute dimension component, the power fluctuation feature descriptor and the transmission efficiency loss cumulative energy time series signal into the pre-constructed servo drive health assessment model to generate servo drive health assessment value. Based on the servo drive health assessment value, a transmission system maintenance prompt instruction is generated. When the servo drive health assessment value is lower than the preset health threshold, the transmission system maintenance prompt instruction is appended to the control parameter optimization instruction.

10. A servo system control optimization system applied to an intelligent creep machine, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the servo system control optimization method for an intelligent creep machine as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images