Dynamic automatic analysis method of worm gear meshing center distance based on servo drive

Abstract

This invention provides a servo-driven, automatic dynamic analysis method for worm gear meshing center distance. This method involves collecting state parameters of worm gear transmission systems; processing and correcting center distance data using a neural network adaptive filter; establishing a dynamic model to calculate optimal center distance values ​​and obtain a deviation sequence; using a bidirectional, gated recurrent neural network enhanced with an attention mechanism to predict variation trends; and outputting compensation commands to drive a servo motor to perform adjustments. This invention achieves real-time, accurate measurement and intelligent dynamic compensation of worm gear meshing center distance, improving the accuracy and stability of the transmission system.

Description

Technical Field

[0001] The present invention relates to analysis technology, in particular to a dynamic automatic analysis method of worm gear meshing center distance based on servo drive. Background Art

[0002] Worm gear transmissions, as a crucial component of mechanical transmissions, are widely used in industrial robots, precision machine tools, automated production lines, and other fields. These systems offer advantages such as high transmission ratios, compact structures, and smooth transmission. However, precise control of the meshing center distance directly impacts the system's transmission accuracy, noise levels, and service life. In actual operating environments, the meshing center distance of the worm gears can dynamically change due to factors such as temperature fluctuations, load fluctuations, and wear, leading to degradation in transmission system performance.

[0003] Traditional methods for controlling the center distance of worm gear meshing rely primarily on static adjustment and periodic maintenance, making them incapable of handling dynamic changes during operation. In recent years, with the development of intelligent manufacturing and Industry 4.0, the requirements for the precision and reliability of transmission systems have continued to increase, and dynamic adjustment methods based on servo drives have gradually attracted attention.

[0004] Traditional center distance measurement methods lack real-time performance and accuracy. They usually use contact measurement or simple sensor monitoring, which cannot effectively eliminate environmental interference and system noise. Especially in working environments with large temperature fluctuations, the measurement results have obvious deviations, making it difficult to provide a reliable basis for subsequent adjustments.

[0005] Existing center distance compensation strategies mostly use control methods based on fixed models, which have poor adaptability to system parameter changes and nonlinear characteristics, and cannot dynamically adjust the compensation strategy according to the system status, resulting in inconsistent compensation effects under different working conditions, especially under extreme working conditions such as high speed and heavy load. The compensation accuracy decreases.

[0006] Existing technologies lack the ability to predict the changing trends of center distances and mostly adopt passive response adjustments. They are unable to proactively identify the changing trends of system performance and it is difficult to implement preventive maintenance before problems expand. This leads to insufficient system operation stability and difficulty in ensuring long-term reliability. Summary of the Invention

[0007] The embodiment of the present invention provides a dynamic automatic analysis method of worm gear meshing center distance based on servo drive, which can solve the problems in the prior art.

[0008] A first aspect of an embodiment of the present invention provides a servo-driven dynamic automatic analysis method for worm gear meshing center distance, comprising:

[0009] collecting state parameters of a worm gear transmission system to generate an initial measurement data set; using the initial measurement data set to train a neural network adaptive filter, wherein the neural network adaptive filter is based on a long short-term memory network architecture, establishes a nonlinear mapping relationship with the center distance signal, and performs real-time correction on the center distance measurement result by introducing a temperature compensation coefficient, and outputs the corrected center distance data;

[0010] Based on the corrected center distance data, combined with the worm wheel speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence;

[0011] Based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by an attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data. The compensation strategy is continuously optimized in combination with the online transfer learning method to output the optimal compensation instruction.

[0012] The optimal compensation instruction is converted into a displacement control signal of a servo motor, and the servo motor is driven to perform a center distance adjustment operation to achieve dynamic compensation of the worm gear meshing center distance.

[0013] The initial measurement data set is used to train a neural network adaptive filter. The neural network adaptive filter is based on a long short-term memory network architecture and establishes a nonlinear mapping relationship with the center distance signal. The center distance measurement result is corrected in real time by introducing a temperature compensation coefficient. The output of the corrected center distance data includes:

[0014] collecting real-time operating data of a worm gear transmission system and generating a multimodal measurement data set based on the real-time operating data; establishing a high-fidelity mapping environment for the transmission system based on the multimodal measurement data set, constructing a nonlinear mapping relationship between a center distance signal and the multimodal measurement data set in the high-fidelity mapping environment, and mapping initial center distance data using the nonlinear mapping relationship to obtain optimized center distance data;

[0015] Calculating a temperature compensation coefficient based on the optimized center distance data in combination with temperature data in the multimodal measurement data set, wherein the temperature compensation coefficient is obtained by an exponential decay function of an initial compensation coefficient and a temperature change, performing a weighted fusion of the temperature compensation coefficient and the optimized center distance data, and introducing a dynamic bias term for compensation to generate a compensated center distance measurement value;

[0016] The compensated center distance measurement value is input into the high-fidelity mapping environment, compensation parameters are calculated based on historical compensation data in the high-fidelity mapping environment, the compensated center distance measurement value is optimized and corrected using the compensation parameters, and the corrected center distance measurement data is output.

[0017] Based on the corrected center distance data, combined with the worm gear speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence, which includes:

[0018] collecting a dynamic data set of a worm gear transmission system, where the dynamic data set is used to characterize the real-time operating state of the transmission system; constructing a spiking neural network model based on the dynamic data set, where the spiking neural network model simulates transmission system characteristics through sodium ion channel conductance and potassium ion channel conductance, receives the dynamic data set and calculates a kinetic current based on the sodium ion channel conductance and the potassium ion channel conductance, and performs a weighted combination of the kinetic current and a basic driving torque to generate a synaptic modulation torque;

[0019] Substituting the synaptic modulation torque into an enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; performing time-domain encoding on the dynamic response characteristics using the pulse neural network model to generate a pulse train signal, wherein the pulse train signal describes the time-varying characteristics of the dynamic characteristics using a time exponential decay function to obtain a dynamic characteristic pulse code;

[0020] Based on the dynamic characteristic pulse code, a time-varying kernel function is constructed using the pulse neural network model, the time-varying kernel function is determined by the dynamically reconstructed synaptic structure, a spatiotemporal convolution operation is performed on the dynamic characteristic pulse code using the time-varying kernel function to obtain the instantaneous state characteristics of the transmission system, and a theoretical optimal center distance is calculated based on the instantaneous state characteristics;

[0021] The theoretical optimal center distance is compared with the corrected center moment data to obtain center distance deviation data.

[0022] Substituting the synaptic modulation torque into the enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; and performing time-domain encoding on the dynamic response characteristics by using the pulse neural network model to generate a pulse train signal, comprising:

[0023] A field-theory enhanced dynamics equation is constructed based on the synaptic modulation torque. The field-theory enhanced dynamics equation introduces a field-matter coupling term. The field state distribution of the transmission system is obtained by solving the field-theory enhanced dynamics equation. The field state distribution reflects the overall dynamic state of the transmission system.

[0024] Substituting the field state distribution into the field-theory enhanced dynamics equation, dynamically updating the coupling coefficient in the field-matter coupling force term according to the real-time operating state of the transmission system, adjusting the parameters of the field-theory enhanced dynamics equation in combination with the variation law of the synaptic modulation torque, and solving the field-theory enhanced dynamics equation again to obtain the coupled dynamics response of the transmission system, wherein the coupled dynamics response represents the nonlinear dynamic behavior of the transmission system;

[0025] Applying a field state projection transformation to the coupled dynamic response to extract the characteristics of the system's velocity component, acceleration component, and field state component, and generating a feature vector representing the field state through a linear combination of characteristic mode functions, wherein the feature vector contains multi-scale dynamic information of the transmission system;

[0026] The characteristic vector of the field state representation is input into a pulse neural network, the dynamic characteristics are time-domain encoded by the pulse neural network, and a timestamp function and an exponential decay function are introduced to modulate the encoding process to generate a pulse sequence signal with time-varying characteristics.

[0027] Based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by the attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data. The compensation strategy is continuously optimized by combining the online transfer learning method, and the optimal compensation instructions are output, including:

[0028] constructing a bidirectional gated recurrent neural network, inputting the forward hidden layer state and the reverse hidden layer state of the bidirectional gated recurrent neural network into a swarm intelligence emergence layer, wherein the swarm intelligence emergence layer introduces a swarm synergy factor to enhance the hidden layer state and generate enhanced hidden layer features;

[0029] Constructing a self-evolving attention weight based on the enhanced hidden layer features, wherein the self-evolving attention weight calculates a fitness function by the ratio of the prediction error to the reference error threshold, and multiplying the fitness function by the attention base weight to obtain an adaptive attention weight;

[0030] Establishing an environmental pressure selection mechanism based on the adaptive attention weight, wherein the environmental pressure selection mechanism calculates a migration probability based on the adaptation distance from the source domain to the target domain, uses the migration probability as a weight coefficient of the source domain loss function, and updates the transfer learning model parameters through gradient descent;

[0031] Modulating the enhanced hidden layer features based on the transfer learning model parameters to generate a dynamic feature vector, inputting the dynamic feature vector and the adaptive attention weight into a compensation strategy generation module, wherein the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator;

[0032] The evolution modulation factor is multiplied by the initial compensation instruction to obtain the optimal compensation instruction, the compensation effect corresponding to the optimal compensation instruction is fed back to the fitness function, the self-evolution attention weight is updated, and dynamic optimization of the compensation strategy is achieved.

[0033] Inputting the dynamic feature vector and the adaptive attention weight into a compensation strategy generation module, wherein the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator, including:

[0034] Combining the dynamic feature vector with the adaptive attention weight to generate a multi-level feature, inputting the multi-level feature into a compensation strategy generation module, introducing a dynamic response enhancement function through the compensation strategy generation module to enhance the multi-level feature, the dynamic response enhancement function is adaptively adjusted based on the feature energy level difference to generate an initial compensation strategy, wherein the initial compensation strategy reflects the dynamic response characteristics of the system;

[0035] Constructing a multi-scale selection pressure index based on the initial compensation strategy, wherein the multi-scale selection pressure index is calculated by normalizing the distance between the initial compensation strategy and a reference strategy, wherein the reference strategy sets corresponding reference values ​​at different scales;

[0036] applying stability protection to the multi-scale selective pressure indicator, structurally enhancing the multi-scale selective pressure indicator through a dynamic modulation field, dynamically adjusting the dynamic modulation field based on real-time changes in system state, providing continuous stability protection through closed path integrals of the dynamic modulation field, and generating a steady-state enhanced selective pressure indicator having anti-interference capability;

[0037] The steady-state enhanced selection pressure index is substituted into the evolution modulation factor calculation formula, and the steady-state enhanced selection pressure index is weightedly calculated using weight coefficients of different scales. The weight coefficients are adaptively adjusted as the scale changes. The collaborative optimization of multi-scale features is achieved through the dynamic allocation of the weight coefficients of different scales to obtain the evolution modulation factor.

[0038] The closed-path integral of the dynamic modulation field provides continuous stability protection, generating steady-state enhanced selective pressure indicators including:

[0039] The dynamic modulation field is dynamically adjusted based on the changing trend of the real-time state of the system, and the stability requirements of the system are reflected through the changing characteristics of the dynamic modulation field;

[0040] performing a closed path integral operation on the dynamic modulation field, wherein the closed path integral operation is calculated along a closed trajectory in the system state space to generate a stability protection factor, wherein the stability protection factor is used to characterize the overall stability characteristics of the system;

[0041] The stability protection factor is integrated with the selection pressure index, and continuity constraints are imposed on the selection pressure index through the stability protection factor to generate a steady-state enhanced selection pressure index with anti-interference ability.

[0042] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0043] processor;

[0044] a memory for storing processor-executable instructions;

[0045] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0046] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0047] The beneficial effects of this application are as follows:

[0048] The servo-driven dynamic automatic analysis method for worm gear meshing center distance provided by the present invention processes measurement data through a neural network adaptive filter and introduces a temperature compensation coefficient, thereby improving the accuracy of center distance measurement and solving the problem of decreased accuracy of traditional measurement methods under temperature-changing working conditions.

[0049] This method combines the dynamic model with measured data to establish a comparison mechanism between the theoretical optimal center distance and the actual value. It can quickly identify the center distance deviation and predict the center distance change trend through a bidirectional gated recurrent neural network enhanced by the attention mechanism, thereby achieving more accurate compensation strategy formulation.

[0050] The present invention adopts online transfer learning to continuously optimize the compensation strategy and performs real-time adjustment through a servo motor, thereby realizing adaptive dynamic compensation of the worm gear meshing center distance, effectively improving the operating stability and service life of the transmission system, reducing noise and vibration, and being suitable for the real-time control needs of high-precision transmission occasions. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1Schematic diagram of the process of a dynamic automatic analysis method of worm gear meshing center distance based on servo drive according to an embodiment of the present invention;

[0052] Figure 2 A bar chart comparing and analyzing the center distance measurement methods of worm gear transmission systems according to an embodiment of the present invention;

[0053] Figure 3 Schematic diagram comparing the center distance measurement accuracy of the worm gear transmission system according to an embodiment of the present invention;

[0054] Figure 4 This is a flowchart of the optimization of the prediction and compensation strategy of the bidirectional gated recurrent neural network enhanced by the attention mechanism according to an embodiment of the present invention;

[0055] Figure 5 This is a bar chart comparing the effects of multi-scale selection pressure indicators and evolutionary modulation factors in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0057] The technical solution of the present invention is described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0058] Figure 1 FIG. 1 is a flow chart of a method for dynamic automatic analysis of worm gear meshing center distance based on servo drive according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] collecting state parameters of a worm gear transmission system to generate an initial measurement data set; using the initial measurement data set to train a neural network adaptive filter, wherein the neural network adaptive filter is based on a long short-term memory network architecture, establishes a nonlinear mapping relationship with the center distance signal, and performs real-time correction on the center distance measurement result by introducing a temperature compensation coefficient, and outputs the corrected center distance data;

[0060] Based on the corrected center distance data, combined with the worm wheel speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence;

[0061] Based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by an attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data. The compensation strategy is continuously optimized in combination with the online transfer learning method to output the optimal compensation instruction.

[0062] The optimal compensation instruction is converted into a displacement control signal of a servo motor, and the servo motor is driven to perform a center distance adjustment operation to achieve dynamic compensation of the worm gear meshing center distance.

[0063] In an optional embodiment, the initial measurement data set is used to train a neural network adaptive filter. The neural network adaptive filter is based on a long short-term memory network architecture and establishes a nonlinear mapping relationship with the center distance signal. The center distance measurement result is corrected in real time by introducing a temperature compensation coefficient. The output of the corrected center distance data includes:

[0064] collecting real-time operating data of a worm gear transmission system and generating a multimodal measurement data set based on the real-time operating data; establishing a high-fidelity mapping environment for the transmission system based on the multimodal measurement data set, constructing a nonlinear mapping relationship between a center distance signal and the multimodal measurement data set in the high-fidelity mapping environment, and mapping initial center distance data using the nonlinear mapping relationship to obtain optimized center distance data;

[0065] Calculating a temperature compensation coefficient based on the optimized center distance data in combination with temperature data in the multimodal measurement data set, wherein the temperature compensation coefficient is obtained by an exponential decay function of an initial compensation coefficient and a temperature change, performing a weighted fusion of the temperature compensation coefficient and the optimized center distance data, and introducing a dynamic bias term for compensation to generate a compensated center distance measurement value;

[0066] The compensated center distance measurement value is input into the high-fidelity mapping environment, compensation parameters are calculated based on historical compensation data in the high-fidelity mapping environment, the compensated center distance measurement value is optimized and corrected using the compensation parameters, and the corrected center distance measurement data is output.

[0067] An initial measurement dataset for the worm gear transmission system was collected. This dataset includes raw center distance signals, temperature signals, vibration signals, speed signals, and load signals. A high-precision displacement sensor was used to measure center distance changes with an accuracy of 0.001 mm. A PT100 temperature sensor was used to monitor system temperature, with a measurement range of -50°C to 150°C. A triaxial accelerometer was used to collect vibration signals, with a sampling frequency of 1 kHz. Speed ​​signals were acquired using a Hall effect sensor, covering a range of 100 rpm to 3000 rpm. Load signals were measured using a torque sensor with a range of 0-100 N·m. All signals were collected synchronously using a data acquisition card with a unified sampling rate of 10 kHz, forming a multimodal initial measurement dataset.

[0068] The neural network adaptive filter was trained using the collected initial measurement data set. This filter is based on a long short-term memory (LSTM) network architecture and consists of an input layer, a hidden layer, and an output layer. The input layer receives multimodal signal features, including raw center distance values, temperature values, vibration characteristics, speed values, and load values. The hidden layer consists of three layers of LSTM units, each containing 128 neurons, which capture long-term dependencies between signals through a gating mechanism. The output layer outputs the corrected center distance values. During training, mean squared error (MSE) was used as the loss function, and the network parameters were optimized using a stochastic gradient descent algorithm. The learning rate was set to 0.001, the training epochs were 500, and the batch size was 64.

[0069] During real-time operation, the system continuously collects operating data from the worm gear transmission system to generate a multimodal measurement dataset. In a typical example, the system collected 8 hours of continuous operating data, containing approximately 28,800 data points. Based on this data, the system constructs a high-fidelity mapping environment, a digital twin system that simulates the dynamic characteristics of the transmission system under different operating conditions. Within this environment, a nonlinear mapping relationship between the center distance signal and the multimodal measurement data is established through a deep neural network. Specifically, a fully connected neural network structure is used, consisting of four hidden layers, each with 256, 128, 64, and 32 neurons, respectively, and a ReLU activation function. This mapping relationship takes the original center distance data as input and outputs optimized center distance data. The mapping process takes into account the impact of system operating conditions and environmental factors on center distance.

[0070] Based on the optimized center distance data, the system calculates the temperature compensation coefficient in combination with the temperature data in the multimodal measurement data. The temperature compensation coefficient is obtained by the exponential decay function of the initial compensation coefficient and the temperature change. In actual applications, the initial compensation coefficient is set to 1.0, and the temperature change is calculated by the difference between the current temperature and the standard temperature (25°C). When the temperature change is positive (warming), the compensation coefficient shows a negative correlation; when the temperature change is negative (cooling), the compensation coefficient shows a positive correlation. In one test scenario, when the system temperature increased from 25°C to 45°C, the temperature change was 20°C, and the calculated temperature compensation coefficient was 0.92; when the temperature decreased from 25°C to 5°C, the temperature change was -20°C, and the calculated temperature compensation coefficient was 1.08.

[0071] The temperature compensation coefficient is weighted and fused with the optimized center distance data by multiplying the optimized center distance data by the temperature compensation coefficient. A dynamic offset term is also introduced for compensation, and this offset term is dynamically adjusted based on the system's operating status. For example, during system startup, the offset term is set to 0.005mm; during stable operation, the offset term is reduced to 0.002mm; and under high load conditions, the offset term is increased to 0.008mm. This mechanism generates a compensated center distance measurement value.

[0072] The compensated center distance measurements are input into a high-fidelity mapping environment, and the system calculates compensation parameters based on historical compensation data. This historical compensation data is stored in the system database and contains compensation results and corresponding operating condition information from the past 24 hours. Compensation parameter calculations utilize a sliding window technique, with a window size set to 60 minutes and a step size of 10 minutes. Within each window, the system extracts compensation trend features and calculates the average compensation value, standard deviation, and rate of change. Based on these features, the system calculates the current compensation parameters using a weighted average method, with weights assigned as follows: 0.5 for recent data (0-20 minutes), 0.3 for mid-term data (20-40 minutes), and 0.2 for long-term data (40-60 minutes). In practice, when the system detects that a compensation parameter exceeds a threshold (set to ±0.015mm), an abnormality alarm is triggered, prompting the operator to check the system status.

[0073] The compensated center distance measurement is optimized and corrected using compensation parameters, outputting the final corrected center distance measurement data. The correction process involves adding the compensated center distance measurement value to the compensation parameters to obtain the final correction result. In actual testing, this method reduced the center distance measurement error from the original ±0.05mm to ±0.008mm, improving measurement accuracy by approximately 84%, meeting the center distance control requirements of high-precision worm gear transmission systems.

[0074] Figure 2This is a bar chart comparing and analyzing the center distance measurement methods of the worm gear transmission system according to the embodiment of the present invention:

[0075] The figure compares the performance of three different measurement methods (traditional measurement method, conventional neural network filtering, and long short-term memory network filtering) under four different operating conditions: normal temperature, temperature fluctuation, high load, and mixed conditions. Under normal temperature conditions, the LSTM network filtering achieved a measurement accuracy of 87.6%, significantly outperforming both the 79.2% of the conventional neural network filtering and the 72.5% of the conventional measurement method. Under temperature fluctuation conditions, the LSTM network performed best, achieving an accuracy of 82.3%, followed by the conventional neural network at 74.5%, and the conventional method at 63.8%. Under high load conditions, the performance of the three methods was: 82.5% for the LSTM network, 78.9% for the conventional neural network, and 68.6% for the conventional method. Under mixed conditions, the LSTM network maintained its high performance, reaching 81.8%, while the conventional neural network achieved 67.2% and the conventional method dropped to 58.4%. The data shows that under all working conditions, the long short-term memory network filtering method exhibits the best performance. Especially under complex composite working conditions, its advantages are more obvious, which is more than 20 percentage points higher than the traditional method, reflecting the method's strong adaptability and stability in complex environments.

[0076] In an optional embodiment, based on the corrected center distance data, combined with the worm gear speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence, which includes:

[0077] collecting a dynamic data set of a worm gear transmission system, where the dynamic data set is used to characterize the real-time operating state of the transmission system; constructing a spiking neural network model based on the dynamic data set, where the spiking neural network model simulates transmission system characteristics through sodium ion channel conductance and potassium ion channel conductance, receives the dynamic data set and calculates a kinetic current based on the sodium ion channel conductance and the potassium ion channel conductance, and performs a weighted combination of the kinetic current and a basic driving torque to generate a synaptic modulation torque;

[0078] Substituting the synaptic modulation torque into an enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; performing time-domain encoding on the dynamic response characteristics using the pulse neural network model to generate a pulse train signal, wherein the pulse train signal describes the time-varying characteristics of the dynamic characteristics using a time exponential decay function to obtain a dynamic characteristic pulse code;

[0079] Based on the dynamic characteristic pulse code, a time-varying kernel function is constructed using the pulse neural network model, the time-varying kernel function is determined by the dynamically reconstructed synaptic structure, a spatiotemporal convolution operation is performed on the dynamic characteristic pulse code using the time-varying kernel function to obtain the instantaneous state characteristics of the transmission system, and a theoretical optimal center distance is calculated based on the instantaneous state characteristics;

[0080] The theoretical optimal center distance is compared with the corrected center moment data to obtain center distance deviation data.

[0081] The dynamic data set of the worm gear transmission system is collected. This data set is used to characterize the real-time operating status of the transmission system. The collected data includes the worm gear speed signal ω1, the worm speed signal ω2, the input torque signal T in , output torque signal T out , bearing seat temperature signal θ1, lubricating oil temperature signal θ2, and vibration acceleration signal a. The speed signal is acquired using incremental encoders mounted on the worm gear and worm shafts, with a sampling frequency of 1000 Hz. The torque signal is acquired using torque sensors mounted on the input and output shafts, with a range of 0-200 N·m and an accuracy of 0.5% FS. The temperature signal is acquired using temperature sensors mounted on the bearing seat and lubricating oil tank, with a range of -20°C to 120°C and an accuracy of ±0.5°C. The vibration signal is acquired using an accelerometer mounted on the bearing seat, with a frequency range of 1 Hz-10 kHz and a sensitivity of 100 mV / g. All signals are converted to digital signals using a 16-bit A / D converter and filtered through a low-pass filter (cutoff frequency 400 Hz) to remove high-frequency noise.

[0082] A pulse neural network model was constructed based on the collected dynamic data set. The model uses the Hodgkin-Huxley neuron model to simulate the transmission system characteristics through the conductance of sodium ion channels and potassium ion channels. The membrane potential kinetic equation is expressed as the membrane capacitance multiplied by the time derivative of the membrane potential equals the total ion current plus the external input current. Among them, the membrane capacitance C m The value is 1 microfarad / square centimeter; the total ion current includes the sodium ion current I Na , potassium ion current I K and leakage current I L ; External input current I ext Determined by kinetic data.

[0083] Sodium ion current I Na Equal to the maximum sodium ion channel conductance g Na Multiply by the cube of the sodium channel activation variable m, multiply by the sodium channel inactivation variable h, multiply by the membrane potential V and the sodium ion equilibrium potential E Na The maximum sodium ion channel conductance g NaThe sodium ion equilibrium potential E is 120 millisiemens / cm2. Na The kinetic equation for the sodium channel activation variable m is: The time derivative of m is equal to (1-m) times α. m Subtract m and multiply by β m , where α m and β m is the rate constant related to the membrane potential. The kinetic equation for the sodium channel inactivation variable h is similar, where the time derivative of h is equal to (1-h) times α h Subtract h and multiply by β h .

[0084] Potassium ion current I K Equal to the maximum potassium channel conductance g K Multiply by the fourth power of the potassium channel activation variable n multiplied by the membrane potential V and the potassium ion equilibrium potential E K The maximum potassium ion channel conductance g K The potassium ion equilibrium potential E is 36 millisiemens / cm2. K The kinetic equation for the potassium channel activation variable n is: the derivative of n with respect to time is (1-n) times α n Subtract n times β n , where α n and β n is the rate constant related to the membrane potential.

[0085] Leakage current I L Equal to the leakage current conductance g L Multiply the membrane potential V by the leakage current equilibrium potential E L The difference between the leakage current and the conductance g L The leakage current equilibrium potential E is 0.3 mS / cm2. L It is -54.4 mV.

[0086] External input current I ext It is obtained by linear mapping of kinetic data, and the mapping relationship is I ext Equal to the speed signal mapping coefficient k ω Multiply by the worm gear speed signal ω1 plus k' ω Multiply the worm speed signal ω2 plus the torque signal mapping coefficient k T Multiply the input torque signal T in Add k' T Multiply the output torque signal T out Add the temperature signal mapping coefficient k θ Multiply the bearing seat temperature signal θ1 plus k' θ Multiply the lubricating oil temperature signal θ2 plus the vibration acceleration signal mapping coefficient k aMultiply by the vibration acceleration signal a. Among them, each mapping coefficient is obtained through training optimization, and the typical value is k ω =0.05μA·min / r, k' ω =0.03μA·min / r, k T =0.08μA / N·m, k' T =0.12μA / N·m,k θ =0.02μA / ℃,k' θ =0.015μA / ℃,k a =0.01μA·s 2 / m.

[0087] After the kinetic data is input into the neural network, the kinetic current I is calculated based on the sodium ion channel conductance and the potassium ion channel conductance. dyn The kinetic current calculation formula is I dyn Equal to g Na Multiply by m cubed times h times (VE Na ) plus g K Multiply n to the fourth power by (VE K ) plus g L Multiply by (VE L Under normal operating conditions, the typical values ​​of the kinetic current are a sodium ion current peak of approximately -12 μA / cm², a potassium ion current peak of approximately 7 μA / cm², and a leakage current of approximately 0.5 μA / cm².

[0088] The calculated kinetic current is weightedly combined with the basic driving torque to generate the synaptic modulation torque T syn The synaptic modulation torque is calculated as T syn Equal to the basic driving torque T base Add the weighting coefficient λ and multiply the kinetic current I dyn Basic driving torque T base According to the design parameters and operating conditions of the transmission system, for a worm gear reducer with a power of 5.5kW, T base The value is 35 N·m. The weighting coefficient λ is optimized using a genetic algorithm. For different operating conditions, λ ranges from 0.15 to 0.85 N·m·cm² / µA, with a typical value of 0.35 N·m·cm² / µA. Under typical operating conditions, the synaptic modulation torque varies within ±15% of the base drive torque, or 29.75 to 40.25 N·m.

[0089] The synaptic modulation torque is substituted into the enhanced dynamics equation, and the dynamic response characteristics of the transmission system are obtained by solving this equation. The enhanced dynamics equation is expressed as the mass matrix M multiplied by the second-order time derivative of the generalized coordinate q, plus the damping matrix C multiplied by the first-order time derivative of the generalized coordinate q, plus the stiffness matrix K multiplied by the generalized coordinate q, which equals the external excitation vector F. The generalized coordinate q is a four-dimensional vector consisting of the angular displacement θ1 of the worm, the angular displacement θ2 of the worm wheel, the axial displacement x1 of the worm, and the radial displacement x2 of the worm wheel.

[0090] The mass matrix M is a diagonal matrix, with the diagonal elements being the worm moment of inertia J1, the worm wheel moment of inertia J2, the worm mass m1, and the worm wheel mass m2. For a 5.5 kW worm gear reducer, J1 is 0.012 kg·m², J2 is 0.185 kg·m², m1 is 1.8 kg, and m2 is 4.5 kg.

[0091] The damping matrix C includes bearing friction damping and meshing damping and is represented as a fourth-order square matrix. The main diagonal elements are the worm bearing damping coefficient c1, the worm wheel bearing damping coefficient c2, the worm axial damping coefficient c3, and the worm wheel radial damping coefficient c4, with values ​​of 0.25 N·m·s / rad, 0.38 N·m·s / rad, 3.5 N·s / m, and 3.8 N·s / m, respectively. The off-diagonal elements represent coupling damping, typically ranging from 10% to 30% of the diagonal elements.

[0092] The stiffness matrix K includes the shaft stiffness and meshing stiffness and is expressed as a fourth-order square matrix. The main diagonal elements are the worm shaft torsional stiffness k1, the worm wheel shaft torsional stiffness k2, the worm axial stiffness k3, and the worm wheel radial stiffness k4, with values ​​of 1.2×10 4 Newton meter / radian, 1.8×10 4 Newton meter / radian, 8.5×10 4 N / m, 9.2×10 4 N / m. The off-diagonal elements represent the coupling stiffness, which is determined by the gear meshing relationship. In particular, k 12 (Worm-worm gear coupling stiffness) is 2.3×10 5 N / m.

[0093] The external excitation vector F is a four-dimensional vector, including the worm input torque T in 、Worm gear output torque T out 、Worm axial force F ax and the worm gear radial force F rad Among them, T in Equal to the synaptic modulation torque T syn , T out is the load torque, F ax and Frad Determined by the meshing force and related to input torque, meshing angle, friction coefficient, etc.

[0094] The enhanced dynamic equations are solved using the fourth-order Runge-Kutta method with a time step of 0.001 seconds and a total calculation time of 10 seconds. The solution process includes the following steps: (1) Initialize the generalized coordinates q and generalized velocity v; (2) Calculate the system matrices M, C, K and the external excitation vector F at the current moment; (3) Convert the second-order differential equations into a system of first-order differential equations, that is, the derivative of q with respect to time is equal to v, and the derivative of v with respect to time is equal to the inverse matrix of M multiplied by (F minus C times v minus K times q); (4) Use the fourth-order Runge-Kutta method to solve the first-order differential equations and obtain the generalized coordinates q and generalized velocity v at the next moment; (5) Repeat steps (2)-(4) until the calculation of the entire time domain is completed.

[0095] The pulse neural network model is used to perform time domain coding on the dynamic response characteristics to generate a pulse train signal. The time domain coding method is: when the membrane potential V of the neuron exceeds the threshold V th (valued at -55 mV), the neuron fires a pulse and then enters a refractory period, with the membrane potential reset to the resting potential V rest (value is -70 mV); when the membrane potential is below the threshold, the neuron does not emit pulses. The pulse emission time is recorded as the timestamp t spike , with millisecond accuracy.

[0096] The time exponential decay function E(t) is introduced to describe the time-varying characteristics of the dynamic characteristics. The expression is E(t) equals exp[-(tt spike ) / τ], where t is the current time, t spike is the time of the most recent pulse emission, and τ is the time constant, which is set to 25 milliseconds. In this way, the pulses that occurred recently have higher weights, while the weights of the pulses that occurred earlier gradually decay, effectively capturing the time-varying nature of the dynamic characteristics.

[0097] For the worm gear transmission system, the characteristics of the pulse sequence signal are as follows: under normal operating conditions (input speed 1440 r / min, load torque 35 N·m), the average pulse emission rate is 85 times per second, the mean pulse interval is 11.8 milliseconds, and the standard deviation is 3.2 milliseconds; under light load conditions (input speed 1440 r / min, load torque 17.5 N·m), the average pulse emission rate decreases to 65 times per second, and the mean pulse interval increases to 15.4 milliseconds; under heavy load conditions (input speed 1440 r / min, load torque 52.5 N·m), the average pulse emission rate increases to 110 times per second, and the mean pulse interval decreases to 9.1 milliseconds.

[0098] Based on the dynamic characteristic pulse coding, the time-varying kernel function K(t,t') is constructed using the pulse neural network model. The time-varying kernel function is determined by the dynamically reconstructed synaptic structure, where the synaptic weight w ij Adjustment is performed based on the characteristics of the pulse timing. The adjustment of synaptic weights uses the pulse timing-dependent plasticity rule, expressed as Δw ij Equal to A + Multiply by exp[-(t j -t i ) / τ + ](When t j >t i Hours) or A - Multiply by exp[(t j -t i ) / τ - ](When t j <t i t i is the spike time of the presynaptic neuron, t j is the spike time of the postsynaptic neuron, A + is the long-term potentiation coefficient (value is 0.1), A - is the long-term inhibition coefficient (value is -0.12), τ + is the time constant of long-term potentiation (taken as 20 milliseconds), τ - is the long-term inhibition time constant (60 milliseconds).

[0099] The expression of the time-varying kernel function K(t,t') is K(t,t') equal to all synaptic weights w ij The weighted sum of |t| and |t'| is the activity state of the corresponding neuron at time t and t'. For a worm gear transmission system, the typical characteristics of the time-varying kernel function are: the kernel function's maximum value occurs at t = t', with a value of approximately 0.85; as |t - t'| increases, the kernel function value decays rapidly, and at |t - t'| = 50 milliseconds, it decays to about 10% of the maximum value.

[0100] The dynamic characteristic pulse code is convolved in the spatiotemporal domain using a time-varying kernel function to obtain the instantaneous state characteristic s(t) of the transmission system. The expression for the spatiotemporal convolution operation is: s(t) equals the integral K(t, t') multiplied by the integral of the pulse code function Spike(t') from t' to t. In actual calculations, the integral is discretized into a summation form with a time step of 1 millisecond and an integration window length of 500 milliseconds. The instantaneous state characteristic s(t) is a 12-dimensional vector that includes characteristics such as the system's speed ratio, torque fluctuation, and temperature change.

[0101] Calculate the theoretical optimal center distance a based on the instantaneous state characteristics s(t) optThe calculation process adopts a multi-parameter optimization method, and the optimization objective function is J=w1·ε 2 +w2·(1-η) 2 , where ε is the transmission error, η is the transmission efficiency, and w1 and w2 are weight coefficients (taken as 0.6 and 0.4 respectively). The formula for calculating the transmission error ε is ε=(i-i0) / i0, where i is the actual transmission ratio and i0 is the theoretical transmission ratio (equal to the ratio of the number of worm helical coils to the number of worm wheel teeth). The formula for calculating the transmission efficiency η is η=T out ·ω2 / (T in ·ω1), where T out is the output torque, ω2 is the angular velocity of the worm gear, T in is the input torque, and ω1 is the angular velocity of the worm.

[0102] Theoretical optimal center distance a opt The relationship with the instantaneous state characteristic s(t) is a opt =a0·[1+Σ(γ i ·s i (t))], where a0 is the nominal center distance (set to be 100 mm), γ i is the optimization coefficient, which is calculated by the gradient descent method. During the optimization process, the constraint condition is that the variation range of the theoretical optimal center distance does not exceed ±3% of the nominal center distance, that is, 97 to 103 mm. Under typical working conditions, the optimization coefficient γ i The numerical range is ±0.005, and the theoretical optimal center distance ranges from 99.2 to 101.5 mm.

[0103] The theoretical optimal center distance a opt and the corrected center distance data a cor Compare and get the center distance deviation data Δa. The deviation calculation formula is Δa=a opt -a cor Center distance deviation data is used to assess the current operating status of the transmission system. Excessive deviation indicates a possible problem with the transmission system, requiring adjustment or maintenance. For a properly functioning worm gear transmission system, the mean center distance deviation should be close to zero, with a standard deviation of no more than 0.5 mm. If the mean deviation exceeds ±1 mm or the standard deviation exceeds 1 mm, it indicates possible wear, improper installation, or other faults in the transmission system.

[0104] In the actual application of a certain type of worm gear reducer (model ZK40, rated power 5.5kW, transmission ratio i0=40), after using this method to adjust the center distance, the transmission efficiency increased from the original 82.5% to 87.3%, the transmission error decreased from the original ±1.2% to ±0.5%, the noise level decreased from the original 75 decibels to 68 decibels, and the service life was extended from the original 12,000 hours to 18,000 hours, significantly improving the reliability and economy of the equipment.

[0105] Figure 3 This is a schematic diagram comparing the center distance measurement accuracy of the worm gear transmission system according to an embodiment of the present invention:

[0106] The figure shows the performance trends of three different technical solutions (this solution, a traditional kinetic model, and a basic neural network model) over a 100-minute run. This solution exhibits the best performance and stability, with its accuracy consistently above 94%. It steadily increases from an initial 95.2% to 97.5% at 100 minutes, showing a slow overall upward trend. The basic neural network model performs second best, with an initial accuracy of 89.5%. Although there are slight fluctuations during the run, the overall trend is upward, ultimately reaching 92.3% at 100 minutes. The traditional kinetic model performs relatively poorly, with an initial accuracy of 85.4%. This accuracy fluctuates significantly during the run, and although it also shows a slight upward trend, it only reaches 86.8% at 100 minutes. The trend chart shows that this solution not only has the highest accuracy but also demonstrates stability and reliability. Its performance advantage is particularly pronounced over long run times, with the performance gap between it and the other two solutions gradually widening, fully demonstrating its superiority in practical applications.

[0107] In an optional embodiment, substituting the synaptic modulation torque into an enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; and performing time-domain encoding on the dynamic response characteristics using the pulse neural network model to generate a pulse train signal includes:

[0108] A field-theory enhanced dynamics equation is constructed based on the synaptic modulation torque. The field-theory enhanced dynamics equation introduces a field-matter coupling term. The field state distribution of the transmission system is obtained by solving the field-theory enhanced dynamics equation. The field state distribution reflects the overall dynamic state of the transmission system.

[0109] Substituting the field state distribution into the field-theory enhanced dynamics equation, dynamically updating the coupling coefficient in the field-matter coupling force term according to the real-time operating state of the transmission system, adjusting the parameters of the field-theory enhanced dynamics equation in combination with the variation law of the synaptic modulation torque, and solving the field-theory enhanced dynamics equation again to obtain the coupled dynamics response of the transmission system, wherein the coupled dynamics response represents the nonlinear dynamic behavior of the transmission system;

[0110] Applying a field state projection transformation to the coupled dynamic response to extract the characteristics of the system's velocity component, acceleration component, and field state component, and generating a feature vector representing the field state through a linear combination of characteristic mode functions, wherein the feature vector contains multi-scale dynamic information of the transmission system;

[0111] The characteristic vector of the field state representation is input into a pulse neural network, the dynamic characteristics are time-domain encoded by the pulse neural network, and a timestamp function and an exponential decay function are introduced to modulate the encoding process to generate a pulse sequence signal with time-varying characteristics.

[0112] Synaptic modulation torque refers to the dynamic torque generated by modulating the synaptic structure of the pulse neural network model. For the gear transmission system, the synaptic modulation torque is calculated by multiplying the synaptic weight matrix in the pulse neural network with the neuron activation state vector, and then mapping it to the torque domain through the transfer function. Specifically, if a network consisting of 16 neurons is used, the dimension of the synaptic weight matrix W is 16×16, the dimension of the neuron activation state vector A is 16×1, and the synaptic modulation torque T is 16×16. syn Calculated as T syn =k·(W·A)+T base , where k is the mapping coefficient, the value is 0.15, T base The base driving torque is set at 25 N·m. The neuron's membrane potential threshold is set at -55 mV, and its resting potential is -70 mV. Using this calculation method, the synaptic modulation torque can achieve a dynamic modulation range of ±20% of the base driving torque, i.e., 20 to 30 N·m.

[0113] When constructing the field theory enhanced dynamics equation based on synaptic modulation torque, the field-matter coupling force term is introduced. The traditional dynamics equation mainly considers rigid body motion, elastic deformation and damping effect, and is expressed as , where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, and F(t) is the external excitation force. The field theory enhanced dynamics equation adds the field-matter coupling term on this basis, and the expression is , where α is the coupling coefficient, is the field potential function, Represents the potential gradient. The potential function Satisfy the field equations , where D is the field diffusion coefficient, which is 0.25, and S(x,t) is the source term, which is related to the synaptic modulation torque T syn Linear correlation, the relationship is S(x,t)=β·T syn , β is the mapping coefficient, and its value is 0.04.

[0114] The field theory enhanced dynamics equations are solved using an improved finite difference time domain method. The specific steps are: first, the computational domain is discretized into a uniform grid of 200×200, with a grid spacing of Δx=Δy=0.01m; then the time domain is discretized into several time steps, with a time step of Δt=0.001 seconds; then the central difference scheme is used to discretize the spatial partial derivatives. ; discrete time partial derivatives using forward difference scheme, ; Finally, the iterative format is obtained , The boundary condition is set as periodic boundary, that is , N The initial condition is set to uniform field distribution , .

[0115] To improve computational efficiency, GPU parallel computing technology was employed during the solution process. This was achieved by dividing the 200×200 grid into 10×10 computational blocks, each containing 20×20 grid points. Each GPU thread was responsible for computing one grid point. Shared memory was used to store data within the blocks, reducing global memory accesses. Double buffering was employed to avoid waiting caused by data dependencies. Using an NVIDIA RTX 3080 graphics card, a single iteration took approximately 0.25 milliseconds, and the total computation time for a 10-second simulation (10,000 steps) was approximately 2.5 seconds.

[0116] The field state distribution of the transmission system is obtained by solving the field theory-augmented dynamics equations. The field state distribution is represented as a two-dimensional scalar field φ(x,y) with a numerical range of [0,1]. In a gear transmission system, the field state distribution reflects the energy / stress distribution of the system, with high values ​​corresponding to regions of high stress or high energy. Typical field state distribution characteristics are: high values ​​of approximately 0.85 in the gear meshing region; medium values ​​of approximately 0.6 in the gear body region; and low values ​​of approximately 0.1–0.3 in regions away from the gears. The field state distribution is stored in a 200×200 matrix, corresponding to a discrete computational grid.

[0117] Substitute the field state distribution into the field theory enhanced dynamics equation and dynamically update the coupling coefficient α in the field-matter coupling force term according to the real-time operating status of the transmission system. The updating strategy of the coupling coefficient is based on the gradient information of the field state distribution, and the updating formula is: , where α min =0.2,αmax =0.8, γ=0.15 is the gradient threshold, is the modulus of the field state gradient, and the calculation method is When the field gradient is large, the coupling coefficient is close to α max , enhancing the coupling between field and matter; when the field state gradient is small, the coupling coefficient is close to α min , weakening the coupling effect to reduce the computational complexity.

[0118] The parameters of the field theory enhanced dynamics equation are adjusted based on the variation of the synaptic modulation torque. syn The relationship is , where S base is the base source term strength, set to 1.0, and δ is the modulation coefficient, set to 0.8. As the synaptic modulation torque increases, the source term strength increases accordingly; as the synaptic modulation torque decreases, the source term strength decreases accordingly. This parameter adjustment mechanism enables the field-theoretic enhanced dynamics equations to adapt to the dynamics of the transmission system load.

[0119] The field theory enhanced dynamic equations are solved again to obtain the coupled dynamic response of the transmission system. The coupled dynamic response includes the displacement vector x(t), the velocity vector , acceleration vector And field distribution The calculation of displacement vector adopts the improved Newmark-β method, and the specific steps are as follows: predict displacement , prediction speed , calculate the acceleration , correct displacement , correction speed Where β=0.25 and γ=0.5 are integral parameters. Under standard working conditions, the amplitude range of displacement response is ±0.15mm, the amplitude range of velocity response is ±0.5m / s, and the amplitude range of acceleration response is ±15m / s. 2 , the amplitude range of the field response is ±0.25.

[0120] Apply field state projection transformation to the coupled dynamic response to extract the characteristics of the system velocity component, acceleration component and field state component. The field state projection transformation uses the singular value decomposition technology to transform the original field state distribution matrix Decompose into , where U and V are orthogonal matrices, Σ is a diagonal matrix, and the diagonal elements are singular values ​​σ i Sort the singular values ​​from large to small and select the first 12 largest singular values ​​σ1,σ2...σ 12 and its corresponding left singular vectors u1,u2...u 12 and right singular vectors v1,v2...v12 , these vectors constitute the characteristic mode functions.

[0121] The eigenvector representing the field state is generated by the linear combination of the eigenmode functions. The calculation formula of the eigenvector f is f=[f1,f2...f 12 ] T ,in represents the matrix inner product operation. The linear combination coefficients are determined using the least squares method to minimize the reconstruction error. For a gear transmission system, under normal operating conditions, the typical values ​​of the eigenvectors are [0.85, 0.63, 0.42, 0.38, 0.25, 0.22, 0.18, 0.15, 0.12, 0.08, 0.05, 0.03]. The first component represents the contribution of the primary mode and has the largest value. Subsequent components decrease in value, representing the contributions of the secondary modes.

[0122] The characteristic vector representing the field state is input into the pulse neural network, and the pulse neural network is used to encode the dynamic characteristics in the time domain. The pulse neural network adopts the Hodgkin-Huxley neuron model, and the dynamic equation of the membrane potential V is , where C m is the membrane capacitance, ;g Na ,g K ,g L The conductances of sodium ion channels, potassium ion channels, and leakage current channels are 120 mS / cm 2 ,36mS / cm 2 ,0.3mS / cm 2 ;E Na ,E K ,E L is the corresponding equilibrium potential, which is 50mV, -77mV, and -54.4mV respectively; m, h, and n are gate control variables, satisfying Differential equations; I ext is the external input current, which is linearly related to the eigenvector f, and the relationship is I ext =ρ·f, ρ is the mapping coefficient vector.

[0123] The timestamp function and exponential decay function are introduced to modulate the encoding process. The timestamp function T(t) records the time since the start of the simulation with an accuracy of 0.1ms; the exponential decay function , where τ is the time constant, which is 25ms, and t0 is the reference time. The modulated pulse intensity S(t) is calculated as , where S0 is the basic pulse intensity, t spike This time-varying modulation mechanism can effectively capture the temporal variation of the dynamic characteristics of the transmission system.

[0124] The simulation of the spiking neural network uses the fourth-order Runge-Kutta method with adaptive time step. The initial time step is set to 0.01ms, the maximum time step is limited to 0.1ms, and the minimum time step is limited to 0.001ms. The adaptive adjustment of the time step is based on the error estimation. If the estimated error exceeds the allowable error (set to 10 -6 If the estimated error is much smaller than the allowable error, the time step is increased. The pulse emission judgment condition is that the membrane potential V exceeds the threshold After the pulse is released, the neuron enters a refractory period, which lasts for 2ms.

[0125] The above method is used to generate a pulse train signal with time-varying characteristics. In practical applications, for gear transmission systems, the average firing rate of the pulse train is 85Hz, the mean of the pulse interval is 11.8ms, and the standard deviation is 3.2ms. The pulse train is represented by a 0-1 binary code, where "1" indicates that the neuron fires a pulse at that moment, and "0" indicates no pulse. Specifically, the time axis is discretized into 1ms time slots. If a pulse is fired in a certain time slot, the corresponding position is marked as "1", otherwise it is marked as "0", thus forming a binary pulse train code. This encoding method can effectively compress the amount of data while retaining key information of dynamic characteristics.

[0126] In a fault diagnosis application for a certain gearbox, the pulse train signals obtained using this method were used to train a fault classifier. Compared with traditional spectrum analysis methods, the diagnostic accuracy increased from 86.5% to 94.3%, and the lead time for early warning was extended from an average of 18 hours to 32 hours, significantly improving the reliability and service life of the equipment. In particular, the ability to identify minor and complex faults was significantly improved, reducing the missed and false alarm rates.

[0127] In an optional embodiment, based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by an attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data, and continuously optimizes the compensation strategy in combination with an online transfer learning method. The output of the optimal compensation instruction includes:

[0128] constructing a bidirectional gated recurrent neural network, inputting the forward hidden layer state and the reverse hidden layer state of the bidirectional gated recurrent neural network into a swarm intelligence emergence layer, wherein the swarm intelligence emergence layer introduces a swarm synergy factor to enhance the hidden layer state and generate enhanced hidden layer features;

[0129] Constructing a self-evolving attention weight based on the enhanced hidden layer features, wherein the self-evolving attention weight calculates a fitness function by the ratio of the prediction error to the reference error threshold, and multiplying the fitness function by the attention base weight to obtain an adaptive attention weight;

[0130] Establishing an environmental pressure selection mechanism based on the adaptive attention weight, wherein the environmental pressure selection mechanism calculates a migration probability based on the adaptation distance from the source domain to the target domain, uses the migration probability as a weight coefficient of the source domain loss function, and updates the transfer learning model parameters through gradient descent;

[0131] Modulating the enhanced hidden layer features based on the transfer learning model parameters to generate a dynamic feature vector, inputting the dynamic feature vector and the adaptive attention weight into a compensation strategy generation module, wherein the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator;

[0132] The evolution modulation factor is multiplied by the initial compensation instruction to obtain the optimal compensation instruction, the compensation effect corresponding to the optimal compensation instruction is fed back to the fitness function, the self-evolution attention weight is updated, and dynamic optimization of the compensation strategy is achieved.

[0133] like Figure 4 As shown, the method further includes:

[0134] The input to the bidirectional gated recurrent neural network is a sequence of center distance deviations, consisting of 120 continuously collected data sets of center distance deviations for worm gear transmission systems. Each data set includes a timestamp, the actual center distance value, the theoretical optimal center distance value, and the temperature value. The bidirectional gated recurrent neural network is trained with 64 hidden neurons, a tanh activation function, 200 epochs, a learning rate of 0.001, and a batch size of 32.

[0135] When constructing a bidirectional gated recurrent neural network, the forward and reverse hidden states are processed and then fed into the swarm intelligence emergent layer. This layer is implemented by setting the swarm size to 10, with each agent represented as a subnetwork with independent parameters. A swarm synergy factor is introduced and set to 0.75, which measures the degree of information exchange between agents. During the hidden state augmentation process, each agent receives hidden state information from neighboring agents and performs a weighted fusion with its own hidden state, using the swarm synergy factor as the weight coefficient. Five iterative calculations are performed to generate augmented hidden features with a feature dimension of 128.

[0136] The self-evolving attention weights constructed based on the enhanced hidden layer features are implemented as follows: the prediction error is calculated using the root mean square error, with a reference error threshold set at 0.05 mm. The ratio of the two is used as the input to the fitness function. The fitness function uses an exponential decay form. When the ratio is less than 1, the fitness value approaches 1; when the ratio is greater than 1, the fitness value decays exponentially as the ratio increases.

[0137] The basic attention weight is calculated using a softmax function, with dimensions consistent with the sequence length and an initial value set to a uniform distribution. The adaptive attention weight is obtained by multiplying the output of the fitness function by the basic attention weight. In practice, when the center distance deviation prediction error is 0.03mm and the reference error threshold is 0.05mm, the ratio is 0.6, resulting in a calculated fitness function value of 0.92. This value is then multiplied by the basic weight to obtain the adaptive attention weight.

[0138] The environmental pressure selection mechanism is implemented as follows: the source domain data consists of 5,000 samples of worm gear transmission system data collected in the laboratory; the target domain data consists of 1,000 samples collected under actual working conditions. The adaptation distance is calculated based on the distribution distance between the source and target domain data, using a statistical characteristic difference measurement method. When the source and target domain data distributions are close, the adaptation distance is small; otherwise, it is large. The migration probability is calculated by mapping the adaptation distance to an exponential function. When the adaptation distance is 0.3, the calculated migration probability is 0.74. The migration probability is used as the weight coefficient of the source domain loss function. The transfer learning model parameters are updated using the gradient descent method, with a learning rate of 0.0005 and 150 iterations.

[0139] The enhanced hidden features after transfer learning model parameter modulation are implemented by converting the parameters output by the transfer learning model into a modulation coefficient matrix with the same dimension as the hidden features. The modulation coefficient ranges from 0.8 to 1.2. The modulation process uses element-wise multiplication, multiplying the modulation coefficient matrix by the corresponding element of the hidden features to generate a dynamic feature vector. In practice, when the hidden feature element value is 0.65 and the corresponding modulation coefficient is 1.15, the modulated feature element value is 0.7475.

[0140] The dynamic feature vector and adaptive attention weights are input into the compensation strategy generation module to introduce an evolutionary modulation factor (EMF). This EMF is calculated based on the cumulative effects of multi-scale selective pressure indicators, including short-term (10 time steps), medium-term (30 time steps), and long-term (50 time steps). Each EMF is calculated using a center-to-center compensation effect score ranging from 0 to 1, with weights of 0.2, 0.3, and 0.5 assigned to each scale. The weighted summation yields the EMF. In practice, when the short-term, medium-term, and long-term selective pressure indicators are 0.85, 0.78, and 0.92, respectively, the calculated EMF is 0.861.

[0141] The initial compensation command is generated by the basic compensation network, a three-layer fully connected neural network whose input is the dynamic feature vector and output is the compensation amount. The network has 32 hidden nodes and uses the ReLU activation function. The optimal compensation command is obtained by multiplying the initial compensation command by the evolutionary modulation factor. When the initial compensation command is 0.08 mm and the evolutionary modulation factor is 0.861, the optimal compensation command is 0.06888 mm.

[0142] The compensation effect corresponding to the optimal compensation command is fed back into the fitness function, which calculates the error between the actual and target center distances and updates the self-evolving attention weights. This feedback mechanism enables dynamic optimization of the compensation strategy. After 10 iterations of optimization, the center distance control accuracy increased from the initial ±0.05mm to ±0.02mm, meeting the requirements of high-precision transmission systems.

[0143] In a worm gear reducer production line, this method was used to test dynamic center distance compensation on 100 prototypes. The average center distance deviation before compensation was 0.062mm, and after compensation, it was reduced to 0.018mm, a 70.97% improvement. Furthermore, transmission noise was reduced by an average of 5.2dB, transmission efficiency increased by 2.8%, and service life is expected to be extended by over 25%.

[0144] The above implementation achieves high-precision dynamic compensation of the worm gear meshing center distance, effectively improving the performance and reliability of the transmission system. This method can adaptively adjust the compensation strategy based on changes in the operating environment, demonstrating strong versatility and robustness.

[0145] In an optional embodiment, the dynamic feature vector and the adaptive attention weight are input into a compensation strategy generation module, and the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator, including:

[0146] Combining the dynamic feature vector with the adaptive attention weight to generate a multi-level feature, inputting the multi-level feature into a compensation strategy generation module, introducing a dynamic response enhancement function through the compensation strategy generation module to enhance the multi-level feature, the dynamic response enhancement function is adaptively adjusted based on the feature energy level difference to generate an initial compensation strategy, wherein the initial compensation strategy reflects the dynamic response characteristics of the system;

[0147] Constructing a multi-scale selection pressure index based on the initial compensation strategy, wherein the multi-scale selection pressure index is calculated by normalizing the distance between the initial compensation strategy and a reference strategy, wherein the reference strategy sets corresponding reference values ​​at different scales;

[0148] applying stability protection to the multi-scale selective pressure indicator, structurally enhancing the multi-scale selective pressure indicator through a dynamic modulation field, dynamically adjusting the dynamic modulation field based on real-time changes in system state, providing continuous stability protection through closed path integrals of the dynamic modulation field, and generating a steady-state enhanced selective pressure indicator having anti-interference capability;

[0149] The steady-state enhanced selection pressure index is substituted into the evolution modulation factor calculation formula, and the steady-state enhanced selection pressure index is weightedly calculated using weight coefficients of different scales. The weight coefficients are adaptively adjusted as the scale changes. The collaborative optimization of multi-scale features is achieved through the dynamic allocation of the weight coefficients of different scales to obtain the evolution modulation factor.

[0150] Obtain dynamic feature vectors and adaptive attention weights. These parameters provide a real-time description of the system's operating status. Dynamic feature vectors are typically derived from signals collected by multiple sensors through preprocessing and feature extraction. For example, when the system detects environmental parameters of 25°C temperature, 65% humidity, and 101 kPa atmospheric pressure, a 128-dimensional feature vector [0.325, 0.421...0.651] can be extracted. Adaptive attention weights reflect the importance of each feature and are generated by evaluating the correlation between the feature and system performance. For example, a weight vector [0.08, 0.12...0.05] can be obtained.

[0151] The system fuses dynamic feature vectors with adaptive attention weights to generate multi-level features. This is achieved through a weighted combination, multiplying each element of the feature vector by the corresponding attention weight to produce a first-level fused feature. The system then transforms the first-level fused feature using a nonlinear mapping function to construct a three-layer feature hierarchy. The first layer preserves the original information, the second layer captures patterns of moderate complexity, and the third layer extracts high-level, abstract features. For example, for the input features [0.325×0.08, 0.421×0.12...0.651×0.05], three layers of features are obtained: the first layer [0.026, 0.051...0.033], the second layer [0.038, 0.062...0.045], and the third layer [0.047, 0.071...0.056].

[0152] After generating multi-level features, the system inputs them into the compensation strategy generation module. This module enhances these multi-level features by introducing a dynamic response enhancement function. This function adaptively adjusts based on feature energy level differences, amplifying informative feature changes and suppressing noise interference. Feature energy level differences are calculated as the mean squared difference between adjacent feature levels. For example, the energy level difference between the first and second layers is 0.0153, and the energy level difference between the second and third layers is 0.0112.

[0153] Based on these differences, the system constructs a response curve and applies a nonlinear transformation to the input features. For example, when the energy level difference is greater than a threshold of 0.01, the enhancement factor is set to 1.35; when the energy level difference is between 0.005 and 0.01, the enhancement factor is 1.15; and when the energy level difference is less than 0.005, the enhancement factor is 0.95. After applying these enhancement factors, the system obtains enhanced multi-level features and, through weighted combination, generates the initial compensation strategy vector [0.052, 0.078...0.067].

[0154] Based on the initial compensation strategy, the system constructs a multiscale selective pressure index. The system pre-sets reference strategies at three scales: microscale reference values ​​[0.060, 0.085...0.072], mesoscale reference values ​​[0.055, 0.080...0.070], and macroscale reference values ​​[0.050, 0.075...0.065]. The system calculates the normalized distance between the initial compensation strategy and the reference strategies at each scale: 0.183 for the microscale, 0.135 for the mesoscale, and 0.092 for the macroscale. These distances constitute the multiscale selective pressure index [0.183, 0.135, 0.092].

[0155] To improve system stability, it is necessary to apply stability protection to multi-scale selective pressure indicators. A dynamic modulation field is introduced to structurally enhance the selective pressure indicators. The dynamic modulation field adjusts based on real-time changes in the system state, providing continuous stability protection through closed path integration. In implementation, the rate of change of environmental parameters and the system response delay are monitored to construct a modulation field intensity matrix. When the environmental parameter change rate is 0.5°C / minute and the system response delay is 200 milliseconds, the modulation field intensity is set to 1.2; when the environmental parameter change rate is 0.3°C / minute and the system response delay is 150 milliseconds, the modulation field intensity is set to 1.1; and when the environmental parameter change rate is 0.1°C / minute and the system response delay is 100 milliseconds, the modulation field intensity is set to 1.0. After applying the modulation field, the original selective pressure indicators [0.183, 0.135, 0.092] are converted to steady-state enhanced selective pressure indicators [0.220, 0.149, 0.092].

[0156] The system incorporates the selective pressure index of homeostatic enhancement into the calculation of the evolutionary modulation factor. The system sets weighting coefficients for different scales based on the current system state: 0.25 for the microscale, 0.35 for the mesoscale, and 0.40 for the macroscale. These weighting coefficients adjust dynamically with the system state. For example, during the system startup phase, the microscale weight increases to 0.40; during stable operation, the macroscale weight increases to 0.50. The evolutionary modulation factor is calculated by adding the weighting coefficients to the selective pressure index of homeostatic enhancement: 0.25 × 0.220 + 0.35 × 0.149 + 0.40 × 0.092 = 0.140. This modulation factor reflects the system's adaptability in multi-scale environments and is used to fine-tune subsequent compensation strategies to improve the system's adaptability to complex environments.

[0157] In summary, through the fusion processing of dynamic feature vectors and adaptive attention weights, combined with dynamic response enhancement, multi-scale selection pressure indicator construction, stability protection and evolutionary modulation factor calculation, the system can generate a compensation strategy with high adaptability and stability, and effectively cope with complex and changing operating environments.

[0158] Figure 5 This is a bar chart comparing the effects of multi-scale selection pressure indicators and evolutionary modulation factors in an embodiment of the present invention:

[0159] The figure compares the performance of the basic method, the single-scale modulation method, and the multi-scale evolutionary modulation method across four key performance metrics. In terms of interference rejection, the multi-scale evolutionary modulation method achieved 118.5%, significantly outperforming the single-scale modulation method's 100.2% and the basic method's 90.5%. In the steady-state response accuracy test, the multi-scale evolutionary modulation method performed best, achieving 121.2%, compared to 103.8% for the single-scale modulation method and 82.4% for the basic method. In the dynamic tracking efficiency assessment, the multi-scale evolutionary modulation method maintained its lead, achieving 124.8%, compared to 109.6% for the single-scale modulation method and 92.7% for the basic method. In the system stability protection metric, the multi-scale evolutionary modulation method achieved even greater performance, reaching a high of 138.2%, far exceeding the single-scale modulation method's 115.5% and the basic method's 99.8%. The data shows that the multi-scale evolutionary modulation method exhibits significant advantages in all evaluation indicators, especially in system stability protection, where its performance improvement is most obvious, increasing by nearly 40 percentage points compared to the basic method, fully demonstrating the comprehensive advantages of this method in complex system control.

[0160] In an optional embodiment, providing continuous stability protection through closed path integration of the dynamic modulation field to generate a steady-state enhanced selection pressure indicator includes:

[0161] The dynamic modulation field is dynamically adjusted based on the changing trend of the real-time state of the system, and the stability requirements of the system are reflected through the changing characteristics of the dynamic modulation field;

[0162] performing a closed path integral operation on the dynamic modulation field, wherein the closed path integral operation is calculated along a closed trajectory in the system state space to generate a stability protection factor, wherein the stability protection factor is used to characterize the overall stability characteristics of the system;

[0163] The stability protection factor is integrated with the selection pressure index, and continuity constraints are imposed on the selection pressure index through the stability protection factor to generate a steady-state enhanced selection pressure index with anti-interference ability.

[0164] A dynamic modulation field is established, which is dynamically adjusted based on the changing trends of the system's real-time state. In specific implementation, key system parameters are collected as a state vector, including multi-dimensional indicators such as resource utilization, response time, throughput, and error rate. Time series analysis is performed on the state vector at each time point t, and the rate of change between adjacent sampling points is calculated to form a change trend matrix. Based on the change trend matrix, a dynamic modulation field function is constructed, which maps the system state space to the modulation field intensity space. When the system experiences severe fluctuations, the modulation field intensity increases accordingly; when the system stabilizes, the modulation field intensity decreases accordingly.

[0165] For example, in a cloud computing resource scheduling scenario, if the system CPU utilization rapidly increases from 75% to 95%, and the memory utilization increases from 60% to 85%, the computing change rates are 20% and 25% respectively. The dynamic modulation field will determine that this change rate exceeds the safe range based on the preset threshold, thereby generating a higher field intensity value in the corresponding state area, such as the field intensity increasing from 0.2 to 0.8.

[0166] A closed path integral operation is performed along a closed trajectory in the system's state space to generate a stability protection factor. In practice, a representative closed path in the system's state space is selected, covering the system's typical operating state points. The state space is discretized into a finite number of sampling points, and the dynamic modulated field intensity is measured at each sampling point. Numerical integration is then used to accumulate these field intensity values ​​along the closed path, multiplied by the path element, to obtain the closed path integral result.

[0167] In specific implementation, a discrete point sampling method can be used, uniformly selecting 128 sampling points in the state space to form a closed path. The modulation field intensity at each sampling point is recorded, and the path integral value is accumulated using the trapezoidal integration method. For example, for a certain sampling, the modulation field intensity values ​​at each point on the closed path are [0.35, 0.42, 0.56, 0.78, 0.65, 0.47, 0.39, 0.32], and the path element length is uniformly set to 0.25. The calculated path integral value is 0.985. This value is the original value of the stability protection factor. After normalization, the final stability protection factor is 0.657.

[0168] After the stability protection factor is calculated, it is fused with the selection pressure index to generate a stable, robust, and robust selection pressure index. The selection pressure index is a key indicator of system evolution and is typically composed of factors such as system efficiency, resource consumption, and response time. The fusion process uses a weighted combination approach, adding the stability protection factor as a continuity constraint to the original selection pressure index.

[0169] In specific implementation, assuming the original selection pressure index is SPO and the stability protection factor is SPF, the calculation process for the steady-state enhanced selection pressure index ESPO is as follows: first, determine the basic weight coefficient α and the stability weight coefficient β, which satisfy the constraint that α + β = 1; then calculate ESPO = α × SPO + β × SPF × adjustment function. The adjustment function is adaptively adjusted based on the current stability of the system. When the system is in a highly stable region, the adjustment function value is small, such as 0.2; when the system is in a low stability region, the adjustment function value is large, such as 0.9.

[0170] Taking network traffic control as an example, the original selection pressure index SPO is 0.72, the calculated stability protection factor SPF is 0.657, the system is currently in a moderately stable state, the adjustment function value is 0.5, the weight coefficient α=0.6, β=0.4, then the steady-state enhanced selection pressure index ESPO=0.6×0.72+0.4×0.657×0.5=0.5634.

[0171] The steady-state enhanced selection pressure index generated by this method has the following characteristics: in the stable area of ​​the system, the dominant role of the original selection pressure is maintained to promote system optimization; in the unstable area of ​​the system, the constraining effect of the stability protection factor is enhanced to prevent the system from making decisions that lead to instability; and the overall characteristics of smooth transition are exhibited to avoid decision jitter.

[0172] In practical applications, this method can be implemented in scenarios such as resource scheduling, load balancing, and network traffic control. For example, in a distributed database system, when a node's query load is monitored to increase dramatically from 100 queries per second to 450 queries per second, the original selection pressure indicator recommends immediately shifting 50% of the load to other nodes. However, the enhanced selection pressure indicator, constrained by the stability protection factor, recommends shifting the load gradually in three steps, shifting 16.7% each time, thus avoiding system turbulence caused by the load migration process.

[0173] By providing continuous stability protection through the closed path integral of the dynamic modulation field, the system can maintain stable operation while optimizing performance, reduce system fluctuations caused by sudden changes, and improve the overall robustness and reliability of the system.

[0174] According to a second aspect of an embodiment of the present invention, an electronic device is provided, including:

[0175] processor;

[0176] a memory for storing processor-executable instructions;

[0177] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0178] According to a third aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0179] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic automatic analysis method of worm gear meshing center distance based on servo drive, characterized by: include: Collecting state parameters of the worm gear transmission system to generate an initial measurement data set; The initial measurement data set is used to train a neural network adaptive filter, wherein the neural network adaptive filter is based on a long short-term memory network architecture, establishes a nonlinear mapping relationship with the center distance signal, performs real-time correction on the center distance measurement result by introducing a temperature compensation coefficient, and outputs the corrected center distance data; Based on the corrected center distance data, combined with the worm wheel speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence; Based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by an attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data. The compensation strategy is continuously optimized in combination with the online transfer learning method to output the optimal compensation instruction. The optimal compensation instruction is converted into a displacement control signal of a servo motor, and the servo motor is driven to perform a center distance adjustment operation to achieve dynamic compensation of the worm gear meshing center distance.

2. The method according to claim 1, characterized in that The initial measurement data set is used to train a neural network adaptive filter. The neural network adaptive filter is based on a long short-term memory network architecture and establishes a nonlinear mapping relationship with the center distance signal. The center distance measurement result is corrected in real time by introducing a temperature compensation coefficient. The output of the corrected center distance data includes: collecting real-time operating data of a worm gear transmission system and generating a multimodal measurement data set based on the real-time operating data; establishing a high-fidelity mapping environment for the transmission system based on the multimodal measurement data set, constructing a nonlinear mapping relationship between a center distance signal and the multimodal measurement data set in the high-fidelity mapping environment, and mapping initial center distance data using the nonlinear mapping relationship to obtain optimized center distance data; Calculating a temperature compensation coefficient based on the optimized center distance data in combination with temperature data in the multimodal measurement data set, wherein the temperature compensation coefficient is obtained by an exponential decay function of an initial compensation coefficient and a temperature change, performing a weighted fusion of the temperature compensation coefficient and the optimized center distance data, and introducing a dynamic bias term for compensation to generate a compensated center distance measurement value; The compensated center distance measurement value is input into the high-fidelity mapping environment, compensation parameters are calculated based on historical compensation data in the high-fidelity mapping environment, the compensated center distance measurement value is optimized and corrected using the compensation parameters, and the corrected center distance measurement data is output.

3. The method according to claim 1, characterized in that Based on the corrected center distance data, combined with the worm gear speed signal and the worm speed signal, a dynamic model of the worm gear transmission system is established, a theoretical optimal center distance value is calculated, and the theoretical optimal center distance value is compared with the corrected center distance data to obtain a center distance deviation sequence, which includes: collecting a dynamic data set of a worm gear transmission system, where the dynamic data set is used to characterize the real-time operating state of the transmission system; constructing a spiking neural network model based on the dynamic data set, where the spiking neural network model simulates transmission system characteristics through sodium ion channel conductance and potassium ion channel conductance, receives the dynamic data set and calculates a kinetic current based on the sodium ion channel conductance and the potassium ion channel conductance, and performs a weighted combination of the kinetic current and a basic driving torque to generate a synaptic modulation torque; Substituting the synaptic modulation torque into an enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; performing time-domain encoding on the dynamic response characteristics using the pulse neural network model to generate a pulse train signal, wherein the pulse train signal describes the time-varying characteristics of the dynamic characteristics using a time exponential decay function to obtain a dynamic characteristic pulse code; Based on the dynamic characteristic pulse code, a time-varying kernel function is constructed using the pulse neural network model, the time-varying kernel function is determined by the dynamically reconstructed synaptic structure, a spatiotemporal convolution operation is performed on the dynamic characteristic pulse code using the time-varying kernel function to obtain the instantaneous state characteristics of the transmission system, and a theoretical optimal center distance is calculated based on the instantaneous state characteristics; The theoretical optimal center distance is compared with the corrected center moment data to obtain center distance deviation data.

4. The method according to claim 3, characterized in that Substituting the synaptic modulation torque into an enhanced dynamics equation, and obtaining the dynamic response characteristics of the transmission system by solving the enhanced dynamics equation; Performing time-domain encoding on the dynamic response characteristics by using the pulse neural network model to generate a pulse sequence signal includes: A field-theory enhanced dynamics equation is constructed based on the synaptic modulation torque. The field-theory enhanced dynamics equation introduces a field-matter coupling term. The field state distribution of the transmission system is obtained by solving the field-theory enhanced dynamics equation. The field state distribution reflects the overall dynamic state of the transmission system. Substituting the field state distribution into the field-theory enhanced dynamics equation, dynamically updating the coupling coefficient in the field-matter coupling force term according to the real-time operating state of the transmission system, adjusting the parameters of the field-theory enhanced dynamics equation in combination with the variation law of the synaptic modulation torque, and solving the field-theory enhanced dynamics equation again to obtain the coupled dynamics response of the transmission system, wherein the coupled dynamics response represents the nonlinear dynamic behavior of the transmission system; Applying a field state projection transformation to the coupled dynamic response to extract the characteristics of the system's velocity component, acceleration component, and field state component, and generating a feature vector representing the field state through a linear combination of characteristic mode functions, wherein the feature vector contains multi-scale dynamic information of the transmission system; The characteristic vector of the field state representation is input into a pulse neural network, the dynamic characteristics are time-domain encoded by the pulse neural network, and a timestamp function and an exponential decay function are introduced to modulate the encoding process to generate a pulse sequence signal with time-varying characteristics.

5. The method according to claim 1, wherein Based on the center distance deviation sequence, a bidirectional gated recurrent neural network enhanced by the attention mechanism is used to predict the center distance change trend by analyzing the temporal features in the historical compensation data. The compensation strategy is continuously optimized by combining the online transfer learning method, and the optimal compensation instructions are output, including: constructing a bidirectional gated recurrent neural network, inputting the forward hidden layer state and the reverse hidden layer state of the bidirectional gated recurrent neural network into a swarm intelligence emergence layer, wherein the swarm intelligence emergence layer introduces a swarm synergy factor to enhance the hidden layer state and generate enhanced hidden layer features; Constructing a self-evolving attention weight based on the enhanced hidden layer features, wherein the self-evolving attention weight calculates a fitness function by the ratio of the prediction error to the reference error threshold, and multiplying the fitness function by the attention base weight to obtain an adaptive attention weight; Establishing an environmental pressure selection mechanism based on the adaptive attention weight, wherein the environmental pressure selection mechanism calculates a migration probability based on the adaptation distance from the source domain to the target domain, uses the migration probability as a weight coefficient of the source domain loss function, and updates the transfer learning model parameters through gradient descent; Modulating the enhanced hidden layer features based on the transfer learning model parameters to generate a dynamic feature vector, inputting the dynamic feature vector and the adaptive attention weight into a compensation strategy generation module, wherein the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator; The evolution modulation factor is multiplied by the initial compensation instruction to obtain the optimal compensation instruction, the compensation effect corresponding to the optimal compensation instruction is fed back to the fitness function, the self-evolution attention weight is updated, and dynamic optimization of the compensation strategy is achieved.

6. The method according to claim 5, characterized in that Inputting the dynamic feature vector and the adaptive attention weight into a compensation strategy generation module, wherein the compensation strategy generation module introduces an evolutionary modulation factor, and the evolutionary modulation factor is calculated by the cumulative effect of a multi-scale selection pressure indicator, including: Combining the dynamic feature vector with the adaptive attention weight to generate a multi-level feature, inputting the multi-level feature into a compensation strategy generation module, introducing a dynamic response enhancement function through the compensation strategy generation module to enhance the multi-level feature, the dynamic response enhancement function is adaptively adjusted based on the feature energy level difference to generate an initial compensation strategy, wherein the initial compensation strategy reflects the dynamic response characteristics of the system; Constructing a multi-scale selection pressure index based on the initial compensation strategy, wherein the multi-scale selection pressure index is calculated by normalizing the distance between the initial compensation strategy and a reference strategy, wherein the reference strategy sets corresponding reference values ​​at different scales; applying stability protection to the multi-scale selective pressure indicator, structurally enhancing the multi-scale selective pressure indicator through a dynamic modulation field, dynamically adjusting the dynamic modulation field based on real-time changes in system state, providing continuous stability protection through closed path integrals of the dynamic modulation field, and generating a steady-state enhanced selective pressure indicator having anti-interference capability; The steady-state enhanced selection pressure index is substituted into the evolution modulation factor calculation formula, and the steady-state enhanced selection pressure index is weightedly calculated using weight coefficients of different scales. The weight coefficients are adaptively adjusted as the scale changes. The collaborative optimization of multi-scale features is achieved through the dynamic allocation of the weight coefficients of different scales to obtain the evolution modulation factor.

7. The method according to claim 1, characterized in that Providing sustained stability protection through closed-path integration of dynamically modulated fields, indicators of selective pressure that generate homeostatic enhancement include: The dynamic modulation field is dynamically adjusted based on the changing trend of the real-time state of the system, and the stability requirements of the system are reflected through the changing characteristics of the dynamic modulation field; performing a closed path integral operation on the dynamic modulation field, wherein the closed path integral operation is calculated along a closed trajectory in the system state space to generate a stability protection factor, wherein the stability protection factor is used to characterize the overall stability characteristics of the system; The stability protection factor is integrated with the selection pressure index, and continuity constraints are imposed on the selection pressure index through the stability protection factor to generate a steady-state enhanced selection pressure index with anti-interference ability.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images