Oil film state online identification and early fault early warning method of hydrostatic pressure rotary table

Abstract

The embodiment of the invention provides an oil film state on-line identification and early fault early warning method for a hydrostatic pressure rotary table, and the method comprises the steps: receiving monitoring results of an oil cavity pressure sensor, an oil cavity temperature sensor, an oil cavity flow sensor, an eddy current sensor, a current sensor and a torque sensor, and carrying out the early fault early warning according to the monitoring results of the oil cavity pressure sensor, the oil cavity temperature sensor, the oil cavity flow sensor, the eddy current sensor, the current sensor and the torque sensor; performing signal-level simplification on the monitoring result through a simplified fusion model, and extracting fingerprint features used for representing statistical, frequency-domain and time-domain properties of the monitoring result; fusing the fingerprint features used for representing the statistical, frequency domain and time domain properties of the monitoring result through weighted summation; and according to the fusion result and preset alarm thresholds corresponding to different alarm categories, identifying to obtain a current alarm category. On-line monitoring can be carried out through various types of sensors, and continuous monitoring of oil film state information and early warning of weak faults can be realized through a model according to a monitoring result.

Description

Technical Field

[0001] This application relates to the field of machinery, and in particular to a method for online identification of oil film status and early fault warning of a hydrostatic rotary table. Background Technology

[0002] Hydrostatic rotary tables are widely used in ultra-precision machine tools due to their high motion accuracy, high rigidity, and low frictional resistance. Oil film thickness is particularly important for the safe and stable operation of the machine tool. However, current oil film condition monitoring is still insufficient, leading to the following problems: Oil film condition is a "black box": The oil film is the core load-bearing and damping element of the hydrostatic rotary table. Its key parameters such as thickness, pressure distribution, temperature, and flow rate are difficult to measure directly, comprehensively, and in real time. Traditional monitoring only measures single physical quantities and lacks systematic fault diagnosis methods, resulting in delayed fault warnings; Weak fault signals are easily drowned out: Early faults in hydrostatic rotary tables (such as minor blockages or slight wear) cause extremely weak changes in oil film parameters. Single parameter measurements are easily drowned out by system noise (pumps, valves, environmental vibrations). Strong coupling of multiple physics fields: The oil film state (fluid dynamics), structural vibration (dynamics), and oil / structural temperature (thermodynamics) of the hydrostatic turntable significantly influence each other, making it difficult for a single signal to accurately characterize the overall state. Challenges in state perception under high precision requirements: Hydrostatic turntables are used for ultra-precision machining, where even minute vibrations (micrometer level), temperature rises (0.5°C level), and slight blockages in the oil circuit can affect machining accuracy, requiring highly sensitive perception and diagnostic strategies. Single, indirect sensor data measurements lack the real-time performance and accuracy for predicting oil film conditions in variable-speed motion turntables. Therefore, how to perform online identification of the oil film state and early fault warning during the operation of a hydrostatic turntable has become an urgent problem to be solved. Summary of the Invention

[0003] The purpose of this application is to provide a method for online identification of oil film status and early fault warning of a hydrostatic rotary table, thereby solving the problem of how to perform online identification of oil film status and early fault warning during the operation of a hydrostatic rotary table. The specific technical solution is as follows: In a first aspect, this application provides a method for online identification of oil film status and early fault warning of a hydrostatic turntable. The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; The method includes: The system receives and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, performs signal-level simplification on the monitoring results using a simplified fusion model. It then extracts fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results. The signal-level simplification includes one or more of the following: frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis. The fingerprint features used to characterize the statistical, frequency domain, and time domain properties of the monitoring results are fused by weighted summation; based on the fusion result and the alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

[0004] In one possible implementation, based on the monitoring results from the oil chamber pressure sensor, the eddy current sensor, and the torque sensor, the monitoring results are simplified at the signal level using a simplified fusion model. Fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are extracted, including: The system receives and performs frequency domain analysis and time domain analysis on the monitoring results of the oil chamber pressure sensor to obtain the average pressure, pressure fluctuation value, and pressure amplitude of the oil film; it receives and performs trend analysis and time domain analysis on the monitoring results of the eddy current sensor to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film; and it receives and performs synchronization analysis and time domain analysis on the monitoring results of the torque sensor to obtain the average torque and torque fluctuation.

[0005] In one possible implementation, based on the monitoring results from the oil chamber pressure sensor, the eddy current sensor, and the torque sensor, the monitoring results are simplified at the signal level using a simplified fusion model to extract fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results; these fingerprint features are then fused through a weighted summation, including: The monitoring results are simplified at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation. These are then vectorized to obtain fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation. A preset fusion formula is then used: R = w1 * (P_rms / P_baseline) + w2 * (h_std / h_baseline) + w3 * (T_avg / T_baseline) is used for fusion, where w1, w2 and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the rotational speed baseline value, and R represents the fusion result.

[0006] In one possible implementation, the training process of the simplified fusion model includes: In a high-fidelity virtual model, labeled simulation data is generated to simulate various normal and abnormal operating conditions. Abnormal operating conditions include: throttle blockage, wear of the oil chamber / sealing edge, and oil deterioration. Normal operating conditions include: operating conditions with various speeds and loads. The simulation data is mapped to the [0,1] interval to obtain normalized data. Through a preset time window, the normalized data is truncated to obtain constructed samples. One-hot encoding is used to encode the constructed samples, resulting in a label code for each constructed sample. Each constructed sample is input into the simplified fusion model to be trained to obtain the model's prediction result; based on the label encoding and the prediction result, the corresponding loss is calculated using the cross-entropy loss function; the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model, and the process of inputting each constructed sample into the simplified fusion model to be trained to obtain the model's prediction result is returned to continue training until the preset iteration stopping condition is met, resulting in the trained simplified fusion model; Obtain actual data of the target hydrostatic turntable, wherein the amount of actual data is less than the amount of simulation data; adjust the parameters of the trained simplified fusion model using the actual data to obtain the trained simplified fusion model.

[0007] In one possible implementation, the training process of the simplified fusion model specifically includes: Freeze all network layers except the classifier, train the trained classifier using the actual data to obtain the weights of the frozen encoder, and adjust the model parameters using the weights of the frozen encoder to obtain the initial training model. Unfreeze all network layers, and adjust the parameters of the initial trained model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

[0008] A second aspect of this application provides an online identification and early fault warning device for the oil film state of a hydrostatic turntable, which is applied to the hydrostatic turntable. The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; The early warning module is used to receive and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, simplify the monitoring results at the signal level using a simplified fusion model, extracting fingerprint features to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The signal level simplification includes one or more of frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis of the monitoring results. The fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are then fused using a weighted summation. Based on the fusion result and pre-set alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

[0009] In one possible implementation, the early warning module is specifically configured to receive and perform frequency domain analysis and time domain analysis on the monitoring results of the oil chamber pressure sensor to obtain the average pressure, pressure fluctuation value, and pressure amplitude of the oil film; receive and perform trend analysis and time domain analysis on the monitoring results of the eddy current sensor to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film; and receive and perform synchronization analysis and time domain analysis on the monitoring results of the torque sensor to obtain the average torque and torque fluctuation.

[0010] In one possible implementation, the early warning module is specifically used to simplify the monitoring results at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation. This is then vectorized to obtain fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation. A preset fusion formula is used: R = w1 * (P_rms / P_baseline) + w2 * (h_std / h_baseline) + w3 * (T_avg / T_baseline) is used for fusion, where w1, w2 and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the rotational speed baseline value, and R represents the fusion result.

[0011] In one possible implementation, the training process of the simplified fusion model includes: In a high-fidelity virtual model, labeled simulation data is generated to simulate various normal and abnormal operating conditions. Abnormal operating conditions include: throttle blockage, wear of the oil chamber / sealing edge, and oil deterioration. Normal operating conditions include: operating conditions with various speeds and loads. The simulation data is mapped to the [0,1] interval to obtain normalized data. Through a preset time window, the normalized data is truncated to obtain constructed samples. One-hot encoding is used to encode the constructed samples, resulting in a label code for each constructed sample. Each constructed sample is input into the simplified fusion model to be trained to obtain the model's prediction result; based on the label encoding and the prediction result, the corresponding loss is calculated using the cross-entropy loss function; the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model, and the process of inputting each constructed sample into the simplified fusion model to be trained to obtain the model's prediction result is returned to continue training until the preset iteration stopping condition is met, resulting in the trained simplified fusion model; Obtain actual data of the target hydrostatic turntable, wherein the amount of actual data is less than the amount of simulation data; adjust the parameters of the trained simplified fusion model using the actual data to obtain the trained simplified fusion model.

[0012] In one possible implementation, the training process of the simplified fusion model specifically includes: Freeze all network layers except the classifier, train the trained classifier using the actual data to obtain the weights of the frozen encoder, and adjust the model parameters using the weights of the frozen encoder to obtain the initial training model. Unfreeze all network layers, and adjust the parameters of the initial trained model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

[0013] Another aspect of the application embodiments also provides an electronic device, including: Memory, used to store computer programs; When the processor executes the program stored in the memory, it implements the method for online identification of the oil film status and early fault warning of any of the above-mentioned hydrostatic rotary tables.

[0014] Another aspect of the embodiments of this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for online identification of oil film status and early fault warning of any of the above-described hydrostatic rotary tables.

[0015] In another aspect of the embodiments of this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the above-described method for online identification of oil film status and early fault warning of any of the hydrostatic rotary tables.

[0016] Beneficial effects of the embodiments in this application: This application provides a method for online identification of oil film status and early fault warning of a hydrostatic turntable. The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle. The oil chamber flow sensor is located inside the oil chamber. The eddy current sensor is located on the turntable's radial bearing and thrust bearing, and is used to monitor the oil film thickness. The oil chamber temperature sensor is used to monitor the oil film temperature. The current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes. The method includes: receiving and... Based on the monitoring results from the oil chamber pressure sensor, oil chamber temperature sensor, oil chamber flow sensor, eddy current sensor, current sensor, and torque sensor, a simplified fusion model is used to simplify the monitoring results at the signal level, extracting fingerprint features to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The signal-level simplification includes one or more of frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis of the monitoring results. The fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are then fused using a weighted summation. Based on the fusion result and pre-set alarm thresholds corresponding to different alarm categories, the current alarm category is identified. This solution allows for continuous online acquisition of parameters such as oil film pressure, temperature, flow rate, and thickness using the oil chamber pressure sensor, oil chamber temperature sensor, oil chamber flow sensor, and eddy current sensor. The model then identifies and provides fault warnings based on the acquired oil film monitoring results, enabling continuous monitoring of oil film status information and early warning of minor faults.

[0017] Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram of a hydrostatic rotary table provided in an embodiment of this application; Figure 2 Another schematic diagram of the hydrostatic rotary table provided in the embodiments of this application; Figure 3A flowchart illustrating a method for online identification of oil film status and early fault warning of a hydrostatic rotary table provided in this application embodiment; Figure 4a A schematic diagram of a simplified fusion model provided in an embodiment of this application; Figure 4b A schematic diagram illustrating the training process of the simplified fusion model provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.

[0021] Before introducing the solution in this application, the application scenario will be described first. See [link to relevant documentation]. Figure 1 and Figure 2 The turntable includes a worktable surface, mirror bearings, thrust bearings, a turntable base, a torque motor, a shaft, and a throttle. Monitoring sensors include temperature sensors, pressure sensors, and laser displacement sensors. It can be used to monitor different locations, such as measuring the table surface temperature and motor temperature rise.

[0022] In a first aspect, this application provides a method for online identification of oil film status and early fault warning of a hydrostatic turntable. The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; See Figure 3 The method includes: Step S31: Receive and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, perform signal-level simplification on the monitoring results using a simplified fusion model, and extract fingerprint features to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The signal-level simplification includes one or more of the following: frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis of the monitoring results. Step S32: The fingerprint features used to characterize the statistical, frequency domain, and time domain properties of the monitoring results are fused by weighted summation; based on the fusion result and the alarm thresholds corresponding to different alarm categories set in advance, the current alarm category is identified.

[0023] Specifically, the sensor setup allows for real-time monitoring of the oil film status, providing feedback signals to the closed-loop control system to maintain stable slider operation. This includes precise oil film pressure sensing via miniature high-frequency pressure sensors embedded in key oil chambers / downstream of the throttle to directly measure dynamic pressure pulsations and obtain pressure distribution gradient information. High-precision oil film thickness measurement utilizes non-contact sensors (eddy current type) installed on the rotary table's radial and thrust bearings to directly and accurately measure minute changes in oil film thickness, estimating local dynamic changes and assessing oil film thickness non-uniformity. Oil film status and temperature field monitoring are also implemented; temperature, viscosity, and flow sensors are placed on key rotary table components and oil chambers to accurately assess oil film status. Finally, vibration and acoustic emission are coordinated; a low-frequency vibration accelerometer is installed on the rotary table stator base to monitor overall operational stability and imbalance. A high-frequency acoustic emission sensor is installed near the load-bearing area of ​​the rotary table stator to sensitively detect early damage events such as microbubble rupture and boundary friction initiation within the oil film. Motor drive current / torque monitoring; monitors the fluctuation of turntable drive motor current / torque to reflect load changes and abnormal frictional resistance.

[0024] The solution proposed in this application can continuously collect parameters such as oil film pressure, temperature, flow rate, and thickness online using oil chamber pressure sensors, oil chamber temperature sensors, oil chamber flow sensors, and eddy current sensors. The collected oil film monitoring results can be identified and fault warnings can be given through a model, thereby achieving continuous monitoring of oil film status information and early warning of minor faults.

[0025] In one possible implementation, the monitoring results from the oil cavity pressure sensor, the eddy current sensor, and the torque sensor are received and, based on these results, the monitoring results are simplified at the signal level using a simplified fusion model. Fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are extracted. This includes: receiving and performing frequency domain and time domain analysis on the monitoring results from the oil cavity pressure sensor to obtain the average pressure, pressure fluctuation, and pressure amplitude of the oil film; receiving and performing trend and time domain analysis on the monitoring results from the eddy current sensor to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film; and receiving and performing synchronization and time domain analysis on the monitoring results from the torque sensor to obtain the average torque and torque fluctuation. When using a simplified fusion model for fault early warning, feature engineering and fusion diagnosis can be applied to oil film condition identification. Specifically, this includes: constructing a core feature set for oil film health, including oil film pressure characteristics, oil film thickness characteristics, oil / temperature characteristics, acoustic emission characteristics, and vibration characteristics; and feature fusion and state indicators based on physical mechanisms, utilizing signals such as dynamic pressure, displacement (oil film thickness change), and driving torque, combined with a simplified oil film dynamics model, to reflect core health indicators reflecting the oil film's load-bearing and vibration reduction capabilities. See [link to relevant documentation]. Figure 4aThis simplified fusion model first simplifies the monitored data to obtain easily computable features, then fuses these features to classify the warning level based on the fusion result and preset classification rules. Signal-level simplification (from raw data to core feature set) transforms the lengthy raw signal waveform into "fingerprint" features that characterize its statistical, frequency domain, and time-frequency domain properties. For example, for simplifying oil film pressure, the raw signal can be: the dynamic pressure signal within the oil cavity (10kHz sampling). Simplified extraction methods can include: time domain analysis: calculating mean, standard deviation, peak-to-peak value, and kurtosis; frequency domain analysis: fast Fourier transform to extract specific frequency amplitudes (such as pump frequency, rotation frequency, and their harmonics); nonlinear analysis: approximate entropy, sample entropy, etc. The extracted features can be: P_avg oil film pressure mean, P_rms pressure fluctuation value, and A_pump pressure amplitude. This oil film pressure can reflect the stability of the oil film load and the presence of periodic disturbances. As another example, for simplifying oil film thickness, the raw signal can be: the raw voltage signal collected by the eddy current sensor (inverted to thickness). A simplified extraction method can be: time-domain analysis: mean, standard deviation, minimum value; trend analysis: calculation of rate of change after moving average filtering. The extracted features can be: h_avg oil film thickness, h_std thickness fluctuation, h_min minimum instantaneous thickness. This oil film thickness directly reflects oil film stability; a small h_min indicates a direct risk of rupture. For example, regarding the simplification of motor torque, the original signal can include: turntable motor torque current signal. A simplified extraction method can be: time-domain analysis: mean, fluctuation value; synchronization analysis: synchronizing with the speed signal to calculate torque fluctuation. The extracted feature can be: T_avg motor torque, Tripple representing torque fluctuation. This motor torque directly reflects load changes; abnormal friction will lead to an increase in torque.

[0026] In one possible implementation, the monitoring results from the oil chamber pressure sensor, the eddy current sensor, and the torque sensor are received and, based on these results, the monitoring results are simplified at the signal level using a simplified fusion model to extract fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results. These fingerprint features are then fused through weighted summation, including: simplifying the monitoring results at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation; vectorizing these values ​​to obtain the fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, ...]. [h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation; fusion is performed using a preset fusion formula: R=w1*(P_rms / P_baseline)+w2*(h_std / h_baseline)+w3*(T_avg / T_baseline), where w1, w2, and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the speed baseline value, and R represents the fusion result. Specifically, multiple extracted features are fused into a health indicator with clear physical meaning. For example: Online estimation model for oil film stiffness: Physical model: Stiffness K = Change in force ΔF / Change in displacement Δh; ΔF (force change): proportional to the product of oil cavity pressure change ΔP and oil cavity area A. ΔF≈A*ΔP, Δh (displacement change): directly obtained by measuring the change in oil film thickness using an eddy current sensor. K_estimated=A*(ΔP / Δh); Diagnostic significance: When the estimated stiffness K decreases by more than 15% compared to the healthy baseline value, it indicates a weakening of the oil film's load-bearing capacity, a sign of early wear or leakage. The oil film rupture risk index is a comprehensive indicator constructed based on multiple physical precursor phenomena. It uses a weighted fusion parameter: R = w1*(P_rms / P_baseline) + w2*(h_std / h_baseline) + w3*(T_avg / T_baseline). All features are normalized to their healthy baseline level to eliminate the influence of dimensions. Weights (w1, w2, w3) are assigned according to the sensitivity of each feature to rupture, for example, w1=0.4, w2=0.3, w3=0.3. Weights are determined using expert experience or regression analysis of historical data. Currently: P_rms is 1.8 times the baseline, h_std is 1.5 times the baseline, and T_avg is 2.5 times the baseline.Therefore, R = 0.4 * 1.8 + 0.3 * 1.5 + 0.3 * 2.5 = 1.92.

[0027] In one possible implementation, the training process of the simplified fusion model includes: simulating labeled simulation data of various normal and abnormal operating conditions in a high-fidelity virtual model; wherein, the abnormal operating conditions include: throttle blockage, oil chamber / sealing edge wear, and oil deterioration; the normal operating conditions include: operating conditions with multiple speeds and multiple loads; mapping the simulation data to the [0,1] interval to obtain normalized data; performing sliding truncation on the normalized data through a preset time window to obtain constructed samples; encoding the constructed samples using one-hot encoding to obtain a label code for each constructed sample; and inputting each constructed sample into the simplified fusion model to be trained. The process involves several steps: First, the model's prediction results are obtained. Then, based on the label encoding and the prediction results, the corresponding loss is calculated using the cross-entropy loss function. Next, the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model. The process then returns to the previous step of inputting each constructed sample into the simplified fusion model to obtain the model's prediction results, continuing training until a preset iteration stopping condition is met. Finally, the actual data of the target hydrostatic turntable is obtained, wherein the amount of actual data is less than the amount of simulation data. The parameters of the trained simplified fusion model are adjusted using the actual data to obtain the trained simplified fusion model. In one possible implementation, the training process of the simplified fusion model specifically includes: freezing all network layers except the classifier; training the trained classifier using the actual data to obtain the weights of the frozen encoder; adjusting the model parameters using the frozen encoder weights to obtain an initial training model; unfreezing all network layers; and adjusting the parameters of the initial training model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

[0028] To illustrate the scheme of this application embodiment, the following description is based on a specific training process, including: Digital Twin: In a high-fidelity virtual model, various normal and abnormal operating conditions are simulated to generate massive amounts of precisely labeled "simulation data". Pre-training stage: The general physical laws and fault modes of the hydrostatic turntable are learned. Transfer learning: The pre-trained model is fine-tuned using a small amount of actual data from the target turntable (mainly normal data, supplemented by a very small amount of fault data). The model can be a Transformer, GNN, or a hybrid model that combines temporal feature extraction and classification capabilities. The principle is the same, and the core workflow is: first, the model is pre-trained using simulation data generated by the digital twin; then, it is fine-tuned using small sample data collected in reality, ultimately achieving high-precision fault diagnosis. Input layer: Receives multi-sensor temporal data. Attention mechanism layer: Enables the model to focus on the sensors and time points most relevant to the current fault. For example, when diagnosing throttle blockage, attention is focused on the pressure signal of the oil chamber. Classifier: Classifies based on extracted features. The specific data processing flow includes: Phase 1: Digital Twin Pre-training; a) Data Generation: Set different operating conditions in the digital twin model: speed = [5, 10, 30] rpm, load = [0, 50%, 100%]. Inject different fault modes: fault type = [normal, throttle blockage, oil chamber / sealing edge wear, oil deterioration], severity = [10%, 30%, 50%, 70%] (e.g., blockage area percentage), run the simulation, generate corresponding multi-sensor time-series data, and automatically label them. b) Data Preprocessing and Sample Construction: Normalization: Normalize all sensor data to the [0,1] interval. Sample Construction: Use a 2-second time window as a sample, sliding the selection. Label Encoding: Use One-hot encoding, such as [1,0,0,0] representing "normal", [0,1,0,0] representing "throttle blockage". c) Example of Model Input and Output (Pre-training Phase): Model Input (X_sim): Shape: (Batch size, time step, number of sensor channels); Example: (32, 2000, 8) → 32 samples, each with 2000 time points and 8 sensor channels (e.g., P1, P2, P3, P4, h, I, T, torque). Specific Data: A sample might be a sequence of readings from 8 sensors within 2 seconds simulating "50% load, 30% throttle blockage". Model Output (y_sim_true) & Loss Function: True Label: [0, 1, 0, 0] (representing "throttle blockage"); Model Prediction: [0.02, 0.85, 0.08, 0.05] (the model is 85% certain it is blockage). Loss Calculation: The classification cross-entropy loss function is used to measure the difference between the prediction and the true label, and backpropagation is used to update the model weights. After this phase, the model learns the fault feature mapping relationship in the simulated world.

[0029] See Figure 4bPhase Two: Fine-tuning with Actual Data; a) Data Preparation: Collect target turntable data: A large amount of normal data: for example, 100 hours of normal data collected under different workdays and operating conditions. Very little fault data: for example, only 10 minutes of data before and after two throttle blockage alarms, and data before and after one oil change (as oil deterioration samples). Data preprocessing: Completely consistent with the pre-training phase (using the same normalization parameters). b) Transfer Learning and Fine-tuning: Model initialization: Load the pre-trained model (including the CNN (neural network model) encoder and classifier). Layered fine-tuning strategy: Initial stage: Only retrain the classifier (the last few fully connected layers), and freeze the weights of the CNN encoder. This is because the general features learned by the encoder (such as pressure fluctuation patterns) are still effective, and we only need to adjust the "decision logic" to adapt to the new equipment. Later stage: Unfreeze all layers and perform overall fine-tuning on all actual data with a very small learning rate. The small learning rate is to avoid "forgetting" the previously learned general knowledge and overfitting to a small number of fault samples. c) Example of Model Input and Output (Fine-tuning Phase): Input 1 (X_actual_normal): Data: A 2-second sensor data segment collected during normal operation of the target turntable. Label: [1, 0, 0, 0] (“Normal”); Input 2 (X_actual_fault): Data: Sensor data from 1 minute prior to the actual throttle blockage on the target turntable. Label: [0, 1, 0, 0] (“Throttle Blockage”) A second aspect of this application provides an online identification and early fault warning device for the oil film state of a hydrostatic turntable, which is applied to the hydrostatic turntable. The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; The early warning module is used to receive and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, simplify the monitoring results at the signal level using a simplified fusion model, extracting fingerprint features to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The signal level simplification includes one or more of frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis of the monitoring results. The fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are then fused using a weighted summation. Based on the fusion result and pre-set alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

[0030] In one possible implementation, the early warning module is specifically configured to receive and perform frequency domain analysis and time domain analysis on the monitoring results of the oil chamber pressure sensor to obtain the average pressure, pressure fluctuation value, and pressure amplitude of the oil film; receive and perform trend analysis and time domain analysis on the monitoring results of the eddy current sensor to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film; and receive and perform synchronization analysis and time domain analysis on the monitoring results of the torque sensor to obtain the average torque and torque fluctuation.

[0031] In one possible implementation, the early warning module is specifically used to simplify the monitoring results at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation. This is then vectorized to obtain fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation. A preset fusion formula is used: R = w1 * (P_rms / P_baseline) + w2 * (h_std / h_baseline) + w3 * (T_avg / T_baseline) is used for fusion, where w1, w2 and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the rotational speed baseline value, and R represents the fusion result.

[0032] In one possible implementation, the training process of the simplified fusion model includes: In a high-fidelity virtual model, labeled simulation data is generated to simulate various normal and abnormal operating conditions. Abnormal operating conditions include: throttle blockage, wear of the oil chamber / sealing edge, and oil deterioration. Normal operating conditions include: operating conditions with various speeds and loads. The simulation data is mapped to the [0,1] interval to obtain normalized data. Through a preset time window, the normalized data is truncated to obtain constructed samples. One-hot encoding is used to encode the constructed samples, resulting in a label code for each constructed sample. Each constructed sample is input into the simplified fusion model to be trained to obtain the model's prediction result; based on the label encoding and the prediction result, the corresponding loss is calculated using the cross-entropy loss function; the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model, and the process of inputting each constructed sample into the simplified fusion model to be trained to obtain the model's prediction result is returned to continue training until the preset iteration stopping condition is met, resulting in the trained simplified fusion model; Obtain actual data of the target hydrostatic turntable, wherein the amount of actual data is less than the amount of simulation data; adjust the parameters of the trained simplified fusion model using the actual data to obtain the trained simplified fusion model.

[0033] In one possible implementation, the training process of the simplified fusion model specifically includes: Freeze all network layers except the classifier, train the trained classifier using the actual data to obtain the weights of the frozen encoder, and adjust the model parameters using the weights of the frozen encoder to obtain the initial training model. Unfreeze all network layers, and adjust the parameters of the initial trained model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

[0034] The solution proposed in this application can continuously collect parameters such as oil film pressure, temperature, flow rate, and thickness online using oil chamber pressure sensors, oil chamber temperature sensors, oil chamber flow sensors, and eddy current sensors. The collected oil film monitoring results can be identified and fault warnings can be given through a model, thereby achieving continuous monitoring of oil film status information and early warning of minor faults.

[0035] In another aspect of the embodiments of this application, an electronic device is also provided, see [link to relevant documentation]. Figure 5 ,include: Memory 501 is used to store computer programs; Processor 502, when executing a program stored in memory, implements: The system receives and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, performs signal-level simplification on the monitoring results using a simplified fusion model. It then extracts fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results. The signal-level simplification includes one or more of the following: frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis. The fingerprint features used to characterize the statistical, frequency domain, and time domain properties of the monitoring results are fused by weighted summation; based on the fusion result and the alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

[0036] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0037] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0038] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0039] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0040] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program. When the computer program is executed by a processor, it implements the method for online identification of oil film status and early fault warning of any of the above-mentioned hydrostatic rotary tables.

[0041] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, enables the computer to implement the above-described method for online identification of oil film status and early fault warning of any of the hydrostatic rotary tables.

[0042] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a solid-state drive (SSD), etc.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0044] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. For related parts, please refer to the description of the method embodiments.

[0045] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A method for online identification of oil film status and early fault warning of a hydrostatic rotary table, characterized in that, The hydrostatic turntable includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module. The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; The method includes: The system receives and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, performs signal-level simplification on the monitoring results using a simplified fusion model. It then extracts fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results. The signal-level simplification includes one or more of the following: frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis. The fingerprint features used to characterize the statistical, frequency domain, and time domain properties of the monitoring results are fused by weighted summation; based on the fusion result and the alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

2. The method according to claim 1, characterized in that, Based on the monitoring results from the oil chamber pressure sensor, the eddy current sensor, and the torque sensor, the monitoring results are simplified at the signal level using a simplified fusion model. Fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are extracted, including: The system receives and performs frequency domain analysis and time domain analysis on the monitoring results of the oil chamber pressure sensor to obtain the average pressure, pressure fluctuation value, and pressure amplitude of the oil film; it receives and performs trend analysis and time domain analysis on the monitoring results of the eddy current sensor to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film; and it receives and performs synchronization analysis and time domain analysis on the monitoring results of the torque sensor to obtain the average torque and torque fluctuation.

3. The method according to claim 2, characterized in that, Based on the monitoring results from the oil chamber pressure sensor, the eddy current sensor, and the torque sensor, the monitoring results are simplified at the signal level using a simplified fusion model, and fingerprint features are extracted to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The fingerprint features used to characterize the statistical, frequency domain, and time domain properties of the monitoring results are fused through weighted summation, including: The monitoring results are simplified at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation. These are then vectorized to obtain fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation. A preset fusion formula is then used: R = w1 * (P_rms / P_baseline) + w2 * (h_std / h_baseline) + w3 * (T_avg / T_baseline) is used for fusion, where w1, w2 and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the rotational speed baseline value, and R represents the fusion result.

4. The method according to claim 3, characterized in that, The training process of the simplified fusion model includes: In a high-fidelity virtual model, labeled simulation data is generated to simulate various normal and abnormal operating conditions. Abnormal operating conditions include: throttle blockage, wear of the oil chamber / sealing edge, and oil deterioration. Normal operating conditions include: operating conditions with various speeds and loads. The simulation data is mapped to the [0,1] interval to obtain normalized data. Through a preset time window, the normalized data is truncated to obtain constructed samples. One-hot encoding is used to encode the constructed samples, resulting in a label code for each constructed sample. Each constructed sample is input into the simplified fusion model to be trained to obtain the model's prediction result; based on the label encoding and the prediction result, the corresponding loss is calculated using the cross-entropy loss function; the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model, and the process of inputting each constructed sample into the simplified fusion model to be trained to obtain the model's prediction result is returned to continue training until the preset iteration stopping condition is met, resulting in the trained simplified fusion model; Obtain actual data of the target hydrostatic turntable, wherein the amount of actual data is less than the amount of simulation data; adjust the parameters of the trained simplified fusion model using the actual data to obtain the trained simplified fusion model.

5. The method according to claim 4, characterized in that, The training process of the simplified fusion model specifically includes: Freeze all network layers except the classifier, train the trained classifier using the actual data to obtain the weights of the frozen encoder, and adjust the model parameters using the weights of the frozen encoder to obtain the initial training model. Unfreeze all network layers, and adjust the parameters of the initial trained model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

6. A device for online identification of oil film status and early fault warning of a hydrostatic rotary table, characterized in that, Applied to a hydrostatic rotary table, the hydrostatic rotary table includes: an oil chamber pressure sensor, an oil chamber temperature sensor, an oil chamber flow sensor, an eddy current sensor, a current sensor, a torque sensor, and an early warning module; The oil chamber pressure sensor is located inside the oil chamber / downstream of the throttle; the oil chamber flow sensor is located inside the oil chamber; the eddy current sensor is located on the rotary table radial bearing and thrust bearing, and the eddy current sensor is used to monitor the oil film thickness; the oil chamber temperature sensor is used to monitor the oil film temperature; the current sensor and torque sensor are used to monitor load changes and identify abnormal friction based on the load changes; The early warning module is used to receive and, based on the monitoring results from the oil chamber pressure sensor, the oil chamber temperature sensor, the oil chamber flow sensor, the eddy current sensor, the current sensor, and the torque sensor, simplify the monitoring results at the signal level using a simplified fusion model, extracting fingerprint features to characterize the statistical, frequency domain, and time domain properties of the monitoring results. The signal level simplification includes one or more of frequency domain analysis, time domain analysis, synchronization analysis, and trend analysis of the monitoring results. The fingerprint features characterizing the statistical, frequency domain, and time domain properties of the monitoring results are then fused using a weighted summation. Based on the fusion result and pre-set alarm thresholds corresponding to different alarm categories, the current alarm category is identified.

7. The apparatus according to claim 6, characterized in that, The early warning module is specifically used to receive and perform frequency domain analysis and time domain analysis on the monitoring results of the oil chamber pressure sensor to obtain the average pressure, pressure fluctuation value and pressure amplitude of the oil film; The monitoring results from the eddy current sensor are received and subjected to trend analysis and time-domain analysis to obtain the average thickness, thickness fluctuation, and minimum instantaneous thickness of the oil film. The monitoring results from the torque sensor are received and subjected to synchronous and time-domain analysis to obtain the average torque and torque fluctuation.

8. The apparatus according to claim 7, characterized in that, The early warning module is specifically used to simplify the monitoring results at the signal level to obtain the average oil film pressure, pressure fluctuation value, pressure amplitude, average oil film thickness, thickness fluctuation, minimum instantaneous thickness, average torque, and torque fluctuation. These are then vectorized to obtain fingerprint features [P_avg, P_rms, A_pump, h_avg, h_std, h_min, T_avg, Tripple], where P_avg represents the average oil film pressure, P_rms represents the pressure fluctuation value, A_pump represents the pressure amplitude, h_avg represents the oil film thickness, h_std represents the thickness fluctuation, h_min represents the minimum instantaneous thickness, T_avg represents the motor torque, and Tripple represents the torque fluctuation. A preset fusion formula is used: R = w1 * (P_rms / P_baseline) + w2 * (h_std / h_baseline) + w3 * (T_avg / T_baseline) is used for fusion, where w1, w2 and w3 represent preset weights, P_baseline represents the pressure baseline value, h_baseline represents the thickness baseline value, T_baseline represents the rotational speed baseline value, and R represents the fusion result.

9. The apparatus according to claim 8, characterized in that, The training process of the simplified fusion model includes: In a high-fidelity virtual model, labeled simulation data is generated to simulate various normal and abnormal operating conditions. Abnormal operating conditions include: throttle blockage, wear of the oil chamber / sealing edge, and oil deterioration. Normal operating conditions include: operating conditions with various speeds and loads. The simulation data is mapped to the [0,1] interval to obtain normalized data. Through a preset time window, the normalized data is truncated to obtain constructed samples. One-hot encoding is used to encode the constructed samples, resulting in a label code for each constructed sample. Each constructed sample is input into the simplified fusion model to be trained to obtain the model's prediction result; based on the label encoding and the prediction result, the corresponding loss is calculated using the cross-entropy loss function; the parameters of the simplified fusion model to be trained are adjusted according to the calculated loss to obtain the adjusted model, and the process of inputting each constructed sample into the simplified fusion model to be trained to obtain the model's prediction result is returned to continue training until the preset iteration stopping condition is met, resulting in the trained simplified fusion model; Obtain actual data of the target hydrostatic turntable, wherein the amount of actual data is less than the amount of simulation data; adjust the parameters of the trained simplified fusion model using the actual data to obtain the trained simplified fusion model.

10. The apparatus according to claim 9, characterized in that, The training process of the simplified fusion model specifically includes: Freeze all network layers except the classifier, train the trained classifier using the actual data to obtain the weights of the frozen encoder, and adjust the model parameters using the weights of the frozen encoder to obtain the initial training model. Unfreeze all network layers, and adjust the parameters of the initial trained model using a small learning rate based on the actual data to obtain the trained simplified fusion model.

Images