Supply and recovery method and device for ultrasonic detection coupling liquid of railway wheel

Abstract

The invention provides a supplying and recycling method and device for ultrasonic detection coupling liquid of railway wheels, and belongs to the technical field of ultrasonic detection of railway rolling stock wheels. The supply and recovery method comprises the following steps: acquiring echo signals of ultrasonic detection and working condition data of the coupling liquid supply and recovery device; signal features reflecting the coupling state are extracted based on the echo signals, and coupling quality evaluation indexes are constructed in combination with the working condition data; inputting the coupling quality evaluation index and the working condition data into a trained prediction model, and outputting a liquid supply control parameter and a recovery control parameter adaptive to the current working condition by the prediction model; and adjusting the coupling liquid supply and recovery device according to the coupling quality evaluation index, the liquid supply control parameter and the recovery control parameter. According to the method, the unit consumption of the coupling liquid can be remarkably reduced, and meanwhile, the automation, standardization and stability of the detection process are improved.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic testing technology for railway locomotive and rolling stock wheels, and particularly to a method and apparatus for supplying and recovering coupling fluid for ultrasonic testing of railway wheels. Background Technology

[0002] Railway locomotive and rolling stock wheels are critical load-bearing components ensuring the safe operation of railway locomotives. During service, the tread, flange, and other parts of the wheel are prone to fatigue cracks, peeling, and other damage, requiring inspection using non-destructive testing methods. Ultrasonic testing, due to its sensitivity to internal defects, strong penetrating power, and ease of quantitative assessment, has been widely used in the field of wheel non-destructive testing and is gradually developing towards automated flaw detection to meet the needs of efficient and standardized operation and maintenance in the field.

[0003] In the scenario of automatic ultrasonic inspection of wheels, the probe and the wheel surface need to form a stable liquid film with water or water-based coupling fluid to ensure the echo amplitude and signal-to-noise ratio of the detection signal, thereby ensuring the sensitivity and accuracy of the evaluation results.

[0004] In existing technologies, coupling fluid supply and recovery are generally achieved using a fixed-parameter control method. The supply side typically consists of a water tank, a water pump, pipelines, and valves. After detection begins, the water pump operates continuously, and the solenoid valve remains at a fixed opening or normally open, allowing the coupling fluid to spray continuously from near the probe at a basically constant flow rate, forming a water film coupling layer between the probe and the wheel tread. Existing coupling fluid supply and recovery devices mostly employ a simple method of water collection structure combined with a return pipeline. A water collection shell, guide channel, or water collection tank is set near the probe, and the return fluid is collected by gravity and returned to the water tank through the return water pipe.

[0005] Existing coupling fluid supply and recovery devices struggle to adapt to changing conditions, such as wheel tread condition and environmental factors, with their fixed flow rate and negative pressure water supply and return methods. This can lead to insufficient coupling or fluid waste. Insufficient coupling can cause significant fluctuations in ultrasonic signal amplitude and signal-to-noise ratio, affecting detection accuracy and stability. In practice, to avoid misjudgments and missed detections due to insufficient coupling, the coupling fluid flow rate and negative pressure are often set too high, resulting in excessive redundant spraying. If fluid recovery is not timely, this can lead to significant fluid waste and potentially impact other maintenance stations. Furthermore, different operators often set different flow rates and negative pressures, lacking a unified and objective standard.

[0006] In addition, existing coupling fluid supply and recovery devices mostly use pressure and flow sensors to achieve constant pressure or constant flow water supply, and adjust them by fixed parameters or manual experience. As a result, these technical solutions only keep the water circuit process parameters stable, and it is difficult to take into account the needs of coupling stability and liquid and energy saving. They do not use the ultrasonic detection signal itself to evaluate the coupling state and make closed-loop adjustment.

[0007] In view of this, based on years of experience in production and design in this and related fields, the inventor has designed a method and device for supplying and recovering coupling fluid for ultrasonic testing of railway wheels through repeated experiments, in order to solve the problems existing in the prior art. Summary of the Invention

[0008] The purpose of this invention is to provide a method and apparatus for supplying and recovering coupling fluid for ultrasonic testing of railway wheels, which can adaptively adjust the coupling state and the amount of coupling fluid used to ensure the quality of ultrasonic signals.

[0009] To achieve the above objectives, this invention proposes a method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels, wherein the supply and recovery method includes:

[0010] Acquire echo signals from ultrasonic testing and operating data of the coupling fluid supply and recovery device;

[0011] Based on the echo signal, signal features reflecting the coupling state are extracted, and coupling quality evaluation indicators are constructed in combination with the operating condition data.

[0012] The coupling quality evaluation index and the operating condition data are input into the trained prediction model, and the prediction model outputs liquid supply control parameters and recovery control parameters adapted to the current operating condition.

[0013] The coupling fluid supply to the recovery device is adjusted according to the coupling quality evaluation index, the fluid supply control parameters, and the recovery control parameters.

[0014] This invention also proposes a supply and recovery device for coupling fluid in ultrasonic testing of railway wheels, used to implement the above-mentioned supply and recovery method, wherein the supply and recovery device includes:

[0015] A wheel detection electromechanical probe assembly includes an electromechanical motion mechanism and a probe assembly. The electromechanical motion mechanism drives the probe assembly to scan the wheel tread along a preset trajectory and perform ultrasonic testing on the wheel.

[0016] An ultrasonic testing and data acquisition unit, connected to the probe assembly, is used to acquire the ultrasonic echo signal obtained by the probe assembly;

[0017] The coupling fluid supply and recovery unit is used to supply coupling fluid to the coupling area between the probe assembly and the wheel and to recover the reflux fluid.

[0018] A signal processing and defect assessment unit, connected to the ultrasonic detection and data acquisition unit, is used to process the ultrasonic echo signal and output signal characteristics;

[0019] The intelligent control unit is communicatively connected to the coupling fluid supply and recovery unit, the ultrasonic detection and data acquisition unit, and the signal processing and defect assessment unit, respectively. The intelligent control unit constructs coupling quality evaluation indicators based on the signal characteristics and operating condition data. The intelligent control unit also has a pre-trained prediction model built in, which generates fluid supply and recovery control parameters based on the coupling quality evaluation indicators and operating condition data.

[0020] Compared with the prior art, the present invention has the following features and advantages:

[0021] The present invention proposes a method and apparatus for supplying and recovering coupling fluid for ultrasonic testing of railway wheels. By real-time acquisition of ultrasonic echo signals and coupling fluid supply and recovery data, signal features reflecting coupling quality are extracted and constructed. Based on a random forest intelligent model, supply control parameters and recovery control parameters matching the current operating conditions are automatically generated. Based on the real-time adjustment of the supply control parameters and recovery control parameters, adaptive maintenance of the coupling state and dynamic optimization of coupling fluid usage are achieved. The corresponding supply and recovery device integrates an intelligent control unit, a wheel testing electromechanical probe assembly, an ultrasonic testing and data acquisition unit, and a coupling fluid supply and recovery unit, forming an intelligent supply and recovery device with ultrasonic echo signals as the core feedback. This device can significantly reduce the unit consumption of coupling fluid while ensuring the quality of the detection signal, and at the same time improve the automation, standardization, and stability of the testing process. Attached Figure Description

[0022] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely illustrative to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. Those skilled in the art, guided by the teachings of this invention, can select various possible shapes and proportions to implement the invention according to specific circumstances.

[0023] Figure 1 This is a schematic diagram of the process of an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the offline training process of the random forest of the present invention;

[0025] Figure 3 This is a functional unit structure diagram of the supply recovery device of the present invention;

[0026] Figure 4 This is a structural diagram of the intelligent control unit of the supply and recovery device of the present invention;

[0027] Figure 5 This is a schematic diagram of the supply recovery method of the present invention.

[0028] Explanation of reference numerals in the attached figures

[0029] 1. Electromechanical motion mechanism; 2. Probe assembly; 3. Coupling fluid supply and recovery unit; 4. Intelligent control unit; 401. Data Interface and Acquisition Unit; 402. Feature Construction and Preprocessing Unit; 403. Random Forest Intelligent Decision-Making Unit; 404. Control Constraints and Smoothing Unit; 405. Execution Instruction Generation Unit; 406. Condition detection and fault diagnosis unit; 407. Parameter Management Unit; 5. Ultrasonic testing and data acquisition unit; 6. Signal processing and defect assessment unit; 7. Test results and repair decision output unit. Detailed Implementation

[0030] The details of the present invention can be more clearly understood by referring to the accompanying drawings and the description of specific embodiments. However, the specific embodiments of the present invention described herein are for illustrative purposes only and should not be construed as limiting the invention in any way. Under the teachings of this invention, those skilled in the art can conceive of any possible modifications based on the invention, and these should all be considered to fall within the scope of the invention.

[0031] like Figure 5 As shown, this invention proposes a method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels, wherein the supply and recovery method includes:

[0032] Acquire echo signals from ultrasonic testing and operating data of the coupling fluid supply and recovery device;

[0033] Based on the echo signal, signal features reflecting the coupling state are extracted, and coupling quality evaluation indicators are constructed by combining them with operating condition data;

[0034] The coupled quality evaluation index and operating condition data are input into the trained prediction model, which then outputs liquid supply control parameters and recovery control parameters adapted to the current operating conditions.

[0035] Adjust the supply of coupling fluid to the recovery device according to the coupling quality evaluation index, fluid supply control parameters, and recovery control parameters.

[0036] The proposed method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels utilizes real-time acquisition of ultrasonic echo signals and operating condition data to construct and utilize coupling quality evaluation indicators. Combined with a trained prediction model, it automatically outputs fluid supply control parameters and recovery control parameters adapted to the current operating conditions, and adjusts the coupling fluid supply and recovery device accordingly. This achieves adaptive and stable maintenance of the coupling state and intelligent optimization of coupling fluid usage, ensuring the quality and stability of ultrasonic testing signals while improving the automation and standardization of the testing process.

[0037] The present invention also proposes a supply and recovery device for coupling fluid in ultrasonic testing of railway wheels, used to implement the above-mentioned method, wherein the supply and recovery device includes:

[0038] The wheel detection electromechanical probe assembly 2 includes an electromechanical motion mechanism 1 and a probe assembly 2. The electromechanical motion mechanism 1 drives the probe assembly 2 to scan the wheel tread according to a preset trajectory and perform ultrasonic detection on the wheel.

[0039] The ultrasonic testing and data acquisition unit 5 is connected to the probe assembly 2 and is used to acquire the ultrasonic echo signal obtained by the probe assembly 2.

[0040] The coupling fluid supply and recovery unit 3 is used to supply coupling fluid to the coupling area between the probe assembly 2 and the wheel and to recover the reflux fluid;

[0041] The signal processing and defect assessment unit 6 is connected to the ultrasonic detection and data acquisition unit 5 and is used to process ultrasonic echo signals and output signal characteristics.

[0042] The intelligent control unit 4 is communicatively connected to the coupling fluid supply and recovery unit 3, the ultrasonic detection and data acquisition unit 5, and the signal processing and defect assessment unit 6, respectively. The intelligent control unit 4 constructs coupling quality evaluation indicators based on signal quality characteristics and operating condition data. The intelligent control unit 4 also has a pre-trained prediction model built in, which generates fluid supply and recovery control parameters based on the coupling quality evaluation indicators and operating condition data.

[0043] This invention proposes a supply and recovery device for coupling fluid in ultrasonic testing of railway wheels, such as... Figure 3 As shown, through the collaborative work of each unit, the ultrasonic detection and data acquisition unit 5 and the coupling fluid supply and recovery unit 3 collect signals and operating condition data in real time. After the signal processing and defect assessment unit 6 extracts the signal features, the intelligent control unit 4 constructs coupling quality evaluation indicators and uses the built-in prediction model to adaptively generate fluid supply and recovery control parameters. This allows for real-time adjustment of the coupling fluid supply and recovery unit 3, thereby achieving dynamic stability maintenance of the coupling state and optimized control of the coupling fluid usage. This ensures the stability of the ultrasonic detection signal quality and improves the automation and standardization level of the entire detection process.

[0044] In an optional embodiment of the present invention, the echo signal is a wheel ultrasonic echo signal, and the operating data includes at least the water tank level, the liquid supply flow rate, the liquid supply pressure, the recovery negative pressure, and / or the wheel speed.

[0045] Specifically, wheel ultrasonic echo signal The data is collected by the ultrasonic detection and data acquisition unit 5. Within the same control cycle, the sensor placed in the coupling fluid supply and recovery device acquires the operating condition data, including the water tank level h, the supply flow rate q, the supply pressure pf, the recovery negative pressure pb, and the wheel speed v. The operating condition data is transmitted to the intelligent control unit 4 through the communication interface. The specific operating condition acquisition items and implementation methods are shown in Table 3.

[0046] Table 3:

[0047]

[0048] The system acquires data through ultrasonic waves emitted and received by probe assembly 2 in contact with the wheel. The water tank level is monitored in real-time by a level sensor installed inside the tank. The supply flow rate and pressure are measured by a flow meter and pressure sensor installed in the supply pipeline, respectively. The recovery negative pressure is detected by a negative pressure sensor installed in the recovery pipeline. The wheel speed is acquired by a speed sensor installed on the wheel axle. This operational data is transmitted in real-time to the intelligent control unit 4, which is used to construct coupling quality evaluation indicators and as input parameters for a predictive model. This generates supply and recovery control parameters adapted to the current operational conditions, enabling adaptive adjustment of the coupling fluid supply and recovery.

[0049] In one alternative embodiment of this implementation, the supply recovery method further includes: preprocessing the echo signal and operating condition data, dividing the echo signal into an effective echo window and a noise window, and performing time synchronization on the operating condition data and removing outliers.

[0050] Specifically, the ultrasonic echo signal is windowed, and the effective echo window is... Noise window is This provides a foundation for subsequent feature calculations. The signal processing and windowing parameters are shown in Table 2. Furthermore, to improve data consistency, time alignment and outlier removal were performed on the operating condition data.

[0051] Table 2:

[0052]

[0053] An effective echo window is used to select the area near the first wave to obtain the main reflected signal characteristics, while a noise window is used to select the area without obvious reflected signals for noise estimation. The operating data is time-aligned to keep the data collected by different sensors synchronized in the time dimension. Statistical analysis methods are used to identify and remove outliers that exceed the preset range to ensure the quality of the data input to the prediction model and provide a reliable foundation for constructing accurate coupled quality evaluation indicators.

[0054] In one alternative embodiment of this implementation, the preprocessed echo signal and operating condition data are output to subsequent steps to construct input features for the coupled quality evaluation index and random forest model, as preparation for subsequent feature vector construction.

[0055] In an optional embodiment of the present invention, ultrasonic detection signals and coupling fluid supply and recovery status data are collected simultaneously.

[0056] In an optional embodiment of the present invention, the signal characteristics include at least one of the first echo amplitude, the mean square value within the noise window, and the signal-to-noise ratio.

[0057] Specifically, the amplitude of the first echo is obtained by identifying the peak value of the first obvious reflected wave in the ultrasonic echo signal. The mean square value within the noise window is obtained by performing mean square calculation on the data in the selected period without obvious reflected signals. The signal-to-noise ratio is calculated by dividing the effective signal amplitude by the noise level. These signal characteristics are extracted from the preprocessed ultrasonic echo signal by the signal processing unit and transmitted to the intelligent control unit 4 to construct coupling quality evaluation indicators, providing key input parameters for the prediction model, thereby achieving precise control of the coupling fluid supply and recovery process.

[0058] In one alternative embodiment of this implementation, the formula for extracting signal features includes:

[0059] (1)

[0060] (2)

[0061] (3)

[0062] In the formula, It is a constant, with a value of 10. -12 ~10 -6 ; The amplitude characteristics of the first echo, The mean square value within the noise window, For signal-to-noise ratio, This is the ultrasonic echo signal from the wheel.

[0063] Specifically, formula (1) is used to calculate the amplitude characteristics of the first echo, formula (2) is used to calculate the mean square value within the noise window, and formula (3) is used to calculate the signal-to-noise ratio. After sampling and preprocessing, the amplitude characteristics of the first echo are obtained by identifying the peak value of the first obvious reflected wave in the ultrasonic signal and combining it with a constant for standardization. The mean square value within the noise window is obtained by averaging the data in the selected no-signal period and then taking the square root. The signal-to-noise ratio is obtained by dividing the amplitude characteristics of the first echo by the mean square value within the noise window. These signal characteristics are transmitted to the intelligent control unit 4 in real time as key elements for constructing the coupling quality evaluation index, providing accurate input for the prediction model, thereby realizing the adaptive adjustment of the coupling fluid supply and recovery parameters.

[0064] In one optional example, the coupling quality evaluation metric is constructed based on signal characteristics:

[0065] (4)

[0066] In the formula, A min A max SNR min SNR maxParameters set for experience; The amplitude characteristics of the first echo, The mean square value within the noise window, This refers to the signal-to-noise ratio.

[0067] Specifically, in formula (4) A min A max SNR min SNR max The parameters are set empirically, determined experimentally based on the actual testing environment and wheel type, and include the characteristics of the first echo amplitude. Mean square value within the noise window Signal-to-noise ratio As input variables, the intelligent control unit 4 receives these signal features extracted from the signal processing unit and calculates the coupling quality evaluation index value in real time according to a preset formula. This index value quantitatively represents the quality of the current coupling state. When the index value is close to 1, it indicates a good coupling state, and when it is close to 0, it indicates a poor coupling state. The parameters are transmitted as key input parameters to the prediction model to generate liquid supply control parameters and recovery control parameters adapted to the current operating conditions, thereby achieving precise regulation of the coupling fluid supply and recovery.

[0068] In an optional example, in coupling quality evaluation metrics After construction, it is necessary to prepare for coupling state discrimination and to use the calculated coupling quality evaluation index. The output is used in subsequent steps to control parameter decisions and threshold determination.

[0069] In an optional embodiment of the present invention, quality evaluation indicators are coupled. This is used to characterize the current coupling state between the probe and the wheel.

[0070] In an optional embodiment of the present invention, the prediction model employs a random forest model.

[0071] Specifically, a random forest model is used, coupled with quality evaluation indicators and operating condition data as input features. By integrating the inference results of multiple decision trees, the liquid supply control parameters and recovery control parameters adapted to the current operating conditions are output. During online monitoring, the intelligent control unit 4 directly calls this solidified model for real-time inference, thereby realizing data-driven intelligent decision-making.

[0072] In one alternative example of this implementation, the random forest model is obtained through offline training, which includes:

[0073] Collect historical test data and construct training samples containing coupled quality feature vectors and corresponding liquid supply and recovery control parameter vectors;

[0074] Preprocess the data in the training samples;

[0075] Extract signal quality features and combine them with operating parameters to construct a feature vector for training;

[0076] The dataset is divided into a training set, a validation set, and a test set;

[0077] A random forest regression model is trained using the training set, and the model parameters are optimized using the validation set.

[0078] Use a test set to evaluate model performance. Once the evaluation is passed, solidify the model parameters and deploy the system.

[0079] Specifically, based on the coupling quality evaluation results obtained from feature extraction and coupling quality assessment, as well as the coupling fluid supply and recovery operation data, a random forest model is used to output control parameters for fluid supply and recovery. The random forest intelligent decision-making unit 403 inputs the feature vectors into the pre-trained random forest regression model and outputs control suggestions. .

[0080] First, the intelligent control unit 4 receives ultrasonic signal quality characteristics from the signal processing and defect assessment unit 6, including the initial amplitude, signal-to-noise ratio, and the coupling quality evaluation index constructed therefrom. Simultaneously, it receives operating data collected by sensors in the liquid supply pipeline, recovery pipeline, and water tank assembly, including water tank level, liquid supply flow rate, liquid supply pressure, recovery negative pressure, and wheel speed.

[0081] Subsequently, the intelligent control unit 4 constructs and normalizes the signal quality characteristics and operating condition data to form feature vectors for input to the random forest model. These feature vectors are then input into the pre-trained and fixed random forest model, which integrates the outputs of multiple decision trees to obtain the liquid supply control parameters and recovery control parameters under the current operating conditions. These control parameters characterize the liquid supply intensity, recovery capacity, or their corresponding target setpoints. The random forest model needs to be trained offline using historical samples before the equipment is put into operation. The trained model parameters are stored in the intelligent control unit 4 and directly accessed during online detection to output the liquid supply and recovery control parameters in real time.

[0082] The random forest model is trained by minimizing the error between the predicted control parameters and the target control parameters. The objective function is:

[0083]

[0084] In the formula, This is the normalized coupling quality feature vector; These are the corresponding liquid supply and recovery control parameters; The parameters are determined by minimizing the coupling quality error cost function, obtained through offline training. The training process of the random forest model is as follows:

[0085]

[0086]

[0087]

[0088] in, For random forest regressors, This refers to the speed of the liquid supply pump; Valve opening degree; To recover negative pressure, the above method enables data-driven decision-making based on ultrasonic signal quality and operating condition information, allowing the supply and recovery control parameters to adaptively adjust according to changes in operating conditions. Simultaneously, a complete offline training process covering data acquisition, preprocessing, feature engineering, model training, validation optimization, and testing evaluation is implemented, enabling the construction of accurate and reliable random forest prediction models using historical data.

[0089] In an optional example, such as Figure 2 As shown, the offline training process of the random forest structure used to output the control parameters for coupling fluid supply and recovery includes:

[0090] Step T1: Training data acquisition and sample construction.

[0091] During the equipment commissioning phase, historical test records, or manual experience control phase, collect ultrasonic test data of the wheels and corresponding coupling fluid supply and recovery condition data.

[0092] Training samples are constructed based on the data. Each sample includes input features and target output. The training sample format is as follows:

[0093]

[0094] in, For coupling quality feature vectors, These are the corresponding liquid supply and recovery control parameters.

[0095] Step T2: Training Data Preprocessing. The ultrasonic signals and operational data in the training samples are preprocessed, including windowing, time alignment of operational data, and feature scale normalization.

[0096] Step T3: Feature Extraction and Training Feature Vector Construction. Based on the preprocessed ultrasonic signal, signal quality features such as initial wave amplitude, echo energy, and signal-to-noise ratio are extracted. Operating parameters such as fluid supply flow rate, recovery negative pressure, and water tank level are used as auxiliary features to construct a training feature vector. .

[0097] Step T4: The preprocessed dataset is randomly divided into training set, test set, and validation set at 70%, 20%, and 10% respectively.

[0098] Step T5: Random Forest Model Training and Parameter Optimization. A random forest regression model is constructed using the training set samples from Step T4, and the model parameters are adjusted and optimized using the validation set.

[0099] Random forests employ a single-model, multiple-output approach to simultaneously output control parameter vectors for both liquid supply and recovery.

[0100]

[0101] In the formula, For the speed of the liquid supply pump, For valve opening, To recover negative pressure.

[0102] For a given sample set S of a node in a decision tree, the node output is the mean of the target output vector within that node:

[0103]

[0104] The node splitting of the regression decision tree uses a weighted squared loss function of the multi-output prediction error:

[0105]

[0106] in, Let be the weight matrix. For candidate splits... The generated left and right child nodes and , choose to The features and thresholds are used to complete node splitting. The outputs of multiple decision trees are ensembled to obtain the final output of the random forest model.

[0107] Step T6: Model evaluation, consolidation, and deployment.

[0108] The final random forest model obtained in step T5 is used to make predictions on the test set to obtain the predicted values ​​of the control parameters corresponding to the test set samples. and compared with the actual control parameters of the test set. A comparative evaluation was conducted. Evaluation metrics for the test set included mean squared error, mean absolute error, or root mean square error. The parameters for this random forest model were set as follows: number of decision trees: 200; maximum depth: 12; minimum number of samples per leaf node: 5; minimum number of split samples: 15; and feature subsampling ratio: 0.6.

[0109] Once the test set evaluation is passed, the structure and parameters of the random forest model are fixed and saved, and the fixed model is stored in the parameter management unit 407.

[0110] The solidified random forest model is deployed to the intelligent control unit 4. During the deployment process, the model inference interface is encapsulated so that its input is the real-time feature vector constructed in the online operation process. The output is a vector of liquid supply and recovery control parameters. .

[0111] In an optional embodiment of the present invention, the prediction model may also employ a linear regression model, a nonlinear regression model, a support vector regression model, a neural network model, a gradient boosting regression model, or a k-nearest neighbor regression model.

[0112] Specifically, the random forest model can be replaced by other data-driven models, such as linear regression, nonlinear regression, support vector regression, neural network models, gradient boosting regression, and k-nearest neighbor regression. The input is a coupled quality feature vector, and the output is the liquid supply and recovery control parameters. By explicitly defining the prediction model, various models, including linear regression, neural networks, and gradient boosting regression, can be adopted, providing flexible technical choices for specific implementations. Different models have different computational complexities, learning capabilities, and applicable scenarios. The most suitable model can be selected based on actual computing power, accuracy requirements, or data characteristics to achieve intelligent decision-making, thereby enhancing the applicability and feasibility of the technical solution while ensuring core functionality.

[0113] In an optional embodiment of the present invention, adjusting the coupling fluid supply and recovery device according to coupling quality evaluation indicators, fluid supply control parameters, and recovery control parameters includes:

[0114] When the coupling quality evaluation index is lower than the coupling quality threshold within a consecutive preset period, it is determined that there is an insufficient coupling state, and liquid supply enhancement or recovery adjustment control is triggered; when the coupling quality evaluation index does not meet the above continuous judgment conditions, the current control state is maintained.

[0115] Specifically, the intelligent control unit 4 determines thresholds, limits, and executes control over the liquid supply enhancement or recovery parameters. The intelligent control unit 4 operates according to the continuous control cycle. Internal Coupling Quality Evaluation Indicators Make a judgment: when Continuous preset number of cycles Below the coupling quality threshold When the coupling is insufficient, the system determines that a coupling deficiency exists and triggers liquid supply enhancement or recovery adjustment control; when the coupling quality evaluation index... If the above continuous judgment conditions are not met, the current control state is maintained. By introducing judgment logic based on continuous periodic monitoring and threshold comparison, stable and interference-resistant monitoring of the coupling state is achieved. This method only triggers adjustments when the trend of insufficient coupling persists, ensuring that the system can respond promptly to the actual deterioration of the coupling state, and effectively preventing frequent oscillations of control commands due to accidental interference, thereby improving the accuracy of the entire supply recovery process control and the stability of the device operation.

[0116] In one optional embodiment of this implementation, the detection termination and fault judgment cycle is performed based on the detection process status and coupling fluid supply and recovery conditions. The system determines the detection termination conditions and abnormal states, and executes corresponding termination control or abnormal handling operations. The status detection and fault diagnosis unit 406 performs abnormality identification based on operating condition data.

[0117] In an optional example, if the detection has not ended and no fault has been triggered, return to the previous step and proceed to the next control cycle.

[0118] In an optional example, the system enters a shutdown state when the detection ends; and enters a protection state when coupling is insufficient, recording abnormal information. Insufficient coupling can be determined by the following formula:

[0119]

[0120] This is a preset coupling quality threshold; The number of consecutive determination periods. The control cycle can be set to 3 to 10 cycles, depending on the control cycle and the dynamic characteristics of the device.

[0121] In an optional example, after the insufficient coupling anomaly is eliminated or the detection process is completed, the supply recovery device performs one of the following operations based on the operating status: if the anomaly is eliminated but the detection task is not completed, it returns to the step of acquiring the echo signal of the ultrasonic detection to continue the detection process; if the detection task is completed or the anomaly cannot be recovered, it exits the detection process and enters a standby or shutdown state.

[0122] In an optional embodiment of the present invention, adjusting the coupling fluid supply and recovery device according to coupling quality evaluation indicators, fluid supply control parameters, and recovery control parameters includes:

[0123] Threshold determination, amplitude limiting and smoothing are performed on the liquid supply control parameters and recovery control parameters. Execution instructions are generated and sent to the actuators of the coupling liquid supply and recovery device to adjust the liquid supply pump speed, valve opening and recovery negative pressure.

[0124] Specifically, to avoid instability in the supply and recovery device caused by sudden changes in control parameters, the liquid supply control parameters and recovery control parameters need to undergo amplitude limiting and smoothing processing before execution to keep their change amplitude or rate within a preset safe range. Control constraint and smoothing unit 404 Limiting and smoothing

[0125]

[0126]

[0127] The processed control parameters are sent by the execution command generation unit 405 to actuators such as the liquid supply pump, valves, and vacuum pump, completing the closed-loop regulation of the coupling fluid supply and recovery process. The execution command generation unit 405 will... The commands are converted into commands for the liquid supply pump speed, solenoid valve opening degree, and recovery negative pressure, and then sent to the coupling fluid supply and recovery unit 3 for execution.

[0128] By using threshold determination and execution control steps, the system can suppress false judgments of transient disturbances while simultaneously maintaining stable coupling states and intelligently optimizing the amount of coupling fluid. Control constraints and smoothing parameters are shown in Table 4.

[0129] Table 4:

[0130]

[0131] By applying threshold judgment, amplitude limiting, and smoothing processing to the control parameters, it is effectively ensured that the commands output to the actuator are always within a safe and reasonable range, preventing abnormal situations such as equipment overload and parameter exceeding limits. Smoothing processing avoids drastic changes in control commands, making the adjustment process of the liquid supply and recovery device more stable and continuous, reducing mechanical and hydraulic shocks, and thus improving the overall stability, reliability, and service life of the supply and recovery device.

[0132] In an optional embodiment of the present invention, the supply recovery method further includes: device startup and device initialization.

[0133] In an optional example, the intelligent control unit 4 is activated, establishing a communication connection with the ultrasonic detection and data acquisition unit 5, the coupling fluid supply and recovery unit 3, the signal processing and defect assessment unit 6, and the detection result output unit 7. During the communication establishment process, the supply and recovery device detects the connection status of each communication interface, and proceeds to the initialization step after confirming that the interface communication is normal.

[0134] In an optional example, based on the requirements of the detection task, the parameter management unit 407 loads the detection operation mode and task parameters, including the control cycle. Coupling quality threshold Control the upper and lower limits of the quantity.

[0135] The control variable is defined as:

[0136]

[0137] In the formula, These are the corresponding liquid supply and recovery control parameters; For the speed of the liquid supply pump, For valve opening, To recover negative pressure, the control operating parameters are shown in Table 1.

[0138] Table 1:

[0139]

[0140] In an optional example, the intelligent control unit 4 loads the pre-trained and fixed parameters of the random forest model and initializes the model state. The random forest model enters the inference state for subsequent processes. The intelligent control unit 4 sends initial control parameters to the coupling fluid supply and recovery unit 3. After completing the above initialization operations, the device enters the detection preparation state and issues a start command to subsequent steps to begin the acquisition of ultrasonic signals and operating condition data.

[0141] In an optional embodiment of the present invention, the electromechanical motion mechanism 1 is used to drive the probe assembly 2 to scan the wheel tread according to a preset trajectory; the probe assembly 2 is used to scan the wheel tread.

[0142] In an optional embodiment of the present invention, the ultrasonic detection and data acquisition unit 5 is connected to the probe assembly 2 for acquiring echo signals.

[0143] In an optional embodiment of the present invention, the coupling fluid supply and recovery unit 3 is used to supply coupling fluid to the probe-wheel coupling area and recover the refluxed coupling fluid.

[0144] In an optional embodiment of the present invention, the signal processing and defect assessment unit 6 is connected to the data acquisition unit 5 and is used to process the echo signal and output the defect assessment result to the detection result and repair decision output unit 7.

[0145] In an optional embodiment of the present invention, the intelligent control unit 4 is connected to the coupling fluid supply and recovery unit 3, the data acquisition unit 5, and the defect assessment unit 6 respectively, and is used to fuse signal quality and operating condition information and output fluid supply and recovery control commands to form a closed-loop regulation.

[0146] In one alternative example of this implementation, such as Figure 4As shown, the intelligent control unit 4 includes: a data interface and acquisition unit 401, a feature construction and preprocessing unit 402, a random forest intelligent decision-making unit 403, a control constraint and smoothing unit 404, an execution instruction generation unit 405, a status detection and fault diagnosis unit 406, and a parameter management unit 407.

[0147] In an optional example, the parameter management unit 407 is used to store thresholds, upper and lower limits, control cycles, and model parameters, etc.; the data interface and acquisition unit 401 is used to receive signal quality information and operating condition data.

[0148] In an optional example, the feature construction and preprocessing unit 402 is used to construct coupling quality features.

[0149] In an optional example, the random forest intelligent decision unit 403 is used to output liquid supply and recovery control parameters.

[0150] In an optional example, the control constraint and smoothing unit 404 is used for amplitude limiting, smoothing, and rate limiting.

[0151] In an optional example, the instruction generation unit 405 is used to generate executable control instructions for the coupling fluid supply and recovery unit 3.

[0152] In an optional example, the status detection and fault diagnosis unit 406 is used to identify anomalies and trigger protection strategies.

[0153] In an optional embodiment of the present invention, the supply recovery device further includes: a detection result and refinishing decision output unit 7, which is used to receive the defect assessment result of the signal processing and defect assessment unit 6, and output the final detection report and decision suggestion on whether the wheel needs to be refinished.

[0154] Example

[0155] The following is a detailed description of the specific implementation process of the method and apparatus for supplying and recovering coupling fluid for ultrasonic testing of railway wheels proposed in this invention, with reference to an embodiment.

[0156] To achieve adaptive maintenance of the coupling state and intelligent optimization of the coupling fluid dosage, the following methods are adopted: Figure 1 The method flow shown is as follows. The method flow includes: S1 Device startup and device initialization; S2 Acquisition of ultrasonic signals and operating data; S3 Feature extraction and coupling quality evaluation; S4 Random forest model output of liquid supply and recovery control parameters; S5 Control parameter limiting and execution control; S6 Detection end, fault diagnosis and loop.

[0157] S1 Device startup and device initialization.

[0158] The intelligent control unit 4 is activated, establishing communication connections with the ultrasonic detection and data acquisition unit 5, the coupling fluid supply and recovery unit 3, the signal processing and defect assessment unit 6, and the detection result output unit 7. During the communication establishment process, the device checks the connection status of each communication interface, and proceeds to the initialization step after confirming that the interface communication is normal.

[0159] Based on the requirements of the detection task, the parameter management unit 407 loads the detection operation mode and task parameters, including the control cycle. Coupling quality threshold Control the upper and lower limits of the quantity.

[0160] Control quantity is defined as

[0161]

[0162] In the formula, These are the corresponding liquid supply and recovery control parameters; For the speed of the liquid supply pump, For valve opening, To recover negative pressure, the control operating parameters are shown in Table 1.

[0163] Table 1:

[0164]

[0165] The intelligent control unit 4 loads the pre-trained and fixed parameters of the random forest model and initializes the model state. The random forest model enters the inference state for subsequent processes. The intelligent control unit 4 sends initial control parameters to the coupling fluid supply and recovery unit 3. After completing the above initialization operations, the supply and recovery device enters the detection preparation state and sends a start command to the subsequent step S2 to begin the acquisition of ultrasonic signals and operating condition data.

[0166] S2 acquires ultrasonic signals and operating data.

[0167] The echo signal from the ultrasonic test and the data on the supply and recovery of the coupling fluid are collected simultaneously, and the collected data are preprocessed to improve the reliability of subsequent coupling quality evaluation and control decisions.

[0168] S2-1 Echo Signal Acquisition. The echo signal from the ultrasonic waves of the wheel was acquired using the ultrasonic testing and data acquisition unit 5. .

[0169] S2-2 Echo Signal Preprocessing. The echo signal is windowed, with the effective echo window being... Noise window is This provides a foundation for subsequent feature calculations. Signal processing and windowing parameters are shown in Table 2.

[0170] Table 2:

[0171]

[0172] S2-3 Operating condition data acquisition.

[0173] Within the same control cycle, operating data is acquired using sensors placed in the coupling fluid supply and recovery device. This data includes the water tank level h, the fluid supply flow rate q, and the fluid supply pressure p. f , recover negative pressure p b Wheel speed v and operating condition data are transmitted to the intelligent control unit 4 via a communication interface. The data acquisition items and implementation methods are shown in Table 3.

[0174] Table 3:

[0175]

[0176] S2-4 Operating condition data preprocessing. To improve data consistency, time alignment and outlier removal are performed on the operating condition data.

[0177] S2-5 Feature Vector Construction Preparation. The preprocessed ultrasonic signals and working condition data are output to the subsequent step S3 to construct the input features for the coupled quality evaluation index and random forest model.

[0178] S3 Feature Extraction and Coupling Quality Evaluation. Based on the ultrasonic echo signals and operational data obtained through S2 preprocessing, signal features reflecting the ultrasonic coupling state are extracted, and coupling quality evaluation indices are constructed. This is used to characterize the current coupling state between the probe and the wheel.

[0179] S3-1 Ultrasonic Signal Feature Extraction. Using the signal processing and defect assessment unit 6, the amplitude characteristics of the first echo of the ultrasonic signal are extracted. Mean square value within the noise window and signal-to-noise ratio .

[0180]

[0181]

[0182]

[0183] In the formula, It is a constant, with a value ranging from 10⁻¹² to 10⁻⁶.

[0184] S3-2 Coupling Quality Evaluation Indicators Construction. Based on the above signal characteristics and optional operating condition characteristics, a coupled quality evaluation index is constructed using the feature construction and preprocessing unit 402. It is used to comprehensively characterize the coupling state within the current detection cycle.

[0185]

[0186] in, A min A max SNR min SNR max Parameters set for experience.

[0187] S3-3 Coupling State Determination Preparation

[0188] The calculated coupling quality evaluation index The output is sent to the subsequent step S4 for control parameter decision-making and threshold determination.

[0189] The S4 random forest model outputs liquid supply and recovery control parameters.

[0190] Based on the coupling quality evaluation results obtained in step S3 and the coupling fluid supply and recovery operation data, the random forest model is used to output the fluid supply and recovery control parameters. The random forest intelligent decision unit 403 inputs the feature vector into the pre-trained random forest regression model and outputs control suggestions. .

[0191] First, the intelligent control unit 4 receives ultrasonic signal quality characteristics from the signal processing and defect assessment unit 6, including the initial amplitude, signal-to-noise ratio, and the coupling quality evaluation index constructed therefrom. Simultaneously, it receives operating data collected by sensors in the liquid supply pipeline, recovery pipeline, and water tank assembly, including water tank level, liquid supply flow rate, liquid supply pressure, recovery negative pressure, and wheel speed.

[0192] Subsequently, the intelligent control unit 4 constructs and normalizes the signal quality characteristics and operating condition data to form feature vectors for input to the random forest model. These feature vectors are then input into the pre-trained and fixed random forest model, which integrates the outputs of multiple decision trees to obtain the liquid supply control parameters and recovery control parameters under the current operating conditions. These control parameters characterize the liquid supply intensity, recovery capacity, or their corresponding target setpoints. The random forest model needs to be trained offline using historical samples before the equipment is put into operation. The trained model parameters are stored in the intelligent control unit 4 and directly accessed during online detection to output the liquid supply and recovery control parameters in real time.

[0193] The random forest model is trained by minimizing the error between the predicted control parameters and the target control parameters, with the objective function being:

[0194]

[0195] In the formula, This is the normalized coupling quality feature vector; These are the corresponding liquid supply and recovery control parameters; The parameters are obtained through offline training and are determined by minimizing the coupling quality error cost function.

[0196] The training process of the random forest model is as follows

[0197]

[0198]

[0199]

[0200] in, For random forest regressors, This refers to the speed of the liquid supply pump; Valve opening degree; To recover negative pressure.

[0201] The above methods enable data-driven decision-making based on ultrasonic signal quality and operating condition information, allowing the liquid supply and recovery control parameters to be adaptively adjusted according to changes in operating conditions.

[0202] S5 Control Parameter Limiting and Execution Control

[0203] The intelligent control unit 4 performs threshold determination, amplitude limiting and execution control on the liquid supply and recovery control parameters output in step S4.

[0204] Intelligent control unit 4 according to continuous control cycle Internal Coupling Quality Evaluation Indicators Make a judgment: when Continuous preset number of cycles Below the coupling quality threshold When this occurs, it is determined that there is an insufficient coupling state, and liquid supply enhancement or recovery adjustment control is triggered; when If the above continuous determination conditions are not met, the current control state shall be maintained.

[0205] To avoid device instability caused by sudden changes in control parameters, the liquid supply control parameters and recovery control parameters need to be limited and smoothed before execution to keep their change range or rate within a preset safe range.

[0206] Control constraint and smoothing unit 404 pairs The amplitude limiting and smoothing processing, control constraints and smoothing parameters are shown in Table 4.

[0207]

[0208]

[0209] The processed liquid supply control parameters and recovery control parameters are sent by the execution command generation unit 405 to the actuators such as the liquid supply pump, valves, and vacuum pump, completing the closed-loop regulation of the coupling fluid supply and recovery process. The execution command generation unit 405 will... The commands are converted into commands for the liquid supply pump speed, solenoid valve opening degree, and recovery negative pressure, and then sent to the coupling fluid supply and recovery unit 3 for execution.

[0210] By using the threshold determination and execution control steps described above, it is possible to suppress momentary interference misjudgments while maintaining the stable coupling state and intelligently optimizing the amount of coupling fluid.

[0211] Table 4:

[0212]

[0213] S6 Detection End, Fault Diagnosis, and Loop

[0214] Based on the detection process status and coupling fluid supply and recovery conditions, the system determines the detection termination conditions and abnormal states, and executes corresponding termination control or abnormal handling operations. The status detection and fault diagnosis unit 406 performs abnormal identification based on the operating data. When the detection is not completed and no fault is triggered, it returns to S2 to enter the next control cycle; when the detection is completed, it enters the shutdown state; when coupling is insufficient, it enters the protection state and records the abnormal information. Insufficient coupling can be determined by the following formula:

[0215]

[0216] This is a preset coupling quality threshold; The number of consecutive determination periods. The control cycle can be set to 3 to 10 cycles, depending on the control cycle and the dynamic characteristics of the device.

[0217] After the insufficient coupling anomaly is eliminated or the detection process is completed, the device performs one of the following operations based on the operating status: if the anomaly is eliminated but the detection task is not completed, it returns to step S2 to continue the detection process; if the detection task is completed or the anomaly cannot be recovered, it exits the detection process and enters standby or shutdown state.

[0218] like Figure 2 As shown, this flowchart is a schematic diagram of the offline training phase of the random forest structure used to output the control parameters for coupling fluid supply and recovery in this invention.

[0219] T1 Training data acquisition and sample construction.

[0220] During the equipment commissioning phase, historical test records, or manual experience control phase, collect ultrasonic test data of the wheels and corresponding coupling fluid supply and recovery condition data.

[0221] Training samples are constructed based on the data. Each sample includes input features and target output. The training sample format is as follows:

[0222]

[0223] in, For coupling quality feature vectors, These are the corresponding liquid supply and recovery control parameters.

[0224] T2 Training Data Preprocessing. Ultrasonic signals and operational data in the training samples are preprocessed, including windowing, time alignment of operational data, and feature scale normalization.

[0225] T3 Feature Extraction and Training Feature Vector Construction. Based on the preprocessed ultrasonic signal, signal quality features such as initial wave amplitude, echo energy, and signal-to-noise ratio are extracted. Operating parameters such as fluid supply flow rate, recovery negative pressure, and water tank level are used as auxiliary features to construct the training feature vector. .

[0226] T4 randomly divides the preprocessed dataset into training, testing, and validation sets at 70%, 20%, and 10% respectively.

[0227] T5 Random Forest Model Training and Parameter Optimization. A random forest regression model is constructed using the training set samples from step T4, and the model parameters are adjusted and optimized using the validation set.

[0228] Random forests employ a single-model, multiple-output approach to simultaneously output control parameter vectors for liquid supply and recovery.

[0229]

[0230] In the formula, For the speed of the liquid supply pump, For valve opening, To recover negative pressure.

[0231] For a given sample set S of a node in a decision tree, the node output is taken as the mean of the target output vector within that node.

[0232]

[0233] The node splitting of the regression decision tree uses a weighted squared loss function of the multi-output prediction error.

[0234]

[0235] in, Let be the weight matrix. For candidate splits... The generated left and right child nodes and , choose to The features and thresholds are used to complete node splitting. The outputs of multiple decision trees are ensembled to obtain the final output of the random forest model.

[0236] T6 model evaluation, solidification, and deployment.

[0237] The final random forest model obtained from T5 is used to make predictions on the test set, yielding the predicted values ​​of the control parameters for the test set samples. and compared with the actual control parameters of the test set. A comparative evaluation was conducted. Evaluation metrics for the test set included mean squared error, mean absolute error, or root mean square error. The parameters for this random forest model were set as follows: number of decision trees: 200; maximum depth: 12; minimum number of samples per leaf node: 5; minimum number of split samples: 15; and feature subsampling ratio: 0.6.

[0238] Once the test set evaluation is passed, the structure and parameters of the random forest model are fixed and saved, and the fixed model is stored in the parameter management unit 407.

[0239] The solidified random forest model is deployed to the intelligent control unit 4. During the deployment process, the model inference interface is encapsulated so that its input is the real-time feature vector constructed in the online operation process. The output is a vector of liquid supply and recovery control parameters. .

[0240] A functional unit structure diagram of intelligent supply and recovery of coupling fluid for ultrasonic testing of railway wheels, as shown below. Figure 3 As shown. The electromechanical motion mechanism 1 drives the probe assembly 2 to scan the wheel tread along a preset trajectory; the probe assembly 2 scans the wheel tread. The ultrasonic testing and data acquisition unit 5 is connected to the probe assembly 2 and is used to acquire echo signals. The signal processing and defect assessment unit 6 is connected to the data acquisition unit 5 and is used to process the echo signals and output defect assessment results to the detection result and repair decision output unit 7. The coupling fluid supply and recovery unit 3 supplies coupling fluid to the probe-wheel coupling area and recovers the returned coupling fluid. The intelligent control unit 4 is connected to the coupling fluid supply and recovery unit 3, the data acquisition unit 5, and the defect assessment unit 6 respectively, and is used to fuse signal quality and operating condition information and output fluid supply and recovery control commands to form a closed-loop regulation.

[0241] The intelligent control unit for intelligent supply and recovery of coupling fluid in ultrasonic testing of railway wheels has the following structure: Figure 4As shown, the intelligent control unit 4 includes: a data interface and acquisition unit 401, a feature construction and preprocessing unit 402, a random forest intelligent decision-making unit 403, a control constraint and smoothing unit 404, an execution instruction generation unit 405, a state detection and fault diagnosis unit 406, and a parameter management unit 407. The parameter management unit 407 stores thresholds, upper and lower limits, control cycles, and model parameters; the data interface and acquisition unit 401 receives signal quality information and operating condition data; the feature construction and preprocessing unit 402 constructs coupling quality features; the random forest intelligent decision-making unit 403 outputs fluid supply and recovery control parameters; the control constraint and smoothing unit 404 limits amplitude, smooths, and restricts rate; the execution instruction generation unit 405 generates executable control instructions for the coupling fluid supply and recovery unit 3; and the state detection and fault diagnosis unit 406 identifies anomalies and triggers protection strategies.

[0242] The detailed explanations of the above embodiments are intended only to explain the present invention so as to facilitate a better understanding of the present invention. However, these descriptions should not be construed as limiting the present invention for any reason. In particular, the various features described in different embodiments can be arbitrarily combined with each other to form other embodiments. Unless there is an explicit description to the contrary, these features should be understood to be applicable to any embodiment, and not limited to the described embodiments.

Claims

1. A method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels, characterized in that, The supply recovery method includes: Acquire echo signals from ultrasonic testing and operating data of the coupling fluid supply and recovery device; Based on the echo signal, signal features reflecting the coupling state are extracted, and coupling quality evaluation indicators are constructed in combination with the operating condition data. The coupling quality evaluation index and the operating condition data are input into the trained prediction model, and the prediction model outputs liquid supply control parameters and recovery control parameters adapted to the current operating condition. The coupling fluid supply to the recovery device is adjusted according to the coupling quality evaluation index, the fluid supply control parameters, and the recovery control parameters.

2. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, The echo signal is the wheel ultrasonic echo signal, and the operating data includes at least the water tank level, liquid supply flow rate, liquid supply pressure, recovery negative pressure and / or wheel speed.

3. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 2, characterized in that, The supply recovery method further includes: The echo signal and the operating condition data are preprocessed by dividing the echo signal into an effective echo window and a noise window, and performing time synchronization on the operating condition data and removing outliers.

4. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, The signal characteristics include at least one of the following: the amplitude of the first echo, the mean square value within the noise window, and the signal-to-noise ratio.

5. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 4, characterized in that, The formula for extracting the signal features includes: （1） （2） （3） In the formula, It is a constant, with a value ranging from 10⁻¹² to 10⁻⁶; The amplitude characteristics of the first echo, The mean square value within the noise window, For signal-to-noise ratio, This is the ultrasonic echo signal from the wheel.

6. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 5, characterized in that, The coupling quality evaluation index is constructed based on the signal characteristics: （4） In the formula, Amin, Amax, SNRmin, and SNRmax are parameters set empirically. The amplitude characteristics of the first echo, The mean square value within the noise window, This refers to the signal-to-noise ratio.

7. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, The prediction model used is a random forest model.

8. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 7, characterized in that, The random forest model is obtained through offline training, which includes: Collect historical test data and construct training samples containing coupled quality feature vectors and corresponding liquid supply and recovery control parameter vectors; The data in the training samples are preprocessed; Extract signal quality features and combine them with operating parameters to construct a feature vector for training; The dataset is divided into a training set, a validation set, and a test set; The random forest regression model is trained using the training set, and the model parameters are optimized using the validation set. The model performance is evaluated using the test set, and once the evaluation is passed, the model parameters are solidified and deployed.

9. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, The prediction model employs linear regression, nonlinear regression, support vector regression, neural network, gradient boosting regression, or k-nearest neighbor regression.

10. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, Adjusting the coupling fluid supply to the recovery device according to the coupling quality evaluation index, the fluid supply control parameters, and the recovery control parameters includes: When the coupling quality evaluation index is lower than the coupling quality threshold within a consecutive preset period, it is determined that there is an insufficient coupling state, and liquid supply enhancement or recovery adjustment control is triggered; when the coupling quality evaluation index does not meet the above continuous determination conditions, the current control state is maintained.

11. The method for supplying and recovering coupling fluid for ultrasonic testing of railway wheels as described in claim 1, characterized in that, Adjusting the coupling fluid supply to the recovery device according to the coupling quality evaluation index, the fluid supply control parameters, and the recovery control parameters includes: The liquid supply control parameters and the recovery control parameters are subjected to threshold determination, amplitude limiting and smoothing processing, and execution instructions are generated and sent to the actuators of the coupling liquid supply and recovery device to adjust the liquid supply pump speed, valve opening and recovery negative pressure.

12. A supply and recovery device for coupling fluid in ultrasonic testing of railway wheels, used to implement the method according to any one of claims 1-10, characterized in that, The supply recovery device includes: A wheel detection electromechanical probe assembly includes an electromechanical motion mechanism and a probe assembly. The electromechanical motion mechanism drives the probe assembly to scan the wheel tread along a preset trajectory and perform ultrasonic testing on the wheel. An ultrasonic testing and data acquisition unit, connected to the probe assembly, is used to acquire the ultrasonic echo signal obtained by the probe assembly; The coupling fluid supply and recovery unit is used to supply coupling fluid to the coupling area between the probe assembly and the wheel and to recover the reflux fluid. A signal processing and defect assessment unit, connected to the ultrasonic detection and data acquisition unit, is used to process the ultrasonic echo signal and output signal characteristics; The intelligent control unit is communicatively connected to the coupling fluid supply and recovery unit, the ultrasonic detection and data acquisition unit, and the signal processing and defect assessment unit, respectively. The intelligent control unit constructs coupling quality evaluation indicators based on signal characteristics and operating condition data. The intelligent control unit also has a pre-trained prediction model built in, which generates fluid supply and recovery control parameters based on the coupling quality evaluation indicators and operating condition data.

Images